Shaduf.Research preview
Jev: Use Cases, Alternatives & Products/Research reports and method
EvidenceUpdated

Jev (TypeSafe AI) catalogue research history and evidence method

Every claim on this site comes from a dated research run. This page lists all reports, newest first, explains the evidence labels, and records what changed in each release, including where every earlier URL now points.

Short answer

42 dated reports cover 20 September to 9 November 2026, the newest being the research of 9 November (three benchmark studies graded from their files, so the ledger has 48 graded rows; the Jev Decision Index re-read with unchanged scores and Artificial Analysis with a changed index; auto-mode 4.0.0 read in source, failure path unchanged and still unsettled; Nace.AI's legal pages now agree, OpenCode's texts still differ; TypeSafe unchanged, with no check between 8 Nov, 05:18 UTC and 9 Nov, 11:07 UTC). Before that, the research of 8 November (ohmyjev 0.14.0 read in source: a new opt-in shadow setting logs what lockdown would block without blocking it, in-repo file moves skip Jev, and its other failure cells and the failure map's 80 are unchanged; pi-jev-router for pi read and counted on the coding-agents page; SubQ's docs and Terms now agree on 60-day retention; TypeSafe unchanged, with no check between 7 Nov, 15:29 UTC and 8 Nov, 05:17 UTC). Before that, the research of 7 November (TypeSafe's models page now lists images as an input for both Jev versions and a Box question type on Jev 1.14 only, while aliases, price and access are unchanged, with no check between 2 Nov and 7 Nov; Base Labs' three decision models added to the hosted alternatives; a dated list of projects that stayed on Jev 1.13). Before that, the research of 2 November (ohmyjev's row on the failure map read again at 0.13.2: its Jev gates now run only with its lockdown setting on, which is off by default; in that mode its failure terms are unchanged, and so are the map's counts; OpenCode lists a free Jev ID, jev-1.13-free, on its own third-party route; TypeSafe unchanged, with no check between 31 Oct and 2 Nov). Before that, the research of 31 October (the 40 hosted decision API comparison cells dated before 17 October read again: 6 changed, among them Microsoft-Decision-1's rate limits, Cloudflare's zero-data-retention wording for Workers AI and Clef-flash's hosts on OpenRouter; Eevo's prices changed; TanStack AI added to the framework packages read in source; access unchanged). No update was published on 9 October; separate evidence notes from 9 October on Jev rerankers and SQL and document projects are not counted among the reports. Each is kept unchanged at its original address. Facts on the question pages carry the date of the report that checked them; a later page edit does not re-date an old fact.

Reports by date and topic

42 reports, 20 September–9 November 2026, by topic; a day without a dot has no report

20 Sep21 Sep22 Sep23 Sep24 Sep25 Sep26 Sep27 Sep28 Sep29 Sep30 Sep1 Oct2 Oct3 Oct4 Oct5 Oct6 Oct7 Oct8 Oct9 Oct10 Oct11 Oct12 Oct13 Oct14 Oct15 Oct16 Oct17 Oct18 Oct19 Oct20 Oct21 Oct22 Oct23 Oct24 Oct25 Oct26 Oct27 Oct28 Oct29 Oct30 Oct31 Oct1 Nov2 Nov3 Nov4 Nov5 Nov6 Nov7 Nov8 Nov9 NovAccess and priceAlternativesProjects and use casesMethodFull scopeBuild and integrationsAccuracy and evidence
Each dot is one report; select it to open the report. The list below has the same reports with summaries.
  1. What changed for Jev builders, 8 to 9 November 2026Accuracy and evidence

    Benchmarks: B49 (S1CogBench, grade B), B50 (FrontierLLMvsJEV, grade A, small sample) and B51 (graph-reflex, grade B, the author's own harness) graded from their files; ledger 48 graded rows (A 15, B 16, C 10, D 7); two leads not graded. Jev Decision Index: scores unchanged (60.24 and 59.22), 275 entries; Artificial Analysis: Jev 1.14 66.96, Jev 1.13 no longer listed, index changed. "Is Jev deterministic?" answered from one option-order test (B41). auto-mode 4.0.0: failure path unchanged, still not settled (counts and the failure map unchanged). Nace.AI's legal pages agree; OpenCode's Terms re-dated 8 Nov. TypeSafe unchanged.

  2. What changed for Jev builders, 7 to 8 November 2026Build and integrations

    ohmyjev re-read at 0.14.0 (fbc4ea4): its four other failure cells are unchanged and lockdown is still off by default; a new opt-in shadow setting, also off by default, logs what lockdown would block without blocking it, and in-repo mkdir, touch, cp and mv now skip Jev; the failure map still counts 80. Coding agents: @caph42/pi-jev-router 2.0.0 (pi) read in source and counted on the page (29 read, 15 counted), off the failure map; 19 newer projects listed as leads. SubQ's docs and Terms agree on 60-day retention; Cloudflare's data-usage page lists a third exception; OpenAI unchanged. langchain-typesafe 0.0.2: failure path unchanged. TypeSafe unchanged; its docs index adds one cookbook.

  3. What changed for Jev builders, 2 to 7 November 2026Build and integrations

    TypeSafe's models page now lists "Text, images" as the input of both jev-1.14.0 and jev-1.13.0, a Box question type on Jev 1.14 only, and no "Language support" section; a new docs changelog dates the Jev 1.14 entry 14 Oct; the Images page still calls images not ready for production (first observed 7 Nov; no image sent). Aliases, price, signup, credit and service unchanged, with no check between 2 Nov and 7 Nov. Alternatives: one Base Labs row (mammoth-1, elephant-1, joey-1 on Baseten; prices, data terms and MIT weights as published; rank claims Reported, not matched). Versions and aliases: retitled "Jev 1.14 vs 1.13" with eight GitHub items, read on 7 Nov, on staying on 1.13 (Reported). OpenCode's free ID unchanged.

  4. What changed for Jev builders, 31 October to 2 November 2026Build and integrations

    ohmyjev re-read at 0.13.2 (commit d307c39): its Jev gates now run only with lockdown on, off by default; in lockdown mode its five failure terms are unchanged (fail-open on a Jev error, a malformed answer and a missing key; acts-anyway below its cut-offs), the bypass cell changed in content, and the failure map still counts 80 implementations with the same totals. OpenCode lists jev-1.13-free ("Free", "for a limited time") on its own third-party route and does not say which Jev version it serves; its docs and Terms are quoted side by side. OpenRouter's ZDR list now names both TypeSafe endpoints. Changed: status-page uptime figures, Laya v0.4.21 and two package versions (not read). TypeSafe unchanged; two notices.

  5. Hosted decision APIs compared: what changed by 31 October 2026Alternatives

    The 40 comparison cells dated before 17 Oct read again: 29 unchanged, 6 changed, 2 re-read in part, 2 not re-read, 1 first read. Changed, each with its date range: Microsoft-Decision-1 rate limits (2,160–15,000 requests and 2.16M–15M tokens per minute by tier), Cloudflare's Workers AI zero-data-retention and no-training wording (page updated 16 Oct), Clef-flash also served by Makora and DeepInfra on OpenRouter, OpenAI's inputs (text, images and JSON) and regions wording. Eevo: $0.10 per 1M input tokens and $0.15 per decision. Kev and Laya: new releases not yet read. Framework packages: TanStack AI read at commit 5651e07 (raises to your code); 7 read in source. Access unchanged; two notices.

  6. What changed for Jev builders, 24–30 October 2026Access and price

    The first check after 24 Oct (no check in between). Unchanged and re-dated 30 Oct: signups open, no new-user credit, API operational (no incident in the status page's history for 15–30 Oct), jev-latest and jev-preview on jev-1.14.0, jev-1.13.0 listed with no deprecation date, $0.042 per million input tokens, six route listings, Opper's alias still on 1.13. Changed, each with its date range: OpenAI's Decisions API retention eligibility, Microsoft-Decision-1 regions, SubQ Snap 1.1 at $0.04 per million input tokens (docs; the launch post still says $0.05), TypeSafe's higher-limit request form; Base Labs' free-week wording not found on two Baseten pages. Catalogs: Bedrock, Microsoft Learn's Foundry pages and Workers AI do not list Jev. Coding agents: acmelabs-15/jev-gate read in source (not settled; off the failure map; 28 read in source), ten newer names classed. Integration releases noted. Three notices.

  7. jev-latest now gives Jev 1.14; decision leaderboards and four new graded studiesAccess and price

    jev-latest on TypeSafe's API now gives jev-1.14.0 (first seen 24 Oct, 05:12 UTC; still jev-1.13.0 at 23 Oct, 05:45 UTC; no dated announcement); jev-1.13.0 still listed by its ID with no deprecation date; Opper's alias still on 1.13. Pages that name a default model now say what an unpinned request gets. Benchmarks: four decision leaderboards that list Jev, with vendor claims checked against their data (Base Labs' "#1 of 208" matches the 20 Oct data, not the 22 Oct data); B45–B48 graded (ledger 45: A 14, B 14, C 10, D 7). OpenAI's Decisions API prices cached input since general availability. Access unchanged; three notices.

  8. Jev with images: what 8 SDKs and integrations do with an image, and the route counts for Jev 1.14Build and integrations

    Limitations: new section "Jev with images": TypeSafe's image docs quoted (PNG or JPEG, opt-in header, 400 and 413, the whole request fails) and 8 integrations read in package files (5 pass images, 1 refuses them, 2 have no image type; of the 5, 2 add the header for you). TypeSafe's n8n node 0.10.0: failure behaviour unchanged, Images field added. pydantic-ai-slim 2.60.0: image check unchanged. jev-latest still on jev-1.13.0; 6 of 6 routes list a 1.14 ID. OpenAI's Decisions API generally available (22 Oct). Access unchanged; two notices. No image was sent.

  9. Coding agents on pi and omp, auto-mode and jcode, and Nace.AI's two sets of termsBuild and integrations

    Coding agents: ohmyjev's pi and omp adapters counted (fail-open on a Jev error); 14 of 27 integrations read in source act on Jev's answer (3 fail-closed, 9 fallback, 2 fail-open); auto-mode 3.0.0 and jcode read, not settled; 13 new leads classed. When Jev fails: ohmyjev's row names pi and omp (still 80). Alternatives: Nace.AI's website Terms and Privacy Policy (9 Oct) and Console terms (19 Oct) quoted side by side; drex-v1.5 price units side by side. Cloudflare's catalog lists typesafe/jev-1.14.0. Access unchanged; the two Jev 1.14 home notices came off after seven days. Nothing was run.

  10. Failure self-check, checklist question 9, and Drex on alternativesBuild and integrations

    When Jev fails: a self-check compares what your code does in five failure conditions with the 80 implementations on the map (map recomputed 19 Oct; same counts as the map; it does not read or test code). Checklist question 9 asks whether your code logs or shows a fallback or fail-open; the map does not record this, so it has no count. Sentences on 14 pages that still said 79 now say 80. Alternatives: Drex (Nace.AI) added as a hosted API from another provider with open weights (38 substitutes; 8 hosted APIs from other providers); its claims against Jev are the vendor's own. LiteLLM 1.106.0: Jev failure path unchanged. n8n 2.44.3: activation refused without a required credential. Access unchanged. Nothing was run.

  11. Failure map completed with ohmyjev, jev-mcp commits, and three hosted decision modelsBuild and integrations

    When Jev fails: ohmyjev joins the map after its pi and omp adapters were read in full, so it counts 80 implementations (was 79), every failure cell recorded for all 80. On a Jev error 40 stop or hold the action, 22 fall back, 10 let it through, 5 return no decision and 3 are advisory; the action goes ahead in 19 of 80. jkudish/jev-mcp: every published cell was read at 0.11.0 (53fe575); 0.13.0 and 0.14.1 are later checks with the failure path unchanged. Content moderation (11 cells) and support-ticket triage (9 cells) now lead with the map terms; no outcome changed. Alternatives: Liquid AI d1-ultra, Mouse-1 and Eevo added with published prices and data terms (37 substitutes; 7 hosted APIs from other providers); no graded comparison. Access unchanged. Nothing was run.

  12. Failure terms aligned, three Claude Code guardrails read, and the products ledger recheckedBuild and integrations

    When Jev fails: claude-model-guard and claude-jev-advisor join the map, so it counts 79 implementations (was 77), every failure cell recorded for all 79. On a Jev error 40 stop or hold the action, 22 fall back, 9 let it through, 5 return no decision and 3 are advisory. Failure terms on 15 pages compared cell by cell with the map: 0 outcome differences; coding agents corrected to 3 fail-closed, 9 fallback and 1 fail-open on a Jev error (was "5 let the call run, 3 hold, 5 fall back"); trading and browser cells use the map terms. Claude Code and MCP: ten guardrails read; two hold or stop, one falls back, two shadow or log-only, one advisory, three fail open, one verdict only; ohmyjev shown, not yet in the map count. Products: 113 rows (99 on 11 Oct), A 100, B 12, C 1; verified in production 0 of 113. Access unchanged; Opper lists Jev 1.14. Nothing was run.

  13. Bypass cells complete, Jev 1.14 preview and Claude Code guardrails recheckedBuild and integrations

    When Jev fails: every failure cell (no key, error, malformed answer, below threshold and bypass) is now recorded for all 77 counted implementations, read at pinned commits. Bypass recorded for 77 of 77 (was 66 of 77; 13 cells are "none found" in the files read); malformed answer: 35 fail closed, 24 fall back, 11 fail open, 4 no decision, 3 advisory, 0 not recorded (competitor-hunter is now no-decision). Jev 1.14 (jev-1.14.0) is a preview under jev-preview; jev-latest still points to jev-1.13.0, and TypeSafe says it moves "in a week". Image input, a Box question type and 262k long context are opt-in experiments, not production-ready per TypeSafe. The models page no longer publishes rate-limit numbers. Claude Code and MCP: short answer unchanged (1 of 7 stops the agent when Jev fails); 32 leads classified; new title. Benchmarks: the Artificial Analysis Decision Index added as a Reported line outside the ledger (41 graded rows, unchanged). Nothing was run.

  14. No key, code paths that skip Jev, and the Python and TypeScript SDKsBuild and integrations

    When Jev fails, no key: of 77 implementations read at pinned commits, 50 fail closed, 17 fall back, 7 fail open, 3 are advisory, 0 not recorded. Code paths that never ask Jev are recorded for 66 of 77; 11 are not recorded (judge tools 4, frameworks 3, browser agents 3, email 1); a cell counts as recorded only when every entry point was read. Below threshold: 0 not recorded. Python: typesafe-sdk 0.7.4 keeps its defaults; TypeScript: ai 7.0.140 renamed experimental_evaluate to experimental_decide, and the AI SDK provider 3.0.18 throws LoadAPIKeyError at the first call without a key. Benchmarks: B43 graded B (SubQ Snap vs Jev; the operator takes part in a Subquadratic prize challenge) and B44 graded A (a Tetris round robin); 41 graded rows (A 12, B 14, C 8, D 7). SubQ Snap added on alternatives (34 substitutes) with two retention texts that disagree. Access unchanged; nothing was run.

  15. Decisions API vs Jev vs Microsoft Decision 1, and what Jev integrations do with no keyAlternatives

    Decision APIs side by side: graded results depend on the task (OpenAI's Decisions API ahead on RPG move accuracy, B38; Jev on Portuguese exact judge score, B39; Jev and Microsoft-Decision-1 not separated on routing macro-F1, both ahead of OpenAI, B41), with a new comparison page. B42 graded B: the ledger has 39 graded rows (A 11, B 13, C 8, D 7). When Jev fails, no key: 43 fail closed, 16 fall back, 7 fail open, 3 advisory, 8 not recorded (of 77); bypass recorded for 52 of 77. Microsoft-Decision-1 is listed as Preview in Foundry at 13 Oct 2026, 05:20 UTC (GA on 12 Oct; cause unknown).

  16. Real-time control demos, Microsoft-Decision-1 and the Cloudflare Clef familyProjects and use cases

    Real-time control: of 3 published game and robot-simulation demos that call Jev, read at pinned commits, 2 keep their last action when a call fails (doom-jev, JevPlaysSnakeGame) and 1 stops (jev-robotics-demo); none is a product. A new page shows each loop's tick against its request timeout. When Jev fails now counts 77 implementations in 12 domain tables: on a Jev error 39 stop or hold the action, 22 fall back, 8 let it through, 5 return no decision and 3 are advisory; the action goes ahead in 17 of 77. Benchmarks: B41 graded A (synthetic bank-withdrawal routing; Jev and Microsoft-Decision-1 not separated, both ahead of OpenAI's Decisions API); the ledger has 38 graded rows (A 11, B 12, C 8, D 7). Alternatives: the Cloudflare Clef family added as two rows (33 listed substitutes); Microsoft-Decision-1's data cell partly read. Jev not listed in Microsoft Foundry's model catalog (12 Oct 2026, 05:15 UTC). Access unchanged; no notice retired. Nothing was run; no Jev, Microsoft or OpenAI call.

  17. Jev benchmark grades and coding-agent settlementAccuracy and evidence

    Benchmarks: three comparisons first listed as leads on 10 Oct are graded A from metric and results files opened at pinned commits (B38, B39, B40); the ledger has 37 graded rows (A 10, B 12, C 8, D 7). Two compare Jev with OpenAI's Decisions API and differ by task: the Decisions API ahead on RPG move choices (B38); Jev ahead on exact judge score in Portuguese, level on routing and injection (B39). B40 is a reranking-pipeline study in which Jev and Cohere Rerank 4 Pro are within noise. Alternatives: Microsoft-Decision-1 added as a second hosted API from another provider (31 listed substitutes). Coding agents: of 24 integrations read in source, 13 act on Jev's answer, 10 do not, 1 is unsettled. When Jev fails now counts 74 implementations: on a Jev error 38 stop or hold the action, 20 fall back, 8 let it through, 5 return no decision and 3 are advisory. Real-time control: the condition for the run-14 page is met (research only). Access unchanged; no notice retired. Nothing was run; no Jev or OpenAI call.

  18. Jev alternatives by layer, and coding-agent integrationsAlternatives

    Alternatives: Jev weights or a local Jev runtime published by TypeSafe were not found on TypeSafe's site, docs, GitHub organisation or a Hugging Face author search (10 Oct 2026, 05:11–05:20 UTC); 30 listed substitutes sorted by layer (12 model weights, 11 local runtimes, 4 request adapters, 2 classifier or LLM approaches, 1 hosted API from another provider: OpenAI's Decisions API, public beta); four Jev-named Hugging Face model families are not published by TypeSafe. Coding agents: of 22 integrations read in source, 10 act on Jev's answer, 10 do not and 2 are unsettled; 35 leads triaged. When Jev fails now counts 71 implementations: on a Jev error 38 stop or hold the action, 17 fall back, 8 let it through, 5 return no decision and 3 are advisory. Benchmarks: jev-pulse graded A; three Jev vs Decisions API or reranking comparisons listed as leads, not graded. Access unchanged; two notices retired. No run took place on 9 Oct. Nothing was run; no Jev call.

  19. Jev reranking and document pipelines: evidence notes, 9 October 2026Evidence notes, 9 Oct 2026

    What five Jev rerankers and context filters and five SQL and document projects did when the Jev call failed, as read in source on 9 October; that day's access and price snapshot, version checks and open questions. Dated observations, not rechecked since; some statements have no surviving source link. Not counted among the 29 dated reports.

  20. Jev reranker and RAG, and Jev inside SQL and document pipelinesProjects and use cases

    Reranking and RAG: the grade-A study measures relevance judgments, not reranking; both rerunnable graded pipeline studies come from parties with a stake; of 5 reranking and passage-filter implementations, on a Jev error 2 stop, 2 fall back and 1 passes unscreened passages. SQL and document pipelines: on a Jev error all 5 projects stop the query or request; pg-redact stores text unredacted when a span's answer is missing or below 0.5. When Jev fails now counts 69 implementations: on a Jev error 38 stop or hold the action, 15 fall back, 8 let it through, 5 return no decision and 3 are advisory. Correction: Hermes jev-approvals falls back to the main agent model on a timeout, connection error, 429 or 402. Access unchanged (checked 8 Oct 2026, 05:14 UTC). Nothing was run; no Jev call.

  21. Judge tools counted when Jev fails, coding-agent cells settled, and new Jev benchmarks gradedBuild and integrations

    When Jev fails now counts 64 implementations in 11 use cases, including 4 eval judges and 2 Hermes Agent plugins; on a Jev error 35 stop or hold the action and 8 let it through. Coding agents: all 5 open cells settled; 8 of 18 act on Jev's answer. Benchmarks: 9 new studies graded (33 graded rows); two short-answer clauses changed. Access unchanged (checked 7 Oct 2026, 05:19 UTC). Nothing was run; no Jev call.

  22. Coding agents beyond Claude Code, and Jev model IDs and aliases by routeBuild and integrations

    Read 18 Jev integrations for coding agents (Codex, OpenCode, Cursor, OpenClaw, Hermes Agent) at pinned commits: 1 is vendor-published (OpenClaw) and none comes from TypeSafe; 6 act on Jev's answer, and 4 of those let the tool call run when Jev errors. Read the model IDs and aliases on 6 routes: 4 accept a versioned ID, Vercel and Cloudflare only an unversioned one; jev-latest points to jev-1.13.0 on TypeSafe's models page. The failure checklist grows to 58 implementations in 10 use cases: 32 stop or hold the action on a Jev error and 7 let it through unchecked. Products: 92 rows (A 75), none independently verified. LiteLLM 1.104.0 failure path unchanged. Vercel AI Gateway now lists TypeSafe as a second provider for Jev. Access unchanged (checked 6 Oct 2026, 05:15 UTC). No agent, plugin or hook installed or run; no Jev call.

  23. Browser and computer use, framework integrations, and Jev on Bedrock or AzureProjects and use cases

    Read 7 Jev browser and computer-use agents at pinned commits: none clicks or types when Jev errors, but 5 act on low-confidence answers and only 1 asks a person before send, buy or delete; 5 of 7 run in the user's own session by default. Read 6 framework packages: 4 are pre-1.0 and none comes from TypeSafe; 4 raise Jev errors to the caller, while the community Spring AI filter and the LlamaIndex reranker fail open. Jev is not listed in the Amazon Bedrock model cards (5 Oct 2026, 05:18 UTC) or the Azure AI Foundry catalog (05:19 UTC). The failure checklist grows to 51 implementations in 9 use cases: 30 stop or hold the action on a Jev error and 6 let it through unchecked. Products: 72 rows (A 55), none independently verified. Access unchanged (checked 5 Oct 2026, 05:13 UTC). No pool Jev call, agent run or install.

  24. Trading and fraud gates, the failure checklist across use cases, and Jev as a judgeProjects and use cases

    Read four Jev trading bots and three fraud or payment gates at pinned commits: 2 of the 4 bots can still place a new order when Jev errors, all 4 can trade without Jev, and none of the 3 gates has a payment call site. Recomputed the failure checklist across 39 implementations in 7 use cases: on a Jev error, 21 stop or hold the action and 5 let it through unchecked, with "not recorded" cells counted. Read four Jev judge tools: DeepEval stops the run on a Jev error; the other 3 drop the item from the score, and openlayer jevals still reports it as passed. Products: 67 rows (A 47), none independently verified. TypeSafe's trust center lists 6 subprocessors (Nebius and CoreWeave added 2 Oct). Access unchanged (checked 4 Oct 2026, 05:15 UTC). No pool Jev call, trade, payment or eval.

  25. Email and lead sorting, the oh-my-claudecode v5.6.0 check, the Products refresh and n8n nodesProjects and use cases

    Read five email and lead sorters at pinned commits: none changes the mailbox when Jev errors; of the 3 that can write to the mailbox, only 1 sends unsure mail to a Review label. oh-my-claudecode v5.6.0 keeps its failure path; its default timeout is now 2,000 ms. Recounted the Products ledger: 61 rows, 39 grade A, none independently verified in production. Identified TypeSafe's own n8n node (the scoped @typesafe-ai package); all 5 nodes read stop the workflow on a Jev error by default. Listed request limit now 80 per second; access unchanged. No pool Jev call.

  26. Model routing, the jev-mcp version check, and content moderationProjects and use cases

    Read OpenRouter's Jev Router docs (failure, long-context and routed-model billing not documented) and six self-hosted Jev routers at pinned commits: on a Jev failure one sends the request to its frontier model, five keep the current model, pick locally or choose no model. Checked @jkudish/jev-mcp 0.12.0 and 0.13.0: failure path unchanged. Read six moderation projects at pinned commits: four leave the post up when Jev fails, one holds it, one returns no decision; no published moderation accuracy found. Access unchanged. No pool Jev call.

  27. Laya vs Kev, data handling by route, and support-ticket triageAlternatives, data handling and use cases

    Compared hosted Jev, Laya and Kev on the same rows: only Opper's test (B11) finds an open model, Kev-4B, within noise of Jev. Quoted training, retention and ZDR terms for TypeSafe direct and four gateways with both layers apart: no text allows training without consent; Vercel also routes Jev through DigitalOcean, marked not ZDR. Read five ticket-triage projects in source: four send unsure tickets to a person, one archives on errors. Access unchanged; rate-limit change narrowed to 29 Sep 19:24 – 30 Sep 20:47 UTC. No pool Jev call.

  28. Evaluation evidence and home status: graded benchmarks, terms and accessAccuracy and evidence

    Graded 14 published Jev studies at their primary source and 10 shorter rows (A–D with flags; all results Reported); found that TypeSafe's MCA barred publishing benchmarks until its 19 Sep version; set out 10 criteria for Jev vs an LLM vs a classifier; built the home status dataset (signups open, no new-user credit, API operational; console HTTP 403). Rate limits changed to 100K tokens/s and 40 requests/s. No pool Jev call.

  29. Integration trust: Claude Code and MCP guardrails, TypeScript paths and testing without a keyBuild and integrations

    Read the failure paths of five Claude Code, MCP and agent-guard integrations at pinned commits, and the official plugin in full: only LangChain AutoModeMiddleware fails closed. Compared four TypeScript paths, twelve no-key test facilities and ten failure branches; ran one hook and one retry client offline. Corrected the oh-my-claudecode timeout note and the jev-mcp product row. Access and price fields unchanged. No pool Jev call.

  30. Release 1 question pages: access, cost, trust, build, limits and videosFull scope

    Checked signup and credit status (reopened 27 Sep, no new-user credit), route prices and limits, lookalike resellers (3–16× list price), billing basis and third-party token counts, production claims, the Python SDK, error codes, confidence semantics, 17 documented limitations, 46 demo videos and new alternatives. Corrected two wrongly 'official' project labels. No pool Jev call.

  31. Jev implementation-path audit and daily discoveryProjects and use cases

    Source-path audit of Jevmail, WordPress comment triage and QuantDinger distinguishes error, no-key and bypass policies from observed production use; no exact-time new item verified in the bounded 27–28 September monitor. Route lists showed no material established change.

  32. Seven-day Jev access checkpoint and daily discoveryAccess and price

    Vercel now shows rounded $0.04/M input instead of its 26 Sep Free display; exact tariff, Jev free-credit eligibility and bill unverified. Other route lists rechecked; QuantDinger documented as optional fail-open integration, not verified live operation.

  33. Post-promotion route and comparison checkAccess and price

    Vercel still displayed Free after its stated 25 Sep promotion end; effective Jev rate unknown. Cloudflare listed $0.042/M input and zero output/cached input. Added Opper route and one operator-reported Jev–Kev suite; narrowed Laya and JevAny artifact boundaries.

  34. Second full-scope Jev source passFull scope

    Expanded projects, use cases and layered alternatives; corrected Vercel Free promotion boundary and monitored exact preceding 24 hours. No authenticated call, independent benchmark or verified production deployment.

  35. Full Jev catalogue rebuild and 24-hour monitorFull scope

    Direct-source rebuild of use cases, independent Jev-like alternatives, project integrations, route pricing and access; no production product or independent benchmark verified. Monitored the preceding 24 hours with documented platform gaps.

  36. Support-ticket workload-evaluation protocolMethod

    Prepared an unfilled Choice evaluation method for one support-ticket route. It separates consecutive and challenge cohorts, freezes labels and comparators, includes review and failure accounting, and records that no Jev call or measured result occurred.

  37. Bounded freshness and change-detection passAccess and price

    Corrected the independent proxy tier record, recorded the MCA displayed Sep 23 update with exact diff unknown, dated the waitlist observation, and preserved older source-check dates and bounded negative findings.

  38. Builder fit and maintenance synthesisMethod

    The 24 September synthesis added a pilot, route fallback, substitute, wait, and stop branch; preserved bounded no-verified-open-weight-Jev and no-verified-production-product findings; proposed source refresh rules; and recorded that no independent workload measurement was run.

  39. Bounded adoption and product verification auditProjects and use cases

    Direct-source audit of killmyidea, Jev Showcase, the TypeSafe playground router, jev-usecases, and the Awesome Jev lead index. killmyidea is a direct-source application with public-deployment metadata but unverified public access and adoption; the other leads remain an unofficial integration, an explicitly mocked demo, an MIT reference harness, or an index. No verified production product was found.

  40. Alternatives substitution matrix refreshAlternatives

    Direct-source audit of exactly six named alternatives, separating model parity, System One-like interface, local/privacy boundary, workflow role, and trade-offs; no verified open-weight Jev model was found.

  41. Alternatives verification refresh (local-only)Alternatives

    A local-artifact-only refresh of the six named alternatives retained direct URLs and substitution layers, upgraded no row, marked all current project facts unknown, and made the no-verified-open-weight-Jev finding explicit without implying that none exists.

  42. Access, pricing, and availability refresh for JevAccess and price

    Current official Jev model, endpoint, aliases, price, limits, text-only input, access ambiguity, legal/data terms, OpenRouter and Cloudflare routes, inaccessible Vercel status, and the independent proxy price were checked; remaining account and gateway gaps stay explicit.

  43. Bootstrap evidence pass for JevFull scope

    First source-led catalogue of Jev use cases, open-source substitutions, directly evidenced projects, and dated access/pricing routes.

Evidence labels

LabelMeaningExample on this site
TestedThe pool ran it. Offline or mock runs are labelled as such.The Python and Node retry clients, run against a local mock server; a Claude Code hook run with no key against a closed local port; the confidence-formula check on documented examples. No live Jev call: the pool has no API key.
DocumentedOfficial TypeSafe or route-owner documentation, or project source code the pool read.The $0.042/M input price; Jevmail's retry code.
ReportedA named third party's own claim or measurement, not rerun by the pool.Opper's Jev–Kev comparison; a README's timing figures.
UnverifiedA lead that has not been checked.Proposed use-case patterns; unchecked repositories.

Release log

DateReleaseWhat changed
9 Nov 2026Benchmark grades and auto-mode 4.0.0No new question page; a new dated report. Benchmarks: B49 to B51 graded (48 graded rows), two leads, leaderboard rows re-read, an answer to "Is Jev deterministic?" and TypeSafe's cookbook figures outside the ledger (also on Jev as a judge). Coding agents: auto-mode row re-pinned at 4.0.0 (counts unchanged). Alternatives: Drex and Eevo rows. OpenCode lines, AI SDK version lines and re-dated checks on the access, route and data pages.
8 Nov 2026ohmyjev 0.14.0 and pi-jev-routerNo new question page; a new dated report. Failure map: ohmyjev row re-pinned at 0.14.0 (fbc4ea4); same terms, colours and counts (80); its bypass cell names the new shadow setting and the in-repo move skip; the Claude Code guardrails, coding agents and products pages follow with the same term, colour and pin. Coding agents: pi-jev-router row (counted on the page, off the map) and a dated lead list. Alternatives and the decision-API comparison: SubQ, Cloudflare and OpenAI data terms. Frameworks: langchain-typesafe 0.0.2 read. Use cases: TypeSafe's cookbook count. Access rechecked 8 Nov; release lines. Evidence: 8 Nov report.
7 Nov 2026Docs change, Base Labs row and Jev 1.14 vs 1.13 (regular release 27)No new question page; a new dated report. TypeSafe's documented inputs (images on both versions), Box on Jev 1.14 only, the removed language section and the docs changelog carried to limitations, what is Jev, the decision-API comparison and the other pages that describe Jev's inputs. Alternatives: Base Labs family row in the hosted layer. Versions and aliases: retitled, with a dated list of projects that stayed on 1.13. Access rechecked 7 Nov; OpenCode lines re-dated; comparison cells re-read; release lines. Evidence: 7 Nov report.
2 Nov 2026ohmyjev map row at 0.13.2 (regular release 26)No new question page; a new dated report. Failure map: ohmyjev row re-read at 0.13.2 and now describes lockdown mode (its default is off); terms and counts unchanged; checklist, copy block, self-check example and every page that shows ohmyjev updated with the same term, colour and pin. Where to use: OpenCode's free Jev ID in its own dated block outside the six-route table; access status "Is Jev free?" and data privacy (OpenCode's terms; OpenRouter's ZDR list). Access rechecked 2 Nov; Decision-1 catalog cells and the unchanged comparison cells re-dated; Laya, Effect and auto-mode release notes.
31 Oct 2026Hosted decision API cells re-read (regular release 25)No new question page; a new dated report. The decision-API comparison: 40 cells dated before 17 Oct read again (6 changed, 29 re-dated), changed-since line and copy block. Alternatives: Eevo prices; Decision-1 and Clef rows; Kev and Laya release notes. Frameworks: TanStack AI row, Pydantic AI 2.64.0 and Effect 4.1.1 checked, counts 7 / 5 / 2. Access rechecked; a resolved 30 Oct status event on the timeline; two notices. Evidence: 31 Oct report.
30 Oct 2026Currency check after 24 Oct (regular release 24)No new question page; a new dated report. Access, aliases, price, routes and catalogs rechecked and re-dated on Home, access status, versions and aliases and where to use. Changed cells on the decision-API comparison and alternatives (OpenAI retention, Microsoft-Decision-1 regions, SubQ Snap 1.1 price); Base Labs line on benchmarks; higher-limit form on errors and limits. Coding agents: one hook read in source off the map, lead triage 21–30 Oct. Version notes on SDK, framework and Claude Code pages. Evidence: 30 Oct research report.
24 Oct 2026jev-latest on Jev 1.14 and decision leaderboards (regular release 23)No new question page; a new benchmarks section and a new dated report. Versions and aliases (title now without "preview"; key finding; six-route table) and the pages that name a default model: jev-latest gives jev-1.14.0 on TypeSafe's API. Benchmarks: leaderboard section; B45–B48 graded (ledger 45); key finding reopened as a dated 24 Oct sentence. Grade lines on alternatives, the decision-API comparison (also cache pricing) and reranking. Access rechecked; three notices. Evidence: 24 Oct research report.
23 Oct 2026Jev with images (regular release 22)No new question page; a new section and a new dated report. Limitations: "Jev with images" (TypeSafe's image docs; 8 integrations; leads by date). Image lines on n8n (0.10.0), TypeScript, frameworks and errors and limits. Versions and aliases: jev-latest sentences give the 23 Oct state; 6 of 6 routes list a 1.14 ID; route image fields. Decision-API comparison: OpenAI's Decisions API generally available. jev-gateway links corrected. Access rechecked; two notices. Evidence: 23 Oct research report.
21 Oct 2026Coding agents on pi and omp (regular release 21)No new question page; a new dated report. Coding agents: ohmyjev's pi and omp row (14 counted of 27 read; 3 not settled); lead triage +13. When Jev fails: ohmyjev's row names pi and omp (80). Alternatives: Nace.AI's two sets of terms. Versions and aliases: Cloudflare 1.14 catalog line. Access rechecked; two notices. Evidence: 21 Oct research report.
20 Oct 2026Failure self-check and Drex (regular release 20)No new question page; a new interactive self-check and a new dated report. When Jev fails: self-check added; checklist question 9 (not counted). 79 → 80 in current sentences on 14 pages. Alternatives: Drex added (38 rows). Frameworks (LiteLLM 1.106.0) and n8n (activation check) rechecked. Access rechecked; four notices. Evidence: 20 Oct research report.
19 Oct 2026Failure map completed and hosted decision models (regular release 19)No new question page; a new dated report. When Jev fails: ohmyjev added (80 implementations, every failure cell recorded). Claude Code and MCP: ohmyjev's bypass cell recorded; jkudish/jev-mcp read and check commits shown separately (also on products). Alternatives: three hosted decision models added (37 rows). Content moderation and support-ticket triage: cells lead with the map terms. Access rechecked; four notices. Evidence: 19 Oct research report.
18 Oct 2026Failure terms, Claude Code guardrails and products (regular release 18)No new question page; a new dated report. When Jev fails: two Claude Code guardrails added (79 implementations, every failure cell recorded). Claude Code and MCP: three guardrails read at pinned commits and a new short answer over ten guardrails. Coding agents: key-finding split corrected; failure cells use the map terms, as on trading and fraud gates and browser and computer use. Products: 14 rows added (113), claims rechecked. Smaller edits: Opper 1.14 on Where to use and versions and aliases; Vercel 32,000 line; @typesafe-ai/sdk 0.7.0 on TypeScript and errors and rate limits; a dated line on alternatives; drift notes. Home and access status rechecked: four notices.
16 Oct 2026Bypass cells, Jev 1.14 preview and Claude Code refresh (regular release 17)No new question page; a new dated report. When Jev fails: the last 11 bypass cells and the last malformed cell read at the pinned commits (bypass recorded for 77 of 77; malformed 35 fail-closed, 24 fallback, 11 fail-open, 4 no-decision, 3 advisory, 0 not recorded), with the matching domain pages. Jev 1.14 preview: versions and aliases (a new "What changed with Jev 1.14" block), Where to use, limitations, data privacy, Python, TypeScript and the other pages that said "text only", gave rate-limit numbers or said both aliases point to 1.13. Claude Code and MCP retitled, with a version block and a 32-lead table (URL unchanged; short answer unchanged). Benchmarks: Decision Index line outside the ledger. Decision API comparison: the Jev "Images" and "Documented errors and limits" cells; prices unchanged. Home and access status: four notices; rate limits now set per organization. Smaller edits: drift notes on frameworks, LLM as a judge, coding agents, browser and computer use and n8n.
14 Oct 2026Failure cells and SDK refresh (regular release 16)No new question page; a new dated report. When Jev fails: no-key, bypass and n8n below-threshold cells settled for ticket triage, email and lead sorting, n8n, browser and computer use, jev-trader, hermes-jev-mana and jev-mcp (no key 50 fail-closed, 17 fallback, 7 fail-open, 3 advisory, 0 not recorded; bypass recorded for 66 of 77, 11 not recorded; below threshold 0 not recorded), with the matching domain pages. Python and TypeScript retitled and checked at typesafe-sdk 0.7.4, @typesafe-ai/sdk 0.6.0, ai 7.0.140 and @ai-sdk/typesafe-ai 3.0.18 (URLs unchanged). Benchmarks: B43 graded B and B44 graded A (41 graded rows: A 12, B 14, C 8, D 7). Alternatives: SubQ Snap added as a hosted row (34 substitutes, 4 hosted APIs from other providers). Decision API comparison: B44 added to the graded tests; no price or status cell changed. Smaller edits: errors and rate limits, testing without a key, Where to use and official vs reseller (SDK versions and the AI SDK rename); drift notes on Claude Code and MCP, coding agents, frameworks and model routing. Evidence: 14 Oct research report.
13 Oct 2026Decision API comparison (regular release 15)Added one page: Decisions API vs Jev (TypeSafe AI) vs Microsoft Decision 1, with Cloudflare Clef (a six-row card for 4 providers, a 13-row table with every cell dated and sourced, and a copy block that repeats the card). Benchmarks: B42 graded B (39 graded rows: A 11, B 13, C 8, D 7; 4 graded comparisons with the Decisions API). When Jev fails: no-key and bypass cells settled for model routing and content moderation (no key: 43 fail-closed, 16 fallback, 7 fail-open, 3 advisory, 8 not recorded; bypass recorded for 52 of 77); malformed answer fail-closed 34 → 35, not recorded 2 → 1. Alternatives: the three hosted rows rechecked and linked to the new page; SubQ Snap, Cygnet and GLiNER2.5-Decide added to "seen, not checked" (33 substitutes, unchanged). Corrections: Where to use (Microsoft-Decision-1 listed as Preview in Foundry at 13 Oct 2026, 05:20 UTC; listed GA on 12 Oct; cause unknown) and the Build hub (71 → 77 implementations). Smaller edits: model routing and content moderation (the same no-key and bypass cells), Jev vs LLM vs classifier (four graded comparisons with the Decisions API), limitations (input types linked to the comparison), Where to use (Bedrock and Foundry rechecked 13 Oct), versions and aliases (six routes rechecked, unchanged), official vs reseller (sites rechecked), data privacy (Vercel providers rechecked), TypeScript and Claude Code and MCP (patch releases noted), five related links (77 implementations) and Home (39 graded studies; a link to the comparison). Home and access status refreshed: status unchanged, checked 13 Oct 2026, 05:18 UTC; two notices shown as on 12 Oct; the hidden Vercel provider notice retired. All earlier URLs and reports kept.
12 Oct 2026Real-time control and Decision-1 (regular release 14)Added one page: Jev plays Doom: what real-time demos do when a Jev call fails. Release 4 is complete: 4 of 4. When Jev fails gained the domain "Real-time control (games and simulations)" with 3 rows (77 implementations, 12 domain tables). Benchmarks: B41 graded A (38 graded rows); the aggregate sentence replaced by "Against small text-generating LLMs, Jev scored ahead in 6 of 10 graded studies, mixed in 2 and not ahead in 2; 5 of the 10 are grade C." Alternatives: two Cloudflare Clef family rows (33 substitutes, 3 hosted APIs from other providers), Microsoft-Decision-1 data and evidence cells, OpenAI Decisions API data-control values. Smaller edits: the Use cases hub and Videos (real-time page links); Where to use (an Azure pointer to Microsoft-Decision-1; Bedrock and Foundry rechecked 12 Oct 2026, 05:15 UTC), with the matching dates on data privacy and official vs reseller; access status ("Is Jev free or paid?"); limitations ("Is Jev multimodal?"); Jev vs LLM vs classifier (B41); frameworks and model routing (LiteLLM 1.105.0 compared); coding agents (drift note); versions and aliases (six routes rechecked, unchanged); Home (the small-LLM count, 38 graded studies). Home and access status refreshed: status unchanged, checked 12 Oct 2026, 05:12 UTC; two notices shown as on 11 Oct, none retired. All earlier URLs and reports kept.
11 Oct 2026Benchmark grades and coding agents (regular release 13)No new page; release 4 stays at 3 of 4 (the condition for the last page, real-time control, was met in this run's research; run 14 decides). Benchmarks: three comparisons graded A from opened files (B38–B40; 37 graded rows), classifier-bakeoff still a lead, 10 new comparison repositories listed as leads. Alternatives: the Decisions API evidence cell rebuilt from graded rows only; Microsoft-Decision-1 added (31 substitutes, 2 hosted APIs from other providers). Coding agents: jev-triage, the jev-for-all launcher and hermes-jev-mana counted (24 read, 13 act). When Jev fails recounted (74 implementations). Products: 99 rows (A 86, B 12, C 1), with jev-triage, the jev-for-all launcher and hermes-jev-mana added. Smaller edits on Jev vs Laya vs Kev, Jev reranker and RAG, Claude Code and MCP, frameworks (pydantic-ai-slim 2.55.0 compared), Python and TypeScript (SDK default timeouts), model routing, Jev as a judge, versions and aliases, official vs reseller, data privacy, Where to use and the hubs. Home and access status refreshed: status unchanged, checked 11 Oct 2026, 05:17 UTC; two notices shown as on 10 Oct, none retired. All earlier URLs and reports kept.
10 Oct 2026Alternatives and coding agents (regular release 12)No run took place on 9 Oct; this release follows the 8 Oct release directly. No new page; release 4 stays at 3 of 4. Alternatives retitled "Jev alternatives (TypeSafe AI): open-source models, local runtimes and hosted APIs, and what each replaces" (URL unchanged): a new "Hosted API from another provider" layer with OpenAI's Decisions API, Jiwo 0.8B and TOD rows, Strands Decider corrected, a box for Jev-named models not published by TypeSafe, and the weights count in the layer table corrected (9 to 12). Coding agents: jev-effort-router and jev-gateway counted, two integrations not settled, 35 leads triaged. When Jev fails recounted (71 implementations). Benchmarks: jev-pulse graded, three comparisons listed as leads, mobile labels on the recompute table. Products: 96 rows (83 grade A). Smaller edits on Claude Code and MCP, Jev reranker and RAG, cost per decision, Jev vs an LLM vs a classifier, versions and aliases (OpenRouter context 64,000), official vs reseller, data privacy, Where to use, frameworks, model routing, TypeScript, Python and the hubs. Home and access status refreshed: status unchanged; the 2 Oct subprocessor notice and the 3 Oct rate-limit notice retired at the end of their seven-day shelf life, so Home shows two notices; both facts stay on data privacy and access status. All earlier URLs and reports kept.
8 Oct 2026Reranking and databases (regular release 11)Added two pages: Jev reranker and Jev for RAG and Jev inside SQL and document pipelines. Release 4: 3 of 4. When Jev fails recounted with 5 new rows and 5 changed cells (69 implementations, 11 use cases; no new domain). Coding agents: the jev-approvals Jev-error cell corrected and jev-curator's three open cells filled. Benchmarks: names replaced by organisations or blog domains, LanceDB no longer provisional, three new leads. Products: 94 rows (81 grade A). Smaller edits on Claude Code and MCP, Jev as a judge, the Use cases and Build hubs, official vs reseller, versions and aliases, data privacy, Where to use, frameworks, model routing and TypeScript. Home and access status refreshed (status unchanged). All earlier URLs and reports kept.
7 Oct 2026Judge rows and benchmarks (regular release 10, maintenance)No new page; release 4 stays at 1 of 4. When Jev fails counts the four eval-judge tools as a new "LLM as a judge" domain and two Hermes Agent plugins (64 implementations, 11 use cases). Coding agents: all 5 open cells settled (8 of 18 counted). Benchmarks retitled "Jev benchmarks (TypeSafe AI): is Jev accurate?" (URL unchanged), with 9 new rows, one regrade and two changed short-answer clauses. Smaller edits on Jev as a judge, Products, official vs reseller, versions and aliases, data privacy, TypeScript, frameworks and Claude Code and MCP. Home and access status refreshed (unchanged status; the WorkflowEvals notice retired). All earlier URLs and reports kept.
6 Oct 2026Coding agents and model versions (regular release 9)Added two pages: How to use Jev with Codex, OpenCode, Cursor and Hermes Agent and Jev 1.13 and jev-latest: model IDs and aliases by route. Release 3 is complete: 9 of 9 planned pages. Release 4: 1 of 4. Claude Code and MCP gained two guard rows. When Jev fails gained 7 rows and a "Coding agents" domain (58 implementations, 10 use cases). Refreshed Products (92 rows, 75 grade A). Where to use and data privacy: Vercel now lists two providers for Jev. Drift notes on model routing (LiteLLM 1.104.0), framework integrations and TypeScript. Failure colours made the same on every page that shows a failure verdict (see the 6 Oct release notes). Home and access status: status unchanged. All earlier URLs and reports kept.
5 Oct 2026Browser, frameworks and Bedrock (regular release 8)Added two pages: Jev browser use and computer use and Jev LangChain and other framework integrations; added the section Is Jev on Amazon Bedrock or Azure AI Foundry? and a route-matrix row to Where to use. Release 3 now has 8 of its 9 planned pages. When Jev fails gained 12 rows (51 implementations, 9 use cases) and a copyable checklist. Refreshed Products (72 rows, 55 grade A). Small additions: a Zapier note (n8n), an EU-hosting line (data privacy), the Strands Decider 2B row (alternatives) and tool-documentation links (Jev as a judge). Home and access status: status unchanged. All earlier URLs and reports kept.
4 Oct 2026Trading, failures and judges (regular release 7)Added three pages: Jev trading bot and fraud gates, when Jev fails (the failure checklist across 7 use cases) and Jev as a judge; release 3 now has 6 of its 9 planned pages. Refreshed Products (67 rows, 47 grade A). Home: the subprocessor notice replaces the WorkflowEvals notice on Home (still current, not shown); status unchanged. Added TypeSafe's subprocessor list to data privacy. Corrections on the email page (two relabels, cost range), Claude Code and MCP (advisory wording, v5.6.1 note) and QuantDinger links (pinned to a5a9f4c). All earlier URLs and reports kept.
3 Oct 2026Email, Products and n8n (regular release 6)Added two pages: Jev email classifier: email and lead sorting and Jev n8n node. Refreshed the Products hub (61 rows, 39 grade A; heading now starts "Jev examples"; URL unchanged). Checked oh-my-claudecode v5.6.0 on Claude Code and MCP (failure path unchanged; 2,000 ms default timeout). Refreshed Home and access status (status unchanged; rate-limit notice replaced: 80 requests per second). Updated errors and rate limits, limitations, Where to use, official vs reseller (n8n npm publisher answered) and the Use cases hub (Jevmail row corrected to "no decision"). All earlier URLs and reports kept.
2 Oct 2026Routing and moderation (regular release 5)Added two pages: Jev Router or your own model router? and content moderation; model routing completes the second set of planned question pages. Replaced the jev-mcp drift note with a version check (failure path unchanged at 0.13.0). Refreshed Home and access status (unchanged status, new check time, uptime 99.826%). Updated the ai drift note, the Jev Router row and check dates on Where to use, data-label check dates on data privacy, and the Use cases hub (routing row now code-backed; WordPress row pinned). All earlier URLs and reports kept.
1 Oct 2026Data privacy and alternatives (regular release 4)Added three pages: Jev vs Laya vs Kev, data handling by route and support-ticket triage. Refreshed Home and access status (unchanged status, new check time, narrowed rate-limit date). Corrected the Vercel data cell and two data caveats on Where to use, the Kev and Laya rows and the he-jev note on Alternatives, the rate-limit dating, and two version-drift notes. All earlier URLs and reports kept.
30 Sep 2026Evaluation evidence (regular release 3)Fixed both tools (fit check and cost worksheet), which did not load their scripts at the address the site serves them. Reworked Home: access status, dated notices, short answer and direct task paths; the dated "What changed" list moved to this page. Added two pages: benchmarks and Jev vs an LLM vs a classifier. Corrected the rate limits, added a TypeScript version-drift note, two official surfaces, 29 Sep status-page events and the alexmolas.com note. All earlier URLs and reports kept.
29 Sep 2026Integration trust (release 2, first part)Added the Build hub and three Build pages: TypeScript, Claude Code and MCP guardrails, and testing without a key. Corrected the oh-my-claudecode timeout note on the errors page and renamed and regraded the jev-mcp row on Products, adding five integration rows. Re-dated route and access fields (values unchanged). Added links from Home, Use cases, Python, Confidence thresholds, Videos, Where to use and Access status. All earlier URLs and reports kept.
28 Sep 2026Question pages, release 1Site reorganised by reader question. Rebuilt Home and the Use cases, Alternatives, Products and Where-to-use hubs; extended What is Jev and this page; added eight pages (limitations, access status, official vs reseller, cost per decision, Python, errors and rate limits, confidence thresholds, videos) and a cost worksheet. Corrected two project labels. All earlier URLs and reports kept.
20–28 Sep 2026Daily catalogue releasesOne release per report listed above, on the previous site layout.

What changed, by date

This dated list used to be on the home page. The home page now shows only current notices; past changes stay here.

9 November 2026

  • Access status rechecked 9 Nov 2026, 11:07 UTC; the previous check was on 8 Nov 2026, 05:17 UTC, with no check in between. Signups open, no new-user credit, API operational; price, rate-limit wording and aliases unchanged; the status page lists no event after 30 Oct; TypeSafe's docs files identical to 8 Nov. No notice added or retired.
  • Changed on benchmarks: three studies graded from their files (B49 grade B, B50 grade A, B51 grade B); 48 graded rows (A 15, B 16, C 10, D 7); the small-LLM count (6 of 10) is kept as counted through 24 Oct. Jev Decision Index re-read (scores unchanged, 275 entries, no position given); Artificial Analysis re-read (Jev 1.14 66.96, Jev 1.13 no longer listed; index changed).
  • Changed on coding agents: auto-mode read at 4.0.0 (7b246a7); failure path unchanged; bypass lists a reply-channel allow; still not settled; counts and the failure map unchanged.
  • Changed on alternatives: Nace.AI's four legal documents (dated 6 Nov 2026) now agree; Drex Max claim now 10.1 points over Jev 1.14; Eevo's announcement table now includes Jev 1.14.
  • Changed on data privacy and where to use: OpenCode's Terms effective 8 Nov 2026 with a Connected Service exception; still not reconciled with its docs. OpenRouter's ZDR list 1,109.

8 November 2026

  • Access status rechecked 8 Nov 2026, 05:17 UTC; the previous check was on 7 Nov 2026, 15:29 UTC, with no check in between. Signups open, no new-user credit, API operational; price, rate-limit wording and aliases unchanged; the status page lists no event after 30 Oct. No notice added or retired; two shown.
  • Changed on the failure map: ohmyjev read at 0.14.0 (published 3 Nov 2026). Its four other failure cells are unchanged and lockdown is still off by default; its bypass cell now names an opt-in shadow setting (off by default) that logs what lockdown would block without blocking it, and in-repo mkdir, touch, cp and mv, which now skip Jev. Every total unchanged (80).
  • New on coding agents: @caph42/pi-jev-router 2.0.0 for pi, counted on the page (29 read in source, 15 counted), not on the failure map; a dated list of 19 coding-agent projects seen 24 Oct–8 Nov 2026 (leads, not read in full).
  • Changed on alternatives: SubQ's docs and Terms now agree on 60-day retention (they disagreed until at least 14 Oct); Cloudflare's data-usage page lists a third retention exception; OpenAI's data controls unchanged.
  • TypeSafe's docs index added one cookbook, "Verifying LLM answers against their source"; use cases gives the new count (6 patterns, 26 cookbooks, 2 demos).

7 November 2026

  • Access status rechecked 7 Nov 2026, 15:29 UTC; the previous check was on 2 Nov 2026, 10:47 UTC, with no check in between. Signups open, no new-user credit, API operational; price, rate-limit wording and aliases unchanged; the status page lists no event for 31 Oct to 7 Nov. The console sign-in page answered again (HTTP 200). No notice added or retired; two shown.
  • Changed in TypeSafe's docs (first observed 7 Nov; after the 2 Nov reads): the models page lists "Text, images" as the input of both versions, adds a Box question type that needs Jev 1.14 and drops its "Language support" section; a new docs changelog dates the Jev 1.14 entry 14 Oct 2026. The Images page still calls images not ready for production. Carried to limitations and the other pages that describe Jev's inputs.
  • New on alternatives: Base Labs (Baseten's research lab), mammoth-1, elephant-1 and joey-1, as one hosted-API row.
  • Changed on versions and aliases: new title "Jev 1.14 vs 1.13" and a dated list of projects that stayed on 1.13 (Reported).
  • TypeSafe's blog has a new post dated 3 Nov 2026 ("Don't Call It a Decision Model", about naming); not a notice.

2 November 2026

  • Access status rechecked 2 Nov 2026, 10:47 UTC; the previous check was on 31 Oct 2026, 05:11 UTC, with no check in between. Signups open, no new-user credit, API operational; price, rate-limit wording and aliases unchanged; the status page lists no event for 31 Oct, 1 Nov or 2 Nov. The console sign-in pages returned an HTTP 403 block page and could not be read. No notice added or retired; two shown.
  • Changed on the failure map: ohmyjev re-read at 0.13.2 (commit d307c39). Its gates now run only with lockdown on, and lockdown is off by default; the row describes lockdown mode, where its five terms are unchanged; the bypass cell changed in content. No count changed (80 implementations).
  • New on where to use: OpenCode's jev-1.13-free ("Free" input and output, "available on OpenCode for a limited time"), a third-party route not run by TypeSafe and not one of the six compared routes.
  • Changed: OpenRouter's ZDR endpoint list now names typesafe/jev-1.14-20261014 as well as typesafe/jev-1.13-20260917 (1,094 endpoints at 10:59 UTC; 941 on 10 Oct); status-page uptime figures; Laya v0.4.21, @effect/ai-typesafe 4.1.2 and auto-mode 3.1.6 released (not read).
  • Rechecked and re-dated: the six route listings, the unchanged comparison cells, Microsoft-Decision-1's catalog cells (unchanged at 2 Nov, 11:00 UTC).

31 October 2026

  • Access status rechecked 31 Oct 2026, 05:11 UTC, one day after the previous check: signups open, no new-user credit, API operational; price, rate-limit wording and aliases unchanged. The status page lists a resolved "System overload" on the us-east-1 endpoint on 30 Oct (degraded for 1 hour and 15 minutes, resolved 21:15 UTC), now on the access timeline. The home page notice about jev-latest moving to jev-1.14.0 ended its seven-day shelf life; the fact stays on versions and aliases. Two notices shown.
  • Changed on the decision-API comparison and alternatives, each dated as a range: Microsoft-Decision-1 rate limits (2,160–15,000 requests and 2.16M–15M tokens per minute by tier; was 60–900 requests per minute), Cloudflare's Workers AI zero-data-retention and no-training wording (page updated 16 Oct), Clef-flash hosts on OpenRouter (Makora and DeepInfra added), OpenAI's inputs and regions wording, Upright's Eevo prices. Kev and Laya: new releases noted, not read.
  • Rechecked and re-dated: 29 comparison cells, the hosted rows for d1-ultra, Drex, Mouse-1 and SubQ Snap, the six route listings (Opper's typesafe/jev-latest still maps to jev-1.13.0) and the Bedrock, Microsoft Learn Foundry and Workers AI pages (Jev not listed at 31 Oct, 05:14 UTC).
  • Updated: framework packages: TanStack AI read at commit 5651e07 (raises to your code; 7 packages read in source, 5 raise errors to your code); Pydantic AI 2.64.0 and Effect 4.1.1 checked, failure paths unchanged.

30 October 2026

  • Access status rechecked 30 Oct 2026, 05:14 UTC, after the check of 24 Oct, 05:12 UTC (no check in between): signups open, no new-user credit, API operational; the status page's history lists no incident dated 15–30 Oct. Price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and alias targets unchanged; jev-1.13.0 still has no deprecation date. Higher limits are now requested through a form (changed between 24 Oct, 05:12 UTC and 30 Oct, 05:14 UTC). Home keeps three notices; none added or retired. Details: access timeline.
  • Rechecked and re-dated: the six route listings (Opper's typesafe/jev-latest still maps to jev-1.13.0), the Bedrock, Microsoft Learn Foundry and Workers AI pages (Jev not listed at 30 Oct, 05:17 UTC; the Azure AI Foundry catalog page could not be read and keeps its 16 Oct date), and the unchanged comparison-card cells.
  • Changed: on the decision-API comparison and alternatives, OpenAI's Decisions API retention eligibility (between 13 and 30 Oct), Microsoft-Decision-1 regions (page dated 19 Oct, first seen 30 Oct) and SubQ Snap 1.1 at $0.04 per million input tokens in SubQ's docs (between 14 and 28 Oct; the launch post still says $0.05). Drex's pricing docs list drex-max at $0.03 (was $0.06), a new drex-max-v1.1 at $0.03 and a $5.00 signup balance (between 20 and 30 Oct). On benchmarks, Base Labs' free-week wording was not found on Baseten's pricing and mammoth-1 pages (30 Oct).
  • Updated: coding agents: acmelabs-15/jev-gate read in source and shown under "Read in source; not on the failure map yet" (28 integrations read in source: 14 counted, 10 not counted, 4 not settled); a dated lead triage for 21–30 Oct. The failure map still has 80 implementations. Version notes for ohmyjev 0.13.2, auto-mode 3.1.5 and newer SDK and framework releases. Evidence: 30 Oct research report.

24 October 2026

  • Access status rechecked 24 Oct 2026, 05:12 UTC: signups open, no new-user credit, API operational; price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and the rate-limit wording unchanged. On TypeSafe's API jev-latest now points to jev-1.14.0: first seen at 05:12 UTC; at 23 Oct 2026, 05:45 UTC it still pointed to jev-1.13.0, and TypeSafe published no dated statement of the move. jev-preview also points to jev-1.14.0 ("There is no preview build available right now"); jev-1.13.0 is still listed with its own ID and price, with no deprecation date. The status page lists no new event; its API uptime figure read as the 90-day window is 99.831% (99.835% on 23 Oct). Home adds a notice for the move, so it shows three (no credit for new users, the jev-latest move, lookalike resellers); none was retired. Details: access timeline.
  • Updated: versions and aliases and the pages that name a default model or describe the move as upcoming (where to use, what is Jev, limitations, Python, TypeScript, Claude Code and MCP, confidence thresholds): on TypeSafe's API an unpinned request now gets Jev 1.14. Opper's typesafe/jev-latest still mapped to jev-1.13.0 at 05:15 UTC; OpenRouter, Vercel, Netlify and Cloudflare publish no alias target (listings, not tested requests). Dated observations are kept.
  • Updated: benchmarks. New section "Jev benchmark leaderboards: who runs them and what you can check": the Jev Decision Index, the Artificial Analysis Decision Index, JevMan and JevBench, each with its operator, version, what is published and Jev's listed position; nine vendor and press claims checked against the leaderboards' own data. Leaderboards sit outside the ledger and carry no grade. Four studies graded: B45 and B46 (grade A), B47 and B48 (grade C); the ledger has 45 graded rows: A 14, B 14, C 10, D 7.
  • Updated: Decisions API vs Jev vs Microsoft Decision 1, alternatives and reranking and RAG: the new grades with the same flags; the Decisions API's cached-input and cache-write prices since general availability (the 13 Oct beta wording is kept as dated). Evidence: 24 Oct research report.

23 October 2026

  • Access status rechecked 23 Oct 2026, 05:18 UTC: signups open, no new-user credit, API operational; price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and the rate-limit wording unchanged. jev-latest still points to jev-1.13.0 and jev-preview to jev-1.14.0, nine days after TypeSafe's 14 Oct post said the move would come "in a week"; no date for the move or for retiring 1.13 has been published. The models, Images and Box pages and the quickstart changed size (the changed lines were not identified); their access, price, alias, input and limit text reads the same. The status page lists no new event; its API uptime figure read as the 90-day window is 99.835% (99.826% on 21 Oct). Home keeps its two notices (no credit for new users, lookalike resellers); none was retired. Details: access timeline.
  • Updated: limitations. New section "Jev with images: does your SDK or integration send them?". Of 8 integrations read in package files, 5 send images to Jev and 2 of those add TypeSafe's opt-in header for you; pydantic-ai refuses images, and both LangChain packages pass them as text. TypeSafe's image docs are quoted with their times (PNG or JPEG only; inline or multipart; 400 without the header or for a bad image, 413 over a size limit; the whole request fails), including the models page's "Text" input next to the images page's "both accept images". No image was sent.
  • Updated: n8n. TypeSafe's node 0.10.0 (commit ec81714) keeps the same failure behaviour in every cell and adds an Images field that sends PNG or JPEG with the opt-in header; the failure map's counts are unchanged. Image lines also on TypeScript, frameworks (pydantic-ai-slim 2.60.0 still refuses images) and errors and limits.
  • Updated: versions and aliases and the pages that described the jev-latest move as upcoming: on 23 Oct 2026, 05:45 UTC, it still pointed to jev-1.13.0. All six routes list a Jev 1.14 ID (Cloudflare's catalog included); a versioned ID can be written on 6 of 6 routes for 1.14 and 5 of 6 for 1.13. Route image fields re-read.
  • Updated: Decisions API vs Jev vs Microsoft Decision 1 and alternatives. OpenAI's Decisions API became generally available on 22 Oct 2026; same model and price; up to 400 questions per request (200 in the beta).
  • Corrected: the pinned jev-gateway links on when Jev fails and coding agents now point to the repository named in its npm record, at the same commit. Evidence: 23 Oct research report.

21 October 2026

  • Access status rechecked 21 Oct 2026, 05:11 UTC: signups open, no new-user credit, API operational; price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and the rate-limit wording unchanged. jev-latest still points to jev-1.13.0 and jev-preview to jev-1.14.0, a week after TypeSafe's post said the move would come "in a week"; no date has been published. The models page and quickstart changed size (the changed lines were not identified); their access, price, alias and limit text reads the same. The status page lists no new event; its 90-day API uptime figure still reads 99.826%. Home now shows two notices (no credit for new users, lookalike resellers): the Jev 1.14 preview and the experimental features notices came off at the end of their seven-day shelf life; both facts stay on versions and aliases and limitations. Details: access timeline.
  • Updated: coding agents. ohmyjev's pi and omp adapters (78bb82c) are a counted row: on every Jev failure the gate returns without blocking, so the host runs the tool (fail-open; acts-anyway below threshold). 27 read in source; 14 act on Jev's answer (3 fail-closed, 9 fallback, 2 fail-open), 10 do not, 3 are not settled (Hoppin-Sorter/jev-router, auto-mode 3.0.0, jcode). 13 new leads classed from one file each; 5 of them act on Jev's answer and wait for a full read.
  • Updated: when Jev fails. ohmyjev's row names its pi and omp adapters, read at the same commit with the same five terms; the map still counts 80 and no count changed. One file reference in its bypass cell corrected.
  • Updated: alternatives. Nace.AI's website Terms and Privacy Policy (9 Oct) say customer content may be used for training by default (opt-out) and kept up to 30 days; its Console Service Terms and Data Processing Addendum (19 Oct) say no training unless you opt in and deletion within 48 hours (paid) or 7 days (free). Which one controls is not stated. drex-v1.5's two prices are in different units and are shown side by side.
  • Updated: versions and aliases. Cloudflare's AI model catalog lists typesafe/jev-1.14.0 through AI Gateway (first seen by this catalogue on 21 Oct 2026; it may have been listed earlier; not tested). Evidence: 21 Oct research report.

20 October 2026

  • Access status rechecked 20 Oct 2026, 05:13 UTC: signups open, no new-user credit, API operational; price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and the rate-limit wording unchanged. jev-latest still points to jev-1.13.0 and jev-preview to jev-1.14.0; TypeSafe's 14 Oct post says jev-latest moves "in a week", and no date for the move has been published. TypeSafe split its docs API reference into separate pages (System One, List models, Request rules and errors); the models and Images pages differ from 19 Oct only in links to the new errors page. The status page lists no new event; its 90-day API uptime figure still reads 99.826%. Home keeps its four notices (no credit for new users, the Jev 1.14 preview, lookalike resellers, the experimental features); none was retired. Details: access timeline.
  • New: a failure self-check on when Jev fails. For each of five failure conditions you pick what your code does and see how many of the 80 implementations made the same choice, with up to three example rows and the matching checklist question. It runs in your browser, sends nothing, and does not read or test code. Counts as recomputed 19 Oct (for example, with no API key 0 of 80 return no decision).
  • Updated: the checklist has nine questions; question 9 asks whether your code logs or shows when it falls back or fails open (not counted: the map does not record it). The copyable checklist repeats it.
  • Corrected: current sentences that still gave the 79-implementation count, or its category numbers, now give 80 on when Jev fails, the use-cases and build hubs, coding agents, Jev as a judge, content moderation, support-ticket triage, model routing, n8n, confidence thresholds, testing without a key, errors and rate limits, the 9 Oct evidence notes and the home page.
  • Updated: alternatives. Drex (Nace.AI) added: a hosted decision API ("Generally available"; $0.06 per 1M input tokens for drex-v2.0 per its pricing docs) with open weights under a modified Open RAIL-M licence; 38 rows (8 hosted). Its "wire-compatible with Jev" statement and its claims against Jev are the vendor's own (Reported); no graded comparison exists.
  • Rechecked: frameworks (LiteLLM 1.106.0: the Jev classifier's failure path is unchanged), n8n (at n8n@2.44.3 a workflow whose node lacks a required credential cannot be activated; an already active one fails at execution), and Jev vs decision APIs (OpenAI's guide now accepts image URLs; no card cell changed). Evidence: 20 Oct research report.

19 October 2026

  • Access status rechecked 19 Oct 2026, 05:22 UTC: signups open, no new-user credit, API operational; price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and the rate-limit wording unchanged. jev-latest still points to jev-1.13.0 and jev-preview to jev-1.14.0; TypeSafe's 14 Oct post says jev-latest moves "in a week", and no date for the move has been published. TypeSafe's models, Images, Box, legal and quickstart pages are unchanged since 16 Oct. The status page lists no new event; its 90-day API uptime figure still reads 99.826%. Home keeps its four notices (no credit for new users, the Jev 1.14 preview, lookalike resellers, the experimental features); none was retired. Details: access timeline.
  • Updated: when Jev fails. ohmyjev (78bb82c; Claude Code module with Bash, Write/Edit and exfiltration gates) joins the Claude Code / MCP guards domain (8 → 9) after its pi and omp adapters were read in full, so the map counts 80 implementations (was 79), with every failure cell recorded for all 80. On a Jev error: 40 fail-closed, 22 fallback, 10 fail-open, 5 no-decision, 3 advisory (was 40 / 22 / 9 / 5 / 3); the action goes ahead in 19 of 80. No key 51 / 17 / 9 / 3; malformed 36 / 24 / 13 / 4 / 3; below threshold 25 / 17 / 16 / 12 / 5 / 5; bypass recorded 80 of 80 (13 "none found", 67 with skip lists).
  • Updated: Claude Code and MCP. ohmyjev's bypass cell is recorded, with two more skip paths in the Claude Code module; the short answer over ten guardrails is unchanged. jkudish/jev-mcp rows now cite the commit each cell was read at (53fe575, 0.11.0) and, separately, the later checks (5e0ca5c, 0.13.0; 86eae78, 0.14.1), here and on when Jev fails and products.
  • Updated: alternatives. Liquid AI d1-ultra, Mouse-1 and Eevo (Upright) added as hosted APIs from other providers, each with its status, published prices and data terms quoted from the provider's own pages; 37 rows (7 hosted). Their claims against Jev are the vendors' own (Reported); no graded comparison exists. The 18 Oct "not yet evaluated" line and the 16 Oct Eevo "seen, not checked" line are replaced by the rows.
  • Updated: content moderation (11 cells) and support-ticket triage (9 cells) lead with the failure checklist's terms; the detail moved to the line below. No outcome changed.
  • Updated: products. Beside the VentureBeat-relayed Vercel adoption claim, Vercel's own AI Gateway production-index figures (35.3% of paid AI Gateway teams in Jev's first two weeks; 0.27% of gateway spend over that period; Reported, Vercel sells the gateway).
  • Rechecked: versions and aliases (aliases and route listings unchanged; 1.14 still not listed on Cloudflare, Bedrock, Foundry or Workers AI) and n8n (missing-credential path unchanged at n8n@2.44.3). Version notes: litellm 1.106.0, @openclaw/typesafe 2026.10.5 and oh-my-claudecode v5.7.4 and v5.7.5 released; not re-read. Evidence: 19 Oct research report.

18 October 2026

  • Access status rechecked 18 Oct 2026, 05:17 UTC: signups open, no new-user credit, API operational; price ($0.042 per million input tokens for jev-1.14.0 and jev-1.13.0) and the rate-limit wording unchanged. jev-latest still points to jev-1.13.0 and jev-preview to jev-1.14.0; TypeSafe's 14 Oct post says jev-latest moves "in a week", and no date for the move has been published. TypeSafe's models, Images, Box and quickstart pages are byte-identical to 16 Oct. The status page lists no new event; its 90-day API uptime figure reads 99.826% (99.840% on 16 Oct). Home keeps its four notices (no credit for new users, the Jev 1.14 preview, lookalike resellers, the experimental features); none was retired. Details: access timeline.
  • Updated: when Jev fails. claude-model-guard (bced870) and claude-jev-advisor (f2c1791) join the Claude Code / MCP guards domain (6 → 8), so the map counts 79 implementations (was 77), with every failure cell recorded for all 79. On a Jev error: 40 fail-closed, 22 fallback, 9 fail-open, 5 no-decision, 3 advisory (was 39 / 22 / 8 / 5 / 3); the action goes ahead in 18 of 79. No key 51 / 17 / 8 / 3; malformed 36 / 24 / 12 / 4 / 3; below threshold 25 / 17 / 15 / 12 / 5 / 5; bypass recorded 79 of 79 (13 "none found"). A line explains why second-look is counted in the map's browser domain but shown separately on the browser page.
  • Updated: Claude Code and MCP. Three guardrails read at pinned commits: claude-model-guard, claude-jev-advisor and ohmyjev. New short answer over ten guardrails: two hold or stop when Jev fails, one falls back, two ship in a shadow or log-only mode, one is advisory, three fail open and one only returns a verdict (replaces "1 of 7"). ohmyjev is shown with all its cells and is not yet in the map count: its bypass cell is a partial read, not counted as recorded.
  • Corrected: coding agents. The key finding said the 13 integrations that act on Jev's answer split "5 let the call run, 3 hold, 5 fall back" on a Jev error; the cells say 3 fail-closed, 9 fallback and 1 fail-open. 8 bypass cells now say "recorded" and 4 cells move "(configured model)" to the detail line.
  • Updated: trading and fraud gates (16 cells use the map terms; legend line; a secondary line: on a Jev error the 4 trading bots are 2 fail-closed and 2 fallback, and both fallbacks can still place a new order) and browser and computer use (6 cells; a line on second-look). Outcomes unchanged.
  • Updated: products. 14 rows added, each with the Jev call read at a pinned commit: 3 Claude Code guardrails, 3 game and robot demos and 8 projects. Ledger 113 (A 100, B 12, C 1; 99 on 11 Oct); independently verified in production 0 of 113. Of 7 published claims, 5 are online with the same text and 2 were not rechecked (VentureBeat returned HTTP 429).
  • Updated: versions and aliases and Where to use. Opper lists typesafe/jev-1.14.0 (alias typesafe/jev-preview); its typesafe/jev-latest still points to 1.13.0. Vercel's 32,000-token context field equals TypeSafe's per-question limit; Vercel does not define the field, so this is not confirmed.
  • Updated: TypeScript and errors and rate limits. Checked at @typesafe-ai/sdk 0.7.0: failure path unchanged from 0.6.0; input-validation errors now reject the returned promise; request bodies of 1 KiB or more are compressed on Node 22.15+; the package needs Node 22.
  • Added: a dated line on alternatives: Liquid AI d1-ultra and Mouse-1 launched; not yet evaluated (launch-page prices; performance claims are vendor claims). Evidence: 18 Oct research report.

16 October 2026

  • Access status rechecked 16 Oct 2026, 05:21 UTC: signups open, no new-user credit, API operational; price unchanged. TypeSafe released Jev 1.14 (jev-1.14.0) under jev-preview (post dated 14 Oct); jev-latest still points to jev-1.13.0. The models page no longer lists rate-limit numbers ("Rate limits are set per organization"). Image input, a Box question type and 262k long context are opt-in experiments. The console now shows a sign-in page instead of HTTP 403. A 45-minute latency event on the us-east-1 endpoint was resolved on 14 Oct, 22:52 UTC. Home now shows four notices: the two new ones cover Jev 1.14 and the experimental features. Retired from the notice list: TypeSafe's Series A announcement of 9 Oct (seven-day shelf life ended 16 Oct; checked 05:24 UTC; never shown on Home).
  • Updated: when Jev fails. The last 11 bypass cells (judge tools 4, frameworks 3, browser agents 3, email 1) and the last malformed cell were read at their pinned commits; every failure cell is now recorded for all 77. Bypass recorded 66 → 77 (13 "none found", 64 with skip paths); malformed answer: competitor-hunter is now no-decision (no-decision 3 → 4, not recorded 1 → 0). Four bypass cells state a narrower read scope (editorial decision, 16 Oct). JevPlaysSnakeGame's bypass cell now rests on the whole repository (still none found). Eight cell texts on seven implementations corrected, terms unchanged. The domain pages carry the same cells.
  • Updated for Jev 1.14: versions and aliases (both model IDs, the aliases, routes for 1.14), Where to use, limitations ("Is Jev multimodal?": text by default; images an opt-in experiment, not production-ready per TypeSafe), data privacy (image data terms not stated on the images page), Python (typesafe-sdk 0.7.5) and TypeScript (@typesafe-ai/sdk 0.7.0, @ai-sdk/typesafe-ai 3.0.20). Earlier dated observations are kept.
  • Updated: Claude Code and MCP, retitled "Jev Claude Code and MCP guardrails (TypeSafe AI): which hook, plugin or server fails open?" (URL unchanged). Versions checked 16 Oct 2026 (oh-my-claudecode 5.7.3: failure path unchanged); 32 leads classified (16 hooks or plugins that allow, block, hold or reroute; 12 return Jev's answer to the agent; 1 skill only; 3 not Jev). Short answer unchanged: 1 of 7 stops the agent when Jev fails.
  • Updated: benchmarks. The Artificial Analysis Decision Index v1.0 (13 Oct 2026) added as a Reported line outside the graded ledger: Jev 1.13 58, GLiDE 54, Microsoft-Decision-1 54; the model ID or route is not stated. Ledger unchanged (41 graded).
  • Updated: Decisions API vs Jev vs Microsoft Decision 1. The Jev "Images" and "Documented errors and limits" cells and the copy block; prices and the other columns unchanged. Alternatives: Eevo (Upright) seen, not checked.

14 October 2026

  • Access status rechecked 14 Oct 2026, 16:46 UTC: unchanged, including the model IDs on all six routes. Two notices shown as on 13 Oct. Retired from the notice list: TypeSafe's Jack & Jill case study (seven-day shelf life ended 14 Oct; checked 16:47 UTC; never shown on Home). The post is still online and stays a vendor lead on benchmarks. The status page now also lists two regional API entries (us-east-1 and us-east-2), with no access change.
  • Updated: when Jev fails. No-key, bypass and n8n below-threshold cells settled for ticket triage, email and lead sorting, n8n, browser and computer use, and three single rows (jev-trader, hermes-jev-mana, jev-mcp), read at the published pins; n8n's own missing-credential path read at release n8n@2.44.1. No key: not recorded 8 → 0; fail-closed 43 → 50, fallback 16 → 17. Bypass recorded 52 → 66, not recorded 25 → 11 (judge tools 4, frameworks 3, browser agents 3, email 1); a cell counts as recorded only when every entry point was read. Below threshold: no-threshold 20 → 25, not recorded 5 → 0. zammad-jev-dispatcher's malformed cell now separates a held off-list choice from the 502 for a missing probabilities map; jev-for-chrome's below-threshold cell notes a second look for a final 'done' or 'blocked' below 0.5.
  • Updated: Python and TypeScript, retitled (URLs unchanged). Checked at typesafe-sdk 0.7.4 (9 Oct 2026; defaults unchanged; example pins now 0.7.4), @typesafe-ai/sdk 0.6.0, ai 7.0.140 and @ai-sdk/typesafe-ai 3.0.18. The AI SDK renamed experimental_evaluate to experimental_decide in ai 7.0.128; the old name still works at 7.0.140 as a deprecated alias. With the AI SDK provider, a missing key throws LoadAPIKeyError at the first call.
  • Updated: benchmarks. B43 (kprzybys/snap-vs-jev-swe, SubQ Snap vs Jev on file localization) graded B, with a conflict-of-interest flag: the operator takes part in Subquadratic's Snap creator challenge, which has prizes. B44 (dante01yoon/decision-model-arena, Tetris round robin, games won of 48: Decision-1 15, Jev 12, Decisions API 11, Laya 8, 2 draws) graded A. The ledger has 41 graded rows (A 12, B 14, C 8, D 7). Graded comparisons with OpenAI's Decisions API: 5; with Microsoft-Decision-1: 2; with SubQ Snap: 1.
  • Updated: alternatives. 33 → 34 listed substitutes; hosted APIs from other providers 3 → 4 (SubQ Snap, from its docs, Terms and launch post, with two retention texts that disagree). SubQ Snap removed from "seen, not checked". Decisions API and Microsoft-Decision-1 rows: B44 added (five and two graded comparisons).
  • Updated: Decisions API vs Jev vs Microsoft Decision 1. B44 (grade A, Tetris round robin) added to the graded tests for Jev, the Decisions API and Microsoft-Decision-1, and to the copy block. "Also launched": SubQ Snap re-dated and linked to its new row. Prices and status cells unchanged (checked 14 Oct 2026, 16:48 UTC).

13 October 2026

  • New: Decisions API vs Jev (TypeSafe AI) vs Microsoft Decision 1, with Cloudflare Clef. Graded results depend on the task: OpenAI's Decisions API ahead on RPG move accuracy (B38); Jev on Portuguese exact judge score (B39); Jev and Microsoft-Decision-1 not separated on routing macro-F1, both ahead of OpenAI (B41).
  • Updated: benchmarks. B42 (etsabary/luna-decisions-benchmark) graded B, with seven flags; the ledger has 39 graded rows (A 11, B 13, C 8, D 7). Graded comparisons with OpenAI's Decisions API: 4 (B38, B39, B41, B42).
  • Updated: when Jev fails. With no API key, of 77 implementations: 43 fail closed, 16 fall back, 7 fail open, 3 are advisory, 8 not recorded (18 not recorded on 12 Oct). Bypass recorded for 52 of 77 (44 on 12 Oct). Malformed answer: fail-closed 34 → 35, not recorded 2 → 1.
  • Correction: Where to use. Microsoft-Decision-1 is listed as Preview in Foundry at 13 Oct 2026, 05:20 UTC; it was listed Generally available on 12 Oct; the cause is unknown.
  • Access status rechecked 13 Oct 2026, 05:18 UTC: unchanged. Two notices shown as on 12 Oct. Retired from the notice list: Vercel adding TypeSafe as a second Jev provider (shelf life ended 13 Oct, 05:15 UTC; checked 05:19 UTC; never shown on Home). The fact is unchanged and stays on Where to use and data privacy.

12 October 2026

  • New: Jev plays Doom: what real-time demos do when a Jev call fails. Of 3 published game and robot-simulation demos that call Jev, read at pinned commits, 2 keep their last action when a call fails (doom-jev, JevPlaysSnakeGame) and 1 stops (jev-robotics-demo); none is a product.
  • Updated: when Jev fails. Of 77 Jev implementations read in source, on a Jev error 39 stop or hold the action, 22 fall back, 8 let it through, 5 return no decision and 3 are advisory. The action goes ahead in 17 of 77; keeping the last action counts as the action going ahead.
  • Updated: benchmarks. B41 graded A, with four flags; the ledger has 38 graded rows (A 11, B 12, C 8, D 7). Against small text-generating LLMs, Jev scored ahead in 6 of 10 graded studies, mixed in 2 and not ahead in 2; 5 of the 10 are grade C.
  • Updated: Jev alternatives. Of 33 listed substitutes: 13 model weights, 11 local runtimes, 4 request adapters, 2 classifier/LLM approaches, 3 hosted APIs from other providers (OpenAI, Microsoft, Cloudflare).
  • Access status rechecked 12 Oct 2026, 05:12 UTC: unchanged. Two notices shown as on 11 Oct; none retired.

11 October 2026

  • Updated: benchmarks. Two independent grade-A comparisons of Jev with OpenAI's Decisions API differ by task: Decisions API ahead on RPG move choices (B38); Jev ahead on exact judge score in Portuguese, level on routing and injection (B39).
  • Updated: Jev alternatives. Jev weights: not found on TypeSafe's site, docs, GitHub or Hugging Face (10 Oct). Of 31 listed substitutes: 12 model weights, 11 local runtimes, 4 request adapters, 2 classifier/LLM approaches, 2 hosted APIs from other providers (OpenAI, Microsoft).
  • Updated: coding agents. Of 24 integrations read in source, 13 act on Jev's answer, 10 do not, 1 is unsettled. On a Jev error the 13: 5 let the call run, 3 hold, 5 fall back.
  • Updated: when Jev fails. Of 74 Jev implementations read in source, on a Jev error 38 stop or hold the action, 20 fall back, 8 let it through, 5 return no decision and 3 are advisory. The action goes ahead in 15 of 74.
  • Updated: Jev reranker and Jev for RAG. Of 4 rerunnable graded reranking-pipeline studies, 2 come from parties with a stake and 2 state none (jev-pulse, 40 queries; jev-rerank-bench, Jev level with Cohere within noise); the other grade-A study measures relevance judgments. On a Jev error, of 5 implementations, 2 stop, 2 fall back, 1 passes unscreened passages.
  • Access status rechecked 11 Oct 2026, 05:17 UTC: unchanged. Two notices shown as on 10 Oct; none retired.

10 October 2026 (no run took place on 9 October)

  • Updated: Jev alternatives. Jev weights: not found on TypeSafe's site, docs, GitHub or Hugging Face (10 Oct). Of 30 listed substitutes: 12 model weights, 11 local runtimes, 4 request adapters, 2 classifier/LLM approaches, 1 OpenAI hosted API.
  • Updated: coding agents. Of 22 integrations read in source, 10 act on Jev's answer, 10 do not and 2 are unsettled. On a Jev error the 10: 3 hold, 1 may ask the main model, 5 let the call run, 1 keeps its configured model.
  • Updated: when Jev fails. Of 71 Jev implementations read in source, on a Jev error 38 stop or hold the action, 17 fall back, 8 let it through, 5 return no decision and 3 are advisory. The action goes ahead in 15 of 71.
  • Updated: benchmarks. jev-pulse graded A; three comparisons with code and data at a pinned commit are listed as leads, not graded, because their metric code was not yet opened.
  • Access status rechecked 10 Oct 2026, 05:12 UTC: unchanged. Two notices shown; the subprocessor notice (2 Oct) and the rate-limit notice (3 Oct) retired when their seven-day shelf life ended. A missed run does not postpone a retirement.
  • Updated: Jev reranker and Jev for RAG. Of 3 rerunnable graded reranking-pipeline studies, 2 come from parties with a stake and 1 (jev-pulse: 40 queries, BM25-only baseline) states none; the other grade-A study measures relevance judgments. On a Jev error, of 5 implementations, 2 stop, 2 fall back, 1 passes unscreened passages.
  • TypeSafe published a funding announcement on 9 Oct 2026; it states no change to signup, credit, prices or limits, so it is not a home-page notice.

8 October 2026

  • New: Jev reranker and Jev for RAG. The grade-A study measures relevance judgments, not reranking; both rerunnable graded pipeline studies come from parties with a stake. On a Jev error, of 5 implementations, 2 stop, 2 fall back, 1 passes unscreened passages.
  • New: Jev inside SQL and document pipelines. On a Jev error, all 5 stop the query or request; pg-redact also stores text unredacted when a span's answer is missing or below 0.5. All 5 send row or document text out.
  • Updated: when Jev fails. Of 69 Jev implementations read in source, on a Jev error 38 stop or hold the action, 15 fall back, 8 let it through, 5 return no decision and 3 are advisory. The action goes ahead in 14 of 69.
  • Corrected: jev-approvals (Hermes Agent) hands the approval question to the main agent model on a timeout, connection error, 429 or 402; that model can approve the command. It was shown as holding for a person on every Jev error.
  • Access status rechecked 8 Oct 2026, 05:14 UTC: unchanged. Four notices shown as on 7 Oct; none retired.

7 October 2026

  • Updated: when Jev fails. Of 64 Jev implementations read in source, on a Jev error 35 stop or hold the action, 13 fall back to another model, rule or default, 8 are advisory or return no decision, and 8 let it through: 4 moderation bots, 2 agent hooks, a passage filter and an eval tool. Eval-judge tools are counted from 7 Oct.
  • Updated: coding agents. Of 18 Jev coding-agent integrations read in source, 1 is vendor-published and none comes from TypeSafe; 8 act on Jev's answer, and 4 of those let the tool call run when Jev errors. 2 of the 8 do nothing until switched on.
  • Updated and retitled: Jev benchmarks (TypeSafe AI): is Jev accurate? 33 graded rows (A 6, B 12, C 8, D 7). Against trained classifiers the evidence is now mixed (one Red Hat study, B28, grade B; the rest grade D); one study found open-weight models scored through their token probabilities about level with Jev overall (B35, grade B).
  • Updated: Jev as a judge (counted on when Jev fails), official vs reseller (thejevai.com row), data privacy (zero-data-retention contact now the sales mailbox).
  • Retired from the notice list: the TypeSafe WorkflowEvals notice (shelf life ended 6 Oct; hidden since 4 Oct). The release stays on benchmarks, B01.

6 October 2026

  • New: How to use Jev with Codex, OpenCode, Cursor and Hermes Agent. Of 18 Jev coding-agent integrations read in source, 1 is vendor-published (OpenClaw) and none comes from TypeSafe; 6 act on Jev's answer, and 4 of those let the tool call run when Jev errors.
  • New: Jev 1.13 and jev-latest: model IDs and aliases by route. On 4 of 6 routes read on 6 Oct 2026 you can write a versioned Jev ID; Vercel and Cloudflare take only an unversioned one. jev-latest points to jev-1.13.0 on TypeSafe's models page.
  • Updated: when Jev fails. Of 58 Jev implementations read in source, 32 stop or hold the action when Jev errors and 7 let it through unchecked: 4 content-moderation bots, 2 agent hooks and one Spring AI passage filter.
  • Updated: Where to use and data privacy. Vercel lists two providers for Jev (checked 6 Oct 2026, 05:15 UTC): TypeSafe AI (zero data retention) and DigitalOcean (no zero data retention); Vercel picks between them by uptime and latency unless you set the order.
  • Updated: Products, 92 rows (A 75, B 16, C 1), none independently verified in production.

5 October 2026

  • New: Jev browser use and computer use. Of 7 Jev browser and computer-use agents read in source, none clicks or types when Jev errors, but 5 act on low-confidence answers, and only 1 asks a person before send, buy or delete.
  • New: Jev LangChain and other framework integrations. Of 6 Jev framework packages read in source, 4 are pre-1.0 and none comes from TypeSafe; 4 raise Jev errors to your code, while the community Spring AI filter passes passages through unscreened. (7 packages since 31 Oct 2026, with TanStack AI; 5 of 7 raise errors to your code.)
  • New section: Is Jev on Amazon Bedrock or Azure AI Foundry? Jev is not listed in the Amazon Bedrock model catalog or the Azure AI Foundry model catalog (checked 5 Oct 2026, 05:18–05:19 UTC), and TypeSafe's docs do not mention either.
  • Updated: when Jev fails. Of 51 Jev implementations read in source, 30 stop or hold the action when Jev errors and 6 let it through unchecked: 4 content-moderation bots, one Claude Code hook and one Spring AI passage filter. A copyable checklist was added.
  • Refreshed: Products. 72 rows (A 55, B 16, C 1); Jev for Chrome, computer-use-jev and jev-android now grade A; 5 new grade-A rows; six framework packages listed as not counted. None independently verified in production.
  • Corrections from the 5 Oct report: langchain-typesafe 0.0.1a3 was uploaded on 20 Sep 2026, not 4 Oct; @effect/ai-typesafe is now 4.0.1; the Spring AI integration is a community package; LlamaIndex has a community integration; the AndroidWorld paper (arXiv 2609.30186) was submitted on 24 Sep, not 25 Sep (5 Oct report).

4 October 2026

  • New: Jev trading bot and fraud gates. Of 4 Jev trading bots read in source, 2 can still place a new order when Jev errors; all 4 can trade without Jev. None of 3 fraud or payment gates moves money.
  • New: when Jev fails. Of 39 Jev implementations read in source, 21 stop or hold the action when Jev errors and 5 let it through unchecked: 4 content-moderation bots and one Claude Code hook.
  • New: Jev as a judge. Of 4 Jev judge tools read in source, DeepEval stops the run on a Jev error; the other 3 drop the item from the score, and openlayer jevals still reports it as passed.
  • Refreshed: Products. 67 rows (A 47, B 19, C 1); jev-trader and Jev Trade now grade A; 6 new grade-A rows; DeepEval's JevEval and openlayer jevals listed as library features, not products. None independently verified in production.
  • New fact: TypeSafe's trust center lists 6 subprocessors; Nebius and CoreWeave were added on 2 Oct, where prompts are "processed, but not stored" (TypeSafe's wording). The SOC 2 report was not seen (data privacy). This is now a Home notice.
  • Corrections: the email page's unsure cells for Jevmail ("acts-anyway, flagged") and spam-leadgen ("fallback: not qualified") were relabelled, and its cost range now starts at about $0.025 per 1,000 very short emails (corrections). QuantDinger links now point to commit a5a9f4c; the 28 Sep line links had gone stale.
  • Version note: oh-my-claudecode v5.6.1 (3 Oct) is a patch; its failure path was checked at v5.6.0 (Claude Code and MCP). Access, price, limits and route listings unchanged (checked 4 Oct, 05:15 UTC).

3 October 2026

  • New: Jev email classifier (email and lead sorting). Of 5 email and lead sorters read in source, none touches the mailbox when Jev errors; of the 3 that can write to Gmail or IMAP, only 1 routes unsure mail to a Review label.
  • New: Jev n8n node. Only the scoped @typesafe-ai/n8n-nodes-typesafe-ai is TypeSafe's (npm provenance, n8n-verified); the unscoped n8n-nodes-typesafe-ai is a third party's. All 5 nodes read stop the workflow on a Jev error by default.
  • Refreshed: Products. 61 ledger rows; the pool read the Jev call in source for 39; none of the 61, nor 7 published claims, is independently verified as running in production. 5 claims rechecked; the 2 relayed by VentureBeat were not rechecked.
  • Version check: oh-my-claudecode v5.6.0 keeps its failure path; its default timeout is 2,000 ms from v5.6.0 (250 ms up to v5.5.0), and OMC_JEV=<point>:active now lets Jev decide (Claude Code and MCP).
  • Change: TypeSafe's models page now lists 80 requests per second (40 on 2 Oct, 05:14 UTC); tokens per second unchanged at 100K (errors and rate limits). Access otherwise unchanged (checked 3 Oct, 05:16 UTC).
  • Correction: Jevmail's behaviour on a Jev error is "no decision" (read-only Gmail access; the email stays unclassified and is retried on the next run), not "stops" (Use cases).

2 October 2026

  • New: Jev Router or your own model router? OpenRouter does not document what Jev Router does when Jev fails; of six self-hosted routers read in source, one sends every Jev failure to its frontier model.
  • New: content moderation with Jev. Of six projects read in source, four leave the post up or let it through when the Jev call fails; one holds it for a person; one returns no decision.
  • Version check: @jkudish/jev-mcp 0.12.0 and 0.13.0 leave the audited failure path unchanged (Claude Code and MCP). oh-my-claudecode v5.6.0 (1 Oct) is not yet checked. ai 7.0.127 noted, not re-audited (TypeScript).
  • Access unchanged (checked 2 Oct, 05:14 UTC); API 90-day uptime 99.826%. Route prices, limits and data labels unchanged.

1 October 2026

  • New: Jev vs Laya vs Kev on the same rows; what data leaves your system, with training, retention and ZDR quoted per route; and support-ticket triage, comparing five projects' failure policies.
  • Correction: Vercel's zdr "none" is a model-level value across two providers. Vercel's docs mark TypeSafe AI as ZDR; its DigitalOcean endpoint is marked not ZDR (details).
  • Correction: the Kev row (base licences, card dates), the Laya row (v0.3.22, base model, uncalibrated multilingual checkpoint) and the he-jev note on Alternatives.
  • The rate-limit change is now dated between 29 Sep, 19:24 UTC and 30 Sep, 20:47 UTC (timeline). Access is otherwise unchanged (checked 1 Oct, 05:16 UTC).
  • Version drift, not re-audited: ai 7.0.126 (TypeScript) and @jkudish/jev-mcp 0.12.0 (Claude Code and MCP).

30 September 2026

  • Both tools work again. The fit check and the cost worksheet did not load their scripts at the address where the site serves them (tools/<name>/), so their buttons and fields did nothing. Each now loads its script from a site-absolute address that works at both tools/<name>.html and tools/<name>/; every button and field was used in a headless browser at both addresses and inside its page before release.
  • Home now opens with the current access status (signups open, no free credit for new users, API operational; checked 30 Sep, 20:47 UTC) and up to four dated notices.
  • New: a graded ledger of published Jev benchmarks, including what TypeSafe's terms said about publishing them, and Jev vs an LLM vs a classifier.
  • Correction: TypeSafe's rate limits are now 100K tokens per second and 40 requests per second, not 250,000 tokens/s and 1,200 requests/min (errors page).
  • Correction: the confidence-thresholds page called the alexmolas.com post "no experiment"; it includes one small example.
  • New official TypeSafe surfaces: the n8n node repository and WorkflowEvals (official vs reseller). ai 7.0.124 and @ai-sdk/typesafe-ai 3.0.12 appeared after the TypeScript audit (drift note).

29 September 2026

  • Four new Build pages: a Build hub, TypeScript, Claude Code and MCP guardrails and testing without a key. Of five Jev guardrails read in source, only LangChain's AutoModeMiddleware stops the agent when Jev fails; the official TypeSafe plugin makes no Jev call.
  • Correction: oh-my-claudecode's Jev timeout was raised to 2000 ms only on its dev branch; the latest release, v5.5.0, still defaults to 250 ms (errors page).
  • Correction: the Products row "jev-mcp" is now "jev-mcp (jkudish)", grade A and advisory. At least four unrelated repositories share that name; five more audited integrations have their own rows (Products).
  • Access, prices, aliases and SDK versions were rechecked: no change since 28 Sep (29 Sep report).

28 September 2026

  • The site is now organised by reader question. Eight new pages cover access status, official vs reseller sites, cost per decision, Python, errors and rate limits, confidence thresholds, limitations and demo videos. Every earlier URL still works; see the old-to-new URL map and release notes.
  • Signups for Jev reopened on 27 Sep (22:33 UTC), but new users get no free credit for now (timeline).
  • Vercel's models API lists Jev at exactly $0.042 per million input tokens; OpenRouter also accepts TypeSafe's SDKs at /api/v1/systemone; OpenRouter's separate Jev Router routes to other models (route matrix).
  • Corrections: two project rows were wrongly labelled as official TypeSafe repositories, and a "250 ms default timeout" repeated online is not TypeSafe's (release 1 report).

Release notes: 9 November 2026

  • No new question page. New dated report: What changed for Jev builders, 8 to 9 November 2026. The 8 November report stays at its address.
  • Check dates: the previous checks were on 8 Nov 2026 (05:17–05:42 UTC) and there was no check between them and 9 Nov, 11:07 UTC. A fact carries a 9 Nov check time only where it was read again on 9 Nov; facts not read again keep their dates.
  • Benchmarks: rows B49 to B51 added to the ledger (grades B, A, B, each with its flags and pinned commit); ledger totals and the grade index recounted (48); two repositories listed as leads with reasons; two Reported blog lines added as leads; the Jev Decision Index and Artificial Analysis rows re-read; a dated "Is Jev deterministic?" section; TypeSafe's cookbook figures outside the ledger. The decision-API comparison page's ledger reference updated.
  • Coding agents: the auto-mode row re-pinned at 4.0.0 (7b246a7) with the bypass content change; every count unchanged (29 / 15 / 10 / 4); the failure map unchanged (80).
  • Data terms: the Drex row's disagreement sentence replaced (Nace.AI's texts now agree), with the earlier texts kept as a dated note; OpenCode's Terms quote and date updated, still not reconciled.
  • Frameworks and releases: @ai-sdk/typesafe-ai 3.0.35 and ai 7.0.165 and 7.0.166 compared with the previous versions (no cell changed); the frameworks and TypeScript version blocks rebuilt, with older checks moved to "Earlier version checks"; n8n 2.47.3 and @openclaw/typesafe 2026.11.2 noted, not re-read.
  • Corrections and re-dates: status-page uptime figures; OpenRouter's ZDR list count (1,109); the six routes, the lookalike sites and the docs index; one stale clause in the site summary that still described jev-latest as pointing to jev-1.13.0 is now dated to 6 Oct.
  • Not read (dates kept): the Amazon Bedrock model lists (partly rendered by script; not determined), the Foundry and Workers AI catalogs, Artificial Analysis's methodology, OpenCode's Privacy Policy and the mod host behind auto-mode.
  • Scan window (method): this update's scan covered 8 Nov 05:14 UTC to 9 Nov 11:05 UTC (29 h 51 min). Coverage of those hours is unknown on Reddit, X, YouTube and TikTok.
  • Home and access status: checked 9 Nov 2026, 11:07 UTC; access state unchanged. Two notices shown; none added or retired.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 9 Nov research report.

Release notes: 8 November 2026

  • No new question page. New dated report: What changed for Jev builders, 7 to 8 November 2026. The 7 November report stays at its address.
  • Check dates: the previous checks were on 7 Nov 2026 (15:29–15:55 UTC) and there was no check between them and 8 Nov, 05:17 UTC. A fact carries an 8 Nov check time only where it was read again on 8 Nov; facts not read again keep their earlier dates.
  • Failure map: ohmyjev's row re-pinned at 0.14.0 (fbc4ea4) with the same five terms and colours; the bypass cell gains the shadow setting and the in-repo move skip; every count unchanged (80). The Claude Code guardrails, coding agents, products and build pages carry the same pin; dated 2 Nov lines stay as history.
  • Coding agents: one row read in source and counted on the page (pi-jev-router, 2.0.0), shown under "Read in source; not on the failure map yet"; the page's totals and per-condition columns recounted; a dated block of 19 leads. auto-mode 4.0.0 noted as not re-read.
  • Data terms: SubQ Snap's retention line replaced (docs and Terms agree on 60 days), with the earlier texts kept as a dated note; Cloudflare's data-usage cells re-dated with the third exception quoted; OpenAI's data controls re-dated, unchanged. The comparison card and its copy block match.
  • Frameworks and releases: langchain-typesafe read at 0.0.2 (failure path unchanged); pydantic-ai-slim 2.66.0's TypeSafe files identical to 2.64.0; @ai-sdk/typesafe-ai 3.0.35, ai 7.0.165 and LiteLLM 1.108.3 noted.
  • Corrections and re-dates: use cases' cookbook count (6 patterns, 26 cookbooks, 2 demos); status-page uptime figures; OpenRouter's ZDR list count (1,117); OpenCode's free ID, the six routes, the lookalike sites and the Foundry and Workers AI catalogs re-read unchanged.
  • Not read (dates kept): the Amazon Bedrock model lists (not readable on 8 Nov), Eevo's announcement, ohmyjev's Codex adapter, and the contents of auto-mode 4.0.0, @ai-sdk/typesafe-ai 3.0.35 and ai 7.0.165.
  • Scan window (method): this update's scan covered 7 Nov 15:27 UTC to 8 Nov 05:14 UTC (13 h 47 min). Coverage of those hours is unknown on Reddit, X, YouTube and TikTok.
  • Home and access status: checked 8 Nov 2026, 05:17 UTC; access state unchanged. Two notices shown; none added or retired.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 8 Nov research report.

Release notes: 7 November 2026

  • No new question page. New dated report: What changed for Jev builders, 2 to 7 November 2026. The 2 November report stays at its address.
  • Check dates: the previous checks were on 2 Nov 2026 and there was no check between 2 Nov and 7 Nov. A fact carries a 7 Nov check time only where it was read again on 7 Nov; facts not read again keep their earlier dates. The access strip runs to 7 Nov because the signup and credit sources show no contrary statement in between; status events for those days come from the status page's own history (none listed).
  • TypeSafe docs: images listed as an input of both versions, Box on Jev 1.14 only, the language section removed, a new changelog page; each sentence that described these in the present tense now says what the docs state on 7 Nov. Older dated sentences stay as dated history. No image or Box request was sent.
  • Alternatives: one Base Labs row in the hosted-API layer; layer counts and the short answer recounted; Base Labs' published weights are a line inside the row. The benchmarks mention of Base Labs links to the row.
  • Versions and aliases: title and H1 now start "Jev 1.14 vs 1.13" (query source in the URL map below); a dated list of eight GitHub items, read on 7 Nov, on staying on 1.13 (Reported).
  • Corrections and re-dates: OpenCode's free ID re-read unchanged (its docs moved to a new address); OpenRouter's ZDR list count; status-page uptime figures; comparison cells due since 3 Nov re-read; new integration releases noted as not re-read (ohmyjev 0.14.0, langchain-typesafe 0.0.2 and patch releases); LiteLLM's stable releases still lack the eu/jev provider.
  • Not read (dates kept): three comparison cells that could not be settled (OpenAI's retention and ZDR wording, Clef's data usage, SubQ Snap's retention); the contents of the new integration releases and Laya v0.4.28; OpenRouter's terms for the Base Labs endpoints.
  • Scan window (method): this update's scan covered 2 Nov 10:46 UTC to 7 Nov 15:27 UTC (124 h 41 min), so the days without a check are inside it. Coverage of those hours is unknown on Reddit, X, YouTube and TikTok.
  • Home and access status: checked 7 Nov 2026, 15:29 UTC; access state unchanged. Two notices shown; none added or retired.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 7 Nov research report.

Release notes: 2 November 2026

  • No new question page. New dated report: What changed for Jev builders, 31 October to 2 November 2026.
  • Check dates: the previous checks were on 31 Oct 2026 and there was no check between 31 Oct and 2 Nov. A fact carries a 2 Nov check time only where it was read again on 2 Nov; facts not read again keep their earlier dates. The access strip runs to 2 Nov because the signup and credit sources show no contrary statement in between; status events for those days come from the status page's own history (none listed).
  • Failure map: ohmyjev row re-read at 0.13.2 and described in lockdown mode (its default is off); same terms, colours and counts; the checklist, copy block, self-check example and the Claude Code, coding-agent and products pages follow with the same term, colour and pin.
  • Third-party route: OpenCode's free Jev ID in its own dated block on where to use, with its data terms quoted on data privacy; not added to the six routes.
  • Corrections: OpenRouter's ZDR list (both TypeSafe endpoints); status-page uptime figures; Microsoft-Decision-1's catalog cells re-read (unchanged).
  • Not read (dates kept): the contents of Laya v0.4.21, @effect/ai-typesafe 4.1.2 and auto-mode 3.1.6; ohmyjev's handlers outside its gates; comparison cells not due before 14 Nov.
  • Scan window (method): this update's scan covered 31 Oct 05:09 UTC to 2 Nov 10:46 UTC (53 h 37 min), so the days without a check are inside it. Coverage of those hours is unknown on Reddit, X, YouTube, TikTok and GitHub code search.
  • Home and access status: checked 2 Nov 2026, 10:47 UTC; access state unchanged. Two notices shown; none added or retired.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 2 Nov research report.

Release notes: 31 October 2026

  • No new question page. New dated report: Hosted decision APIs compared: what changed by 31 October 2026.
  • Check dates: comparison cells dated before 17 Oct were read again. A cell carries a 31 Oct check time only where it was read again on 31 Oct; cells not read again keep their earlier dates, and a change whose source gives no date is dated as a range between two reads.
  • Corrections: Microsoft-Decision-1 rate limits; Cloudflare's data-retention and training wording and Clef's OpenRouter hosts; OpenAI's inputs and regions wording (the per-request image limit read on 13 Oct is kept as dated history); Eevo's prices. The comparison card's changed-since line and copy block follow.
  • Framework packages: one new row (TanStack AI); counts 7 read in source, 4 pre-1.0, 0 from TypeSafe, 5 raise errors to your code, 2 decide by themselves; old counts replaced in current-tense sentences. The failure map is unchanged (80).
  • Not re-read (dates kept): Microsoft-Decision-1's catalog cells and graded-test cells, Eevo's errors and authentication pages, the new Kev and Laya releases.
  • Scan window (method): this update's scan covered 30 Oct 05:13 UTC to 31 Oct 05:09 UTC (23 h 56 min). Coverage of those hours is unknown on Reddit, X, YouTube, TikTok and GitHub code search.
  • Home and access status: checked 31 Oct 2026, 05:11 UTC; access state unchanged. Two notices shown; one retired at the end of its shelf life, none added.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 31 Oct research report.

Release notes: 30 October 2026

  • No new question page. New dated report: What changed for Jev builders, 24–30 October 2026.
  • Check dates: the previous checks were on 24 Oct 2026 and there was no check between 24 and 30 Oct. A fact carries a 30 Oct check time only where it was read again on 30 Oct; facts not read again keep their earlier dates, and a change whose source gives no date is dated as a range between two reads. The access strip runs to 30 Oct because the signup and credit sources show no contrary statement dated in between; incidents for those days come from the status page's own history.
  • Re-dated: Home and access status, the key alias line and six-route lines on versions and aliases, the catalog lines on where to use, the unchanged comparison-card cells, the TypeSafe SDK version lines.
  • Corrections: OpenAI's Decisions API retention row (no longer "Pending confirmation"), Microsoft-Decision-1 regions and SubQ Snap 1.1's price, each with its date range; the comparison card's changed-since list and copy block follow. Base Labs' free-week line on benchmarks. TypeSafe's higher-limit form on errors and limits and limitations.
  • Coding agents: one hook read in source, off the failure map; counts 28 read in source (14 + 10 + 4); old counts replaced in current-tense sentences; lead triage block (ten names, README level).
  • Not re-read (dates kept): OpenAI's "128 images" cell, Microsoft-Decision-1's retention wording, the d1-ultra line and the Drex row apart from its prices, ohmyjev's failure-map cells at 0.13.2.
  • Scan window (method): this update's scan covered 24 Oct 05:11 UTC to 30 Oct 05:13 UTC (144 hours, by day). Coverage of those dates is unknown on Reddit, X, YouTube, TikTok and GitHub code search, and for 6 GitHub names on 28 Oct beyond the search limit.
  • Home and access status: checked 30 Oct 2026, 05:14 UTC; access state unchanged. Three notices shown; none added or retired.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 30 Oct research report.

Release notes: 24 October 2026

  • No new question page. New section Jev benchmark leaderboards on benchmarks, and a new dated report: jev-latest now gives Jev 1.14; decision leaderboards and four new graded studies (24 Oct 2026).
  • Aliases: jev-latest moved to jev-1.14.0 on TypeSafe's API between the 23 Oct, 05:45 UTC and 24 Oct, 05:12 UTC checks. Current-tense sentences that named 1.13 as the alias target or described the move as upcoming were replaced or extended; dated observations stay. Versions and aliases: title and H1 without "preview", new key finding and a six-route table.
  • Benchmarks: leaderboard section (key finding, a table of what each leaderboard lets you check, the Base Labs claim at two versions, nine claims checked, method notes); B45–B48 graded; grade index 45; the key finding reopened as a dated 24 Oct sentence with the 12 Oct form kept below; coverage lines on images, model versions and long state updated; nine new leads listed, not opened.
  • Cross-page: grade lines on alternatives (Decisions API and SubQ Snap rows), the decision-API comparison (graded tests row and copy block) and reranking and RAG (study types stated); the old ledger total was replaced in current-tense sentences.
  • Correction: the Decisions API price cell gains OpenAI's cached-input and cache-write prices since general availability; the 13 Oct beta wording is kept as dated.
  • Scan window (method): this update's scan covered 23 Oct 05:18 UTC to 24 Oct 05:11 UTC.
  • Home and access status: checked 24 Oct 2026, 05:12 UTC; access state unchanged. Three notices shown (one added: the jev-latest move); none retired.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 24 Oct research report.

Release notes: 23 October 2026

  • No new question page. New section Jev with images on limitations, and a new dated report: Jev with images: what 8 SDKs and integrations do with an image, and the route counts for Jev 1.14 (23 Oct 2026).
  • Limitations: key finding, a figure and a table for 8 integrations (plain terms; not part of the failure map), TypeSafe's image docs with quotes and times, the route-listing line, image leads by date, and what was not verified. The "Is Jev multimodal?" box links to it.
  • Cross-page: n8n (0.10.0 checked; Images field; data sent; current version), TypeScript, Python, frameworks (pydantic-ai-slim 2.60.0; LangChain and LiteLLM image lines), errors and limits (image requests row). No failure-map cell or count changed.
  • Aliases: 11 sentences that described the jev-latest move as upcoming now give its state on 23 Oct 2026, 05:45 UTC; dated lines and quotes unchanged.
  • Routes: "list a 1.14 ID" 6 of 6; "a versioned ID can be written" 6 of 6 for 1.14 and 5 of 6 for 1.13; "you cannot pin" on Cloudflare now refers to 1.13.
  • Alternatives and the comparison page: OpenAI's Decisions API status changed to generally available (22 Oct); price unchanged; no new graded cell.
  • Correction: jev-gateway pin links point to vinilana/jev-gateway at the same commit; cells unchanged.
  • Scan window (method): this update's scan covered 21 Oct 05:10 UTC to 23 Oct 05:18 UTC.
  • Home and access status: checked 23 Oct 2026, 05:18 UTC; access state unchanged. Two notices shown; none retired. jev-latest still points to jev-1.13.0.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 23 Oct research report.

Release notes: 21 October 2026

  • No new question page. New dated report: Coding agents on pi and omp, auto-mode and jcode, and Nace.AI's two sets of terms (21 Oct 2026).
  • Coding agents: new counted row for ohmyjev's pi and omp adapters (fail-open a–d, acts-anyway e, bypass recorded; the tool call still runs). Read in source 24 → 27; counted 13 → 14 (fail-open 1 → 2); not settled 1 → 3 (auto-mode, jcode). Hosts 17 for 14 rows (pi 1, omp 1). Lead triage: 13 lines from the 19–20 Oct scans. Kicker, key finding, short answer, figure and install-and-check list recounted from the table.
  • When Jev fails: ohmyjev's row names pi and omp (text added to the row only; every count, and every string the self-check uses, unchanged; 80 implementations). Bypass-cell file reference corrected.
  • Cross-page: the build hub's coding-agents row, the products ledger's ohmyjev row and the Claude Code and MCP guards ohmyjev row carry the same numbers and terms.
  • Alternatives: the Drex row quotes Nace.AI's website terms next to its Console terms, its zero-retention mode and its sub-processor schedule; drex-v1.5 price units side by side. 38 rows; layer counts unchanged.
  • Versions and aliases: alias line re-dated (21 Oct, 05:11 UTC; unchanged); Cloudflare's typesafe/jev-1.14.0 catalog page added.
  • Drift: TypeSafe's n8n node 0.10.0 (image support per its release note; not read); @ai-sdk/typesafe-ai 3.0.24, ai 7.0.148, @jkudish/jev-mcp 0.14.4, @delt/claude-jev-advisor 0.5.1 and litellm 1.106.1 (patches; not re-read).
  • Scan window (method): this update's scan covered 20 Oct 05:13 UTC to 21 Oct 05:10 UTC.
  • Home and access status: checked 21 Oct 2026, 05:11 UTC; access state unchanged. Two notices shown; the Jev 1.14 preview and experimental-features notices retired after seven days. jev-latest still points to jev-1.13.0.
  • Tools: the three tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 21 Oct research report.

Release notes: 20 October 2026

  • No new question page. New dated report: Failure self-check, checklist question 9, and Drex on alternatives (20 Oct 2026).
  • When Jev fails: new self-check (five questions, one per failure condition; "N of 80 made the same choice" with the map term, examples and checklist question; "not sure" points to the checklist; "Clear choices"). Its data is generated from the map and every count equals the map's (error 40 / 22 / 10 / 5 / 3; malformed 36 / 24 / 13 / 4 / 3; below threshold 17 / 16 / 12 / 5 / 5 / 25; no key 51 / 17 / 9 / 0 / 3; bypass 13 none found, 67 with skip paths). Checklist question 9 (no count); heading "Nine questions"; copy block regenerated.
  • Corrections in this release: 23 current-tense sentences on 13 pages, and one on the home page, gave the earlier map count of 79 (and in several cases its category numbers); all now give 80. Dated lines keep their counts.
  • Alternatives: Drex (Nace.AI) added; 38 rows; hosted APIs from other providers 7 → 8. Price as published in its docs; the product page's drex-v1.5 figure is in a different unit and not reconciled. No comparison-card column.
  • Drift: LiteLLM 1.106.0 (failure path unchanged); ai 7.0.146, @ai-sdk/typesafe-ai 3.0.23, @tanstack/ai-typesafe 0.1.13 and n8n 2.44.4 (patches; not re-read).
  • Scan window (method): this update's scan covered 19 Oct 05:21 UTC to 20 Oct 05:13 UTC.
  • Home and access status: checked 20 Oct 2026, 05:13 UTC; access state unchanged. Four notices shown; none retired. jev-latest still points to jev-1.13.0.
  • Tools: the new self-check and the two existing tools were each opened through their page, at tools/<name>/ and at tools/<name>.html, with every button and field used.
  • Evidence: 20 Oct research report.

Release notes: 19 October 2026

  • No new question page. New dated report: Failure map completed with ohmyjev, jev-mcp commits, and three hosted decision models (19 Oct 2026).
  • When Jev fails: ohmyjev added in Claude Code / MCP guards (8 → 9); 80 implementations in 12 domains, every failure cell recorded for all 80. No key 51 fail-closed, 17 fallback, 9 fail-open, 3 advisory; error 40 / 22 / 10 / 5 / 3 (fail-closed / fallback / fail-open / no-decision / advisory); malformed 36 / 24 / 13 / 4 / 3; below threshold 25 no-threshold, 17 held, 16 acts-anyway, 12 fallback, 5 no-decision, 5 advisory; bypass 80 recorded, 0 not recorded (13 "none found", 67 with skip lists). The 10 that let the action through on a Jev error: 4 moderation bots, 4 agent hooks, a passage filter and an eval tool. veermshah's error note now says its 10 s / 30 s values are SDK defaults, not read in the repository.
  • Claude Code and MCP: ohmyjev's pi and omp adapters read in full at the pinned commit; its bypass cell is recorded and it is counted on the map (its built Codex hook was checked at its gate lines; its source was read in full). Short answer unchanged (ten guardrails). jkudish/jev-mcp: one convention on four pages, cells cite 53fe575 (0.11.0), later checks are listed as "checked at … failure path unchanged".
  • Alternatives: three rows added (Liquid AI d1-ultra, Mouse-1, Eevo); 37 rows; hosted APIs from other providers 4 → 7. Data terms are quoted, not summarised as "private" or "safe". No comparison-card column.
  • Domain pages: content moderation 11 cells and support-ticket triage 9 cells reworded to the map terms; 0 outcome or colour changes.
  • Products: Vercel's own production-index figures added beside the relayed Vercel claim (Reported). The two VentureBeat-relayed claims were still not rechecked (HTTP 429; no archived copy found); their line keeps its 18 Oct date.
  • Drift: litellm 1.106.0 (minor; failure path not rechecked at this version), @openclaw/typesafe 2026.10.5 and oh-my-claudecode v5.7.4 and v5.7.5 (patches; not re-read). n8n's missing-credential path is unchanged at n8n@2.44.3.
  • Scan window (method): each scan starts at the previous scan's cutoff, so windows neither gap nor overlap. This update's scan covered 18 Oct 05:16 UTC to 19 Oct 05:21 UTC.
  • Home and access status: checked 19 Oct 2026, 05:22 UTC; access state unchanged. Four notices shown; none retired. jev-latest still points to jev-1.13.0.
  • Tools: the fit check and the cost worksheet were each opened at 3 addresses, at desktop and phone widths; 482 of 482 checks passed. Final-shell browser check of all 72 routes (overflow, iframe scroll, keyboard, focus, 200% text): 0 issues.
  • Evidence: 19 Oct research report.

Release notes: 18 October 2026

  • No new question page. New dated report: Failure terms aligned, three Claude Code guardrails read, and the products ledger rechecked (18 Oct 2026).
  • When Jev fails: claude-model-guard and claude-jev-advisor added in Claude Code / MCP guards (6 → 8); 79 implementations in 12 domains, every failure cell recorded for all 79. No key 51 fail-closed, 17 fallback, 8 fail-open, 3 advisory; error 40 / 22 / 9 / 5 / 3 (fail-closed / fallback / fail-open / no-decision / advisory); malformed 36 / 24 / 12 / 4 / 3; below threshold 25 no-threshold, 17 held, 15 acts-anyway, 12 fallback, 5 no-decision, 5 advisory; bypass 79 recorded, 0 not recorded (13 "none found", 66 with skip lists). The 9 that let the action through on a Jev error: 4 moderation bots, 3 agent hooks, a passage filter and an eval tool.
  • Failure terms across pages: 15 pages compared cell by cell with the map; 0 outcome differences. Coding agents: key-finding split corrected (3 fail-closed, 9 fallback, 1 fail-open on a Jev error) and its cells reworded to the map terms (bypass cells say "recorded"; hermes-jev-mana's four cells keep the term and move "configured model" to the detail). Trading and fraud gates: 16 cells and the legend line. Browser and computer use: 6 cells and a second-look line. Wording-only cells on content moderation (11) and support-ticket triage (9) are left for a later update. The jkudish/jev-mcp pin difference between the Claude Code page (86eae78) and the map (53fe575) is unchanged.
  • Claude Code and MCP: three guardrails read at pinned commits; new short answer over ten guardrails (replaces "1 of 7"); ohmyjev shown with all its cells, not yet in the map count (editorial decision, 18 Oct: its pi and omp adapters were read at their tool-call gates only). Title and URL unchanged.
  • Products: 14 rows added (3 guardrails, 3 demos, 8 projects); ledger 113 (A 100, B 12, C 1); independently verified in production 0 of 113. Claims rechecked 18 Oct 2026, 05:19 UTC: 5 of 7 same text, 2 of 7 not rechecked (VentureBeat HTTP 429), 0 changed. Directory counts re-dated; they are listings, not adoption.
  • Routes and SDKs: Opper lists typesafe/jev-1.14.0 (alias typesafe/jev-preview); its typesafe/jev-latest still 1.13.0. Cloudflare, Bedrock, Microsoft Learn's partner-models list and Workers AI do not list 1.14 (05:20 UTC; the Foundry catalog itself was last read on 16 Oct). Vercel 32,000 line. @typesafe-ai/sdk 0.7.0: failure path unchanged from 0.6.0.
  • Alternatives: one dated line: Liquid AI d1-ultra and Mouse-1 launched, not yet evaluated; no rows and no comparison columns; layer counts unchanged.
  • Drift: patches ai 7.0.143, @ai-sdk/typesafe-ai 3.0.21, deepeval 4.3.1 (JevEval files not rechecked), n8n 2.44.3 (Jev path last read at 2.44.1); pre-release @openclaw/typesafe 2026.10.10-beta.2.
  • Scan window (method): each scan starts at the previous scan's cutoff, so windows neither gap nor overlap. This update's scan covered 16 Oct 05:19 UTC to 18 Oct 05:16 UTC.
  • Home and access status: checked 18 Oct 2026, 05:17 UTC; access state unchanged. Four notices shown; none retired. jev-latest still points to jev-1.13.0.
  • Tools: the fit check and the cost worksheet were each opened at 3 addresses, at desktop and phone widths; 386 of 386 checks passed. Final-shell browser check of all 71 routes (overflow, iframe scroll, keyboard, focus, 200% text): 0 issues.
  • Evidence: 18 Oct research report.

Release notes: 16 October 2026

  • No new question page. New dated report: Bypass cells complete, Jev 1.14 preview and Claude Code guardrails rechecked (16 Oct 2026).
  • When Jev fails: the last 11 bypass cells and the last malformed cell settled at the published pins. Bypass recorded for 77 of 77, 0 not recorded (66 / 11 on 14 Oct): judge tools 4, frameworks 3, browser and computer use 3, email 1 newly recorded; 13 cells are "none found" in the files read and 64 list skip paths. Malformed answer: 35 fail-closed, 24 fallback, 11 fail-open, 4 no-decision, 3 advisory, 0 not recorded (competitor-hunter now no-decision). No key, error and below threshold unchanged, each with 0 not recorded. The domain pages carry the same cells.
  • Read scope for four bypass cells (editorial decision, 16 Oct): openlayer jevals, competitor-hunter, the ilyamk Gmail labeler and LlamaIndex JevRerank count as recorded; each cell names the files or lines read in a narrower way (a documentation helper, page markup checked by script, entry functions found by a scripted scan, provider clients read at branch, return and raise lines).
  • Jev 1.14 preview: jev-1.14.0 under jev-preview since 14 Oct; jev-latest still jev-1.13.0 (TypeSafe: it moves "in a week"). Image input, the Box question type and 262k long context are labelled experimental and not production-ready per TypeSafe. Routes listing 1.14 at 16 Oct 2026, 05:21–05:25 UTC: TypeSafe, OpenRouter, Vercel AI Gateway, Netlify; not listed: Cloudflare, Opper, Amazon Bedrock, Microsoft Foundry, Workers AI. SDKs: typesafe-sdk 0.7.5 and @typesafe-ai/sdk 0.7.0 (error path not yet re-read); @ai-sdk/typesafe-ai 3.0.20.
  • Claude Code and MCP: new title (URL unchanged), one version block with an "Earlier version checks" list, and a 32-lead table that replaces the "Leads seen, not yet read" box. Short answer unchanged. Three gates read at pinned commits join the page in a later update.
  • Benchmarks: Decision Index line, Reported, outside the ledger; no grade or count change (41 graded: A 12, B 14, C 8, D 7).
  • Comparison and alternatives: the Jev "Images" and "Documented errors and limits" card cells and the copy block; prices and other columns unchanged. Eevo (Upright) seen, not checked.
  • Drift: pydantic-ai-slim 2.56.0 and 2.57.0 (a request with images raises UserError before any call; failure path otherwise unchanged); DeepEval 4.3.0 (JevEval files unchanged); patches @ai-sdk/typesafe-ai 3.0.20, ai 7.0.142, @effect/ai-typesafe 4.0.5, @openclaw/typesafe 2026.10.4, @jkudish/jev-browser 0.8.8; n8n 2.44.2 published (Jev path last read at 2.44.1).
  • Scan window (method): each scan starts at the previous scan's cutoff and ends at the current check, so windows neither gap nor overlap. This update's scan covered 14 Oct 16:45 UTC to 16 Oct 05:19 UTC.
  • Home and access status: checked 16 Oct 2026, 05:21 UTC; access state unchanged. Four notices shown (two new: the Jev 1.14 preview and the experimental features). The Series A notice retired at the end of its seven-day shelf life (checked 05:24 UTC); it was hidden and never shown on Home. Rate limits are now set per organization, kept as an access condition. The console shows a sign-in page; a 45-minute us-east-1 latency event was resolved on 14 Oct.
  • Tools: the fit check and the cost worksheet were each opened at 3 addresses, at desktop and phone widths; 810 of 810 checks passed.
  • Evidence: 16 Oct research report.

Release notes: 14 October 2026

  • No new question page. New dated report: Jev (TypeSafe AI) research report, 14 October 2026.
  • When Jev fails: 31 open cells settled at the published pins (ticket triage, email and lead sorting, n8n, browser and computer use, jev-trader, hermes-jev-mana, jev-mcp). No key: 50 fail-closed, 17 fallback, 7 fail-open, 3 advisory, 0 not recorded (43 / 16 / 7 / 3 / 8 on 13 Oct). Bypass recorded for 66 of 77, 11 not recorded (52 / 25): judge tools 4, frameworks 3, browser and computer use 3, email 1. Below threshold: 25 no-threshold, 16 held, 15 acts-anyway, 12 fallback, 5 no-decision, 4 advisory, 0 not recorded. Error and malformed counts unchanged. Two cells corrected without a count change (zammad-jev-dispatcher malformed answer; jev-for-chrome below threshold). The domain pages carry the same cells.
  • Rule applied strictly: bypass counting (editorial decision, 14 Oct). Under the rule of 13 Oct, a bypass cell counts as recorded only when every entry point was read: every skip path listed with file and line, or "none found" with the list of entry-point files read from one tree listing at the pin. Four cells that were only partly read (the ilyamk Gmail labeler, jev-android, jev-for-chrome and second-look) stay not recorded; for the Gmail labeler and second-look the skip paths found are still shown, marked "partial read; not counted as recorded".
  • Scan window (method): each scan starts at the previous scan's cutoff and ends at the current check, so windows neither gap nor overlap. This update's scan covered 13 Oct 2026, 05:17 UTC to 14 Oct 2026, 16:45 UTC.
  • SDKs: Python checked at typesafe-sdk 0.7.4 and TypeScript at @typesafe-ai/sdk 0.6.0, ai 7.0.140 and @ai-sdk/typesafe-ai 3.0.18, each with one version block and an "Earlier version checks" list. New titles (URLs unchanged): "Jev Python SDK (TypeSafe AI): a version-pinned call, timeouts, retries and missing-key errors" and "Jev TypeScript (TypeSafe AI): @typesafe-ai/sdk vs AI SDK vs fetch, with timeouts and missing-key errors". Python examples pin 0.7.4; the AI SDK example keeps its audited pins because the full option shape of experimental_decide was not read. The same versions and names are carried on errors and rate limits, testing without a key, Where to use and official vs reseller.
  • Benchmarks: B43 graded B (6 flags, among them the operator's part in Subquadratic's prize challenge) and B44 graded A (5 flags). 41 graded rows (A 12, B 14, C 8, D 7). The sentence on small text-generating LLMs (6 of 10 graded studies) is unchanged. The Artificial Analysis Decision Index v1.0 is a watch item, not graded.
  • Alternatives and the comparison page: SubQ Snap added as a hosted row (34 substitutes). B44 added to the comparison card's graded tests and copy block; no price or status cell changed.
  • Drift: patches @jkudish/jev-mcp 0.14.3, @effect/ai-typesafe 4.0.4, @openclaw/typesafe 2026.10.3, litellm 1.105.1; minor pydantic-ai-slim 2.56.0 (drift notes, not re-audited).
  • Home and access status: status unchanged, checked 14 Oct 2026, 16:46 UTC. Two notices shown, one hidden. The Jack & Jill case-study notice retired at the end of its seven-day shelf life (checked 16:47 UTC); it was never shown on Home.
  • Tools: the fit check and the cost worksheet were each opened at 5 addresses, at desktop and phone widths; 1,030 of 1,030 checks passed.
  • Evidence: 14 Oct research report.

Release notes: 13 October 2026

  • New page: Decisions API vs Jev (TypeSafe AI) vs Microsoft Decision 1, with Cloudflare Clef. A six-row card (graded tests, price as published, images, request shape, your data, documented errors and limits) for Jev, OpenAI's Decisions API, Microsoft-Decision-1 and the Cloudflare Clef family; a 13-row table in which every cell carries its quote, date and source; a copy block that repeats the card word for word; a not-tested list; vendor claims, Reported and attributed. Key finding: "Graded results depend on the task: OpenAI's Decisions API ahead on RPG move accuracy (B38); Jev on Portuguese exact judge score (B39); Jev and Microsoft-Decision-1 not separated on routing macro-F1, both ahead of OpenAI (B41)." SubQ Snap (Subquadratic, launched 12 Oct 2026 per the vendor post) is named on the page as also launched and not compared; its docs and data terms were not read.
  • New rule: side-by-side comparisons of hosted decision APIs (decided 13 Oct). On the comparison page and anywhere its rows are quoted, graded rows lead, and each names its task, its metric and who came out ahead on that metric, as the operator reported it. No sentence, card cell or copy line says or implies that one API is better, more accurate, cheaper per decision or the best overall, and no row is summed, averaged or ranked. Prices are quoted as published, per million tokens, with the page and date; there is no per-call arithmetic. Every non-Jev column names its provider and says where your data goes. Each cell carries its own source date, and a fact not found is written with its evidence class and check time. Vendor accuracy, speed and price claims are Reported and attributed, outside the graded rows. Reason: the page puts three providers' hosted APIs next to Jev, and the graded studies point in different directions by task.
  • New rule: a bypass cell of "none found" (decided 13 Oct). A bypass cell on when Jev fails counts as recorded when every code path that reaches the action without calling Jev is listed with its file and line at the pin, or when it says "none found" together with the list of entry-point files read, taken from one tree listing at the pin. "None found" without that list stays not recorded. Reason: an absence is only as good as the files that were read; this matches how JevPlaysSnakeGame's "none found" was counted on 12 Oct.
  • New standing runner step: comparator review (decided 13 Oct). Each run, a script extracts every decision model, hosted decision API or open-weight substitute that a benchmarks ledger row names as a comparator, and checks it against the alternatives rows. Each ends as "has a row", "covered by family row" or "no row", with the reason. Rows are added only on the manager's decision. Reason: Cloudflare's Clef had been a benchmarks comparator since 7 Oct but had no alternatives row until 12 Oct. First result: of 9 comparator entries, 5 have a row, 2 are covered by a family row (Clef; Winnow through the Ollaya row) and 2 have none (Cygnet, B07; GLiNER2.5-Decide, B29). Those 2 are listed as seen, not checked, and queued for run 16.
  • Benchmarks: B42 (etsabary/luna-decisions-benchmark at 7a5ea13, 1,000 five-option logic questions) graded B with seven flags, among them Jev answers reused from a 22 Sep run, a withheld answer key and calls through OpenRouter's alpha Decisions endpoint. "On 1,000 five-option logic questions (B42, grade B), Jev answered 817 correctly and OpenAI's Decisions API 612; Jev's answers come from a 22 Sep run, and the answer key is not published." The ledger has 39 graded rows (A 11, B 13, C 8, D 7); graded comparisons with the Decisions API: 4 (B38, B39, B41, B42), each stated per task. B42 sits below B38, B39 and B41 and is not in the comparison page's key finding. The sentence on small text-generating LLMs (6 of 10 graded studies) is unchanged. utakatano/decision-models-demo stays a lead, not graded (an empty repository at 13 Oct 2026, 05:22 UTC).
  • When Jev fails: all 18 open no-key and bypass cells for model routing and content moderation settled at the published pins. No key: 43 fail-closed, 16 fallback, 7 fail-open, 3 advisory, 8 not recorded (38 / 14 / 4 / 3 / 18 on 12 Oct). Bypass recorded for 52 of 77, 25 not recorded (44 / 33). The LiteLLM bypass cell counts as recorded: the complexity router's own bypass paths were read at the pin. agent-router's malformed-answer cell is now fail-closed (it exits with code 2), so the malformed split is 35 fail-closed, 24 fallback, 11 fail-open, 3 no-decision, 3 advisory, 1 not recorded. Four other cells gain text qualifiers with no change to the counts. The ticket-triage bypass cells move to run 16.
  • Alternatives: the three hosted rows rechecked (Decision-1 requests-per-minute limits by tier, Product Terms and catalog licence text, Foundry lifecycle Preview; OpenAI refusal behaviour, 128-image limit and region table; Clef platform errors and OpenRouter hosts; Clef base-model licences) and linked to the new page. SubQ Snap, Cygnet and GLiNER2.5-Decide are added to "seen, not checked". 33 substitutes, unchanged.
  • Corrections: Where to use said Microsoft-Decision-1 was GA in Foundry; it is listed as Preview at 13 Oct 2026, 05:20 UTC (listed GA on 12 Oct; cause unknown). The Build hub said 71 implementations; it now says 77. The 12 Oct report and release notes keep their dated wording.
  • Drift: patches only: @ai-sdk/typesafe-ai 3.0.18, ai 7.0.138, @jkudish/jev-mcp 0.14.2, @jkudish/jev-agent-tools 0.2.1 (drift notes, not re-audited).
  • Home and access status: rendered from the run's status file: status unchanged, checked 13 Oct 2026, 05:18 UTC. Two notices shown, two hidden. The Vercel provider notice retired at the end of its shelf life (13 Oct, 05:15 UTC; checked 05:19 UTC); it was never shown on Home. Neither the Decision-1 lifecycle change nor the SubQ Snap launch is a home-page notice.
  • Tools: the fit check and the cost worksheet were each opened at 5 addresses, at desktop and phone widths; 944 of 944 checks passed.
  • Evidence: 13 Oct research report.

Release notes: 12 October 2026

  • New page; release 4 complete (4 of 4): Jev plays Doom: what real-time demos do when a Jev call fails. Five harnesses read at pinned commits: 3 counted (doom-jev, jev-robotics-demo, JevPlaysSnakeGame), 2 listed and not counted (ra2web-jev-player calls a local substitute model by default; llm-invaders is a research harness with no model player as the default). Each counted row shows the loop tick next to the request timeout.
  • New rule: counting real-time harnesses (decided 12 Oct). A game or simulation harness counts on when Jev fails when its own code performs the action Jev chose and it calls TypeSafe's Jev by default at the pin. A harness that calls a substitute model by default is listed, not counted. An uncaught exception that stops the demo is fail-closed, with "the demo stops; no action is taken". Below threshold follows the decision that gates the physical action. A new domain is opened only at 3 or more counted rows. "Keep the last action" (fallback-last action) counts as the action going ahead, because the loop keeps acting on a choice Jev did not make for this tick. Reason: a game loop does not wait for Jev, so a demo that silently keeps moving on an old choice is a different risk from one that stops; and a harness that runs another model by default says nothing about what a Jev failure does.
  • When Jev fails: new domain "Real-time control (games and simulations)" with 3 rows; n = 77 in 12 domain tables. On a Jev error: fail-closed 39, fallback 22, fail-open 8, no-decision 5, advisory 3. The action goes ahead in 17 of 77 (15 of 74 on 11 Oct; doom-jev and JevPlaysSnakeGame keep their last action). Routing fallbacks: 8.
  • New standing check: 72-hour decision-API look-back (decided 12 Oct). Each scan also looks back 72 hours for launches of hosted decision models or decision APIs by any provider, on timestamped surfaces and known provider pages. It is not a full 72-hour scan. Reason: Microsoft-Decision-1 was published on 9 Oct, when no run took place, and was first recorded on 11 Oct. See the scan section.
  • Benchmarks: B41 (kartikanand73/decision-api-benchmark, synthetic bank-withdrawal routing) graded A with four flags (one synthetic task; thresholds set with Jev-shaped probabilities, the operator's own note; all three models called through OpenRouter; code version not pinned). Jev and Microsoft-Decision-1 were not separated on macro-F1 and both were ahead of OpenAI's Decisions API on this task. The ledger has 38 graded rows (A 11, B 12, C 8, D 7); graded comparisons with the Decisions API: 3 (B38, B39, B41), each stated per task. The aggregate sentence on benchmarks now reads: "Against small text-generating LLMs, Jev scored ahead in 6 of 10 graded studies, mixed in 2 and not ahead in 2; 5 of the 10 are grade C."
  • Alternatives: Jev is not listed in Microsoft Foundry's model catalog (checked 12 Oct 2026, 05:15 UTC); Foundry lists Microsoft-Decision-1. Its data cell is partly read (training text quoted; no retention period stated; Product Terms not read). The Cloudflare Clef family is added as two rows, a hosted API from another provider and open weights (33 listed substitutes). OpenAI's Decisions API row gains its data-control values.
  • Drift: LiteLLM 1.105.0 checked: the Jev classifier's failure path is unchanged. @openclaw/typesafe 2026.10.2 released (drift note only). The ra2web-jev-player pin moved from af8caca to 0e4e8cf; the cited files are byte-identical.
  • Home and access status: rendered from the run's status file: status unchanged, checked 12 Oct 2026, 05:12 UTC. Two notices shown; three current notices stay off the home page. Another provider's decision model is never a home-page notice.
  • Evidence: 12 Oct research report.

Release notes: 11 October 2026

  • New rule: grades rest on opened files (decided 11 Oct). A comparison of Jev with a substitute is graded only from its metric file and its results file, opened at a pinned commit; a README number alone keeps it a lead, not graded, with the missing file named. A lead is never shown with a grade or a grade colour, and no page sums, averages or ranks the comparisons into an overall verdict: a summary names each task and who came out ahead in it, as the operator reported. Reason: on 10 Oct three published comparisons could only be listed as leads because their files had not been opened, while the alternatives page needed graded evidence to compare Jev with another provider's hosted decision API; their READMEs pointed in different directions by task.
  • Benchmarks: Two independent grade-A comparisons of Jev with OpenAI's Decisions API differ by task: Decisions API ahead on RPG move choices (B38); Jev ahead on exact judge score in Portuguese, level on routing and injection (B39). B38 (komzweb, RPG move choices), B39 (mateusfonsek, Portuguese routing, injection detection and judging) and B40 (anessbelbati, a reranking-pipeline study: Jev and Cohere Rerank 4 Pro within noise) are graded A with their flags; the ledger has 37 graded rows (A 10, B 12, C 8, D 7; 34 on 10 Oct). In B39 the metric matters: Jev is ahead on the exact judge score (80.7% vs 72.0%), the Decisions API is ahead within ±1 point (100.0% vs 98.7%) and on injection precision, and the operator calls the routing and injection differences not significant. classifier-bakeoff stays a lead, not graded (report opened, metric code not opened); decisionbench was not reached; 10 new comparison repositories are listed as leads.
  • Alternatives: Jev weights: not found on TypeSafe's site, docs, GitHub or Hugging Face (10 Oct). Of 31 listed substitutes: 12 model weights, 11 local runtimes, 4 request adapters, 2 classifier/LLM approaches, 2 hosted APIs from other providers (OpenAI, Microsoft). The Decisions API's "Parity or quality evidence" cell is rebuilt from the graded rows B38 and B39 only. Microsoft-Decision-1 (published 9 Oct; Microsoft says it is "available in Microsoft Foundry and through OpenRouter") is a second hosted API from another provider; it was not in OpenRouter's model list at 11 Oct 2026, 05:17 UTC, so that availability rests on Microsoft's statement. No graded comparison exists for it yet.
  • Coding agents: Of 24 integrations read in source, 13 act on Jev's answer, 10 do not, 1 is unsettled. On a Jev error the 13: 5 let the call run, 3 hold, 5 fall back. jev-triage is counted because OpenClaw's host code was read at a pinned commit (a plugin result marked handled returns before the main model is called); every jev-triage failure sends the message to the main model. The jev-for-all launcher is counted; its OpenCode plugin part is not settled. hermes-jev-mana is counted, with the note that it is inert after the documented install (every Jev call is refused because admit_unlabeled defaults to false). Hoppin-Sorter/jev-router is not settled. README-only candidate gates: 8 to 6.
  • When Jev fails: 3 new rows under Coding agents, all routing fallbacks; n = 74 in 11 use cases. On a Jev error: fail-closed 38, fallback 20, fail-open 8, no-decision 5, advisory 3. The action goes ahead in 15 of 74 (unchanged). Routing fallbacks: 8.
  • Products: 99 rows (A 86, B 12, C 1), with jev-triage, the jev-for-all launcher and hermes-jev-mana added.
  • Real-time control (research only): two third-party harnesses that call Jev were read at pinned commits with every failure cell (a ViZDoom game loop that keeps its last action on a Jev error, and a MuJoCo simulation that stops); a third calls a local substitute model by default. The condition for the release-4 page is met; run 14 decides whether to build it. Nothing is published on that topic in this release.
  • Drift and defaults: pydantic-ai-slim checked at 2.55.0 (failure path unchanged; image input now rejected before the request). Patch releases @effect/ai-typesafe 4.0.3 and @openclaw/typesafe 2026.10.1 (the jev-triage host branch has the same effect up to the line read). SDK defaults read from the published packages: typesafe-sdk 0.7.4, 10 s per HTTP operation, 2 retries, 30 s retry budget; @typesafe-ai/sdk 0.6.0, 10,000 ms per attempt, 2 retries. LanceDB's TypeSafe reranker with no key raises at the first rerank call, with no fallback.
  • Home and access status: rendered from the run's status file: status unchanged, checked 11 Oct 2026, 05:17 UTC; API 90-day uptime 99.838%; the status page also lists api.us.typesafe.ai (not in TypeSafe's docs). Two notices shown; three current notices stay off the home page because none changes whether or how a new user can get access.
  • Audience research ran on 11 Oct; it does not change this release's pages.
  • Tools: the fit check and the cost worksheet were opened at tools/<name>.html, at tools/<name>/, through a simulated redirect and in their pages, at desktop and phone widths; every button and field was used.
  • Evidence: 11 Oct research report.

Release notes: 10 October 2026

  • No run on 9 October. The 10 October scan covered the full 48 hours since the 8 October cutoff (8 Oct 2026, 05:12 UTC to 10 Oct 2026, 05:11 UTC), with no gap and no overlap. A missed run does not postpone a notice's retirement: the subprocessor notice (shelf life ended 9 Oct) retired on 10 Oct.
  • New rule: absence is stated as an evidence class with its search boundary and time (decided 10 Oct). A sentence that something does not exist is written as "not found on <surfaces> at <time>". Reason: readers asking "is Jev open weight?" or "jev ollama" can land on a Jev-named model from another account; "Jev has no open weights" would claim more than a bounded search shows, and a later TypeSafe release would make it silently wrong.
  • New rule: alternatives are sorted by the layer they replace (decided 10 Oct). A hosted API from another provider, open weights, a local runtime and a Jev-shaped request adapter are separate layers in the layer table and the filter, and the layer counts equal the table's rows by layer. A Jev-named third-party model carries "Jev-named; not published by TypeSafe" and sits in a separate box, not in the table. Reason: OpenAI's Decisions API replaces the whole hosted service, while a local runtime replaces only where the model runs; mixing them hid which part of a Jev setup each substitute changes. The layer table had said 9 weights entries while the table had 10; with this run's rows it is 12.
  • Alternatives: Jev weights: not found on TypeSafe's site, docs, GitHub or Hugging Face (10 Oct). Of 30 listed substitutes: 12 model weights, 11 local runtimes, 4 request adapters, 2 classifier/LLM approaches, 1 OpenAI hosted API. The Decisions API is shown as an alternative, never as a product built on Jev, with its published price only; its quality evidence is "lead, not graded" (the leads point in different directions by task). Strands Decider: copies are published by Amazon's verified Hugging Face organisation (Documented); The New Stack reports it as an AWS project (Reported).
  • Coding agents: Of 22 integrations read in source, 10 act on Jev's answer, 10 do not and 2 are unsettled. On a Jev error the 10: 3 hold, 1 may ask the main model, 5 let the call run, 1 keeps its configured model. jev-effort-router's code rewrites the request to deepseek-v4.1-flash below its threshold, unlike its catalog text. jev-gateway falls back to the agent's own model and the call runs. jev-triage and hermes-jev-mana stay unsettled until their host paths are read. 35 leads triaged into six classes.
  • When Jev fails: 2 new rows under Coding agents; n = 71 in 11 use cases. On a Jev error: fail-closed 38, fallback 17, fail-open 8, no-decision 5, advisory 3. The action goes ahead in 15 of 71 (8 fail-open plus 7 fallbacks that can still end in the action, now including jev-gateway). jev-effort-router joins the routing fallbacks (5).
  • Benchmarks: jev-pulse graded A (code, metric and data opened at a1e796f; flags: small-n (40 queries), BM25-only baseline, single task, run outputs not committed). jev-rerank-bench, openai-decisions-vs-jev-vs-laya and jev-vs-decisions-rpg are leads, not graded: code and data are present at the pin, but their metric code was not yet opened. The 20 recompute-table cells now carry mobile labels.
  • Jev reranker and RAG: key finding rewritten because jev-pulse is now graded: "Of 3 rerunnable graded reranking-pipeline studies, 2 come from parties with a stake and 1 (jev-pulse: 40 queries, BM25-only baseline) states none; the other grade-A study measures relevance judgments. On a Jev error, of 5 implementations, 2 stop, 2 fall back, 1 passes unscreened passages." The 8 Oct finding said both rerunnable graded pipeline studies came from parties with a stake. jev-rerank-bench's task now uses the 10 Oct README wording.
  • Route and drift: OpenRouter's typesafe/jev-1.13 endpoint now states a 64,000-token context with a 32,000-token prompt limit (32,000 on 8 Oct). Patch releases: LiteLLM 1.104.2, typesafe-sdk 0.7.4, ai 7.0.137, @ai-sdk/typesafe-ai 3.0.17; minor release pydantic-ai-slim 2.55.0 (not yet diffed). Not re-audited.
  • Home and access status: rendered from the run's status file: status unchanged, checked 10 Oct 2026, 05:12 UTC; API 90-day uptime 99.835%. Two notices shown; three current notices not shown (a Vercel provider change, a vendor case study and a funding announcement), because none changes whether or how a new user can get access.
  • Tools: the fit check and the cost worksheet were opened at tools/<name>.html, at tools/<name>/ and in their pages, at desktop and phone widths; every button and field was used.
  • Evidence: 10 Oct research report.

Release notes: 8 October 2026

  • Counting rule for rerankers, SQL functions and document pipelines (decided 8 Oct): a reranker counts on when Jev fails when its own code reorders or cuts results; a document pipeline counts when its own code acts on Jev's answer. SQL extensions that only return Jev's answer into the user's query are not counted. Reason: the count is about code that acts on Jev's answer; a SQL function hands the value back to a query the builder writes, which decides what to do with it, while a reranker or pipeline decides itself what reaches the prompt or the stored document. The SQL extensions' failure cells are still published on Jev inside SQL and document pipelines.
  • When Jev fails: 5 new rows under the nearest existing domains (LiteLLM's compaction guardrail, jev-reranker and LanceDB's TypeSafeReranker under Frameworks; pg-redact under Content moderation; docjev under Email and lead); n = 69 in 11 use cases. On a Jev error: fail-closed 38, fallback 15, fail-open 8, no-decision 5, advisory 3. The action goes ahead in 14 of 69. Every count was recomputed by the publisher from the page's cells and matched the runner's script. Three counting choices were decided by the manager: LiteLLM's compaction guardrail is counted; pg-redact's malformed cell is fail-open (a broken body gives 502); openlayer's no-key cell is fallback-another backend (it fails closed if no other backend is configured).
  • Reranking studies: each is labelled as a relevance-judgment study or a reranking-pipeline study, with grade and conflict-of-interest flags. iwhalen (grade A) measures agreement with human relevance grades, not a reranking pipeline. jev-pulse and anessbelbati/jev-rerank-bench are listed as ungraded candidates whose operators state no stake; the key finding therefore speaks of graded studies. LanceDB's benchmark script was opened, so its grade B is no longer provisional.
  • Corrections: jev-approvals' Jev-error cell (fallback-main agent model; held only on other HTTP errors) on coding agents, when Jev fails and Products; rh-guard's below-threshold text (0.5 is Jev's choice confidence; REVIEW_THRESHOLD 0.45 is a risk-label floor); the Use cases hub showed B14 as grade A provisional (grade B since 7 Oct).
  • Cells filled: jev-curator malformed (fail-closed), below threshold (held) and no key (fail-closed); openlayer no key. oh-my-claudecode v5.6.2 and @jkudish/jev-mcp 0.14.1: failure paths unchanged.
  • Benchmarks hygiene: personal names replaced by the organisation or blog domain (B16, B20, the leads paragraph); B29 note; MindStudio's post named as not graded; leads added for jev-pulse, anessbelbati/jev-rerank-bench and TypeSafe's Jack & Jill case study (vendor-and-customer claim, Reported, no code or data; not shown on Home by manager decision).
  • Cross-page consistency: the five reranking and passage-filter rows carry the same cells, colours and pins on Jev reranker and Jev for RAG, frameworks and when Jev fails; the pg-redact and docjev rows match on Jev inside SQL and document pipelines and when Jev fails; jev-approvals, jev-curator and openlayer match on coding agents, Jev as a judge, Products and when Jev fails. The hubs, confidence thresholds and Home carry the new counts.
  • Home and access status: rendered from the run's status file: status unchanged, checked 8 Oct 2026, 05:14 UTC; the API's 90-day uptime moved from 99.828% to 99.831% with no new incident listed. Four notices shown; two current notices not shown (the Vercel provider notice and the case study). The access timeline has an 8 Oct recheck row.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/, through a simulated redirect and inside their pages (both addresses), at 1280 and 390 pixels wide. Every button and field was used. 810 of 810 checks passed, with no script error and no failed load of a pool file. The tool files did not change after the test.
  • Evidence: 8 Oct research report.

Release notes: 7 October 2026

  • Counting rule for judges (decided 7 Oct): a judge or eval tool whose own code turns Jev's answer into the score, pass/fail result, ranking or gate that its run, report or CI uses is an action and counts on when Jev fails. An error that becomes a pass is fail-open; a dropped item is no-decision. Reason: a judge that turns a failed Jev call into a pass lets a broken release through as surely as a hook that lets a tool call run. DeepEval counts although it is a library, because its own test runner aborts the run or fails the test case. The Jevals harness is not counted: its code is not public.
  • When Jev fails: 6 new rows (4 judge tools, 2 Hermes Agent plugins) and a new "LLM as a judge" domain; n = 64 in 11 use cases. On a Jev error: fail-closed 35, fail-open 8, advisory 3, fallback 13, no-decision 5. The action goes ahead in 13 of 64. The key finding was edited in publication review to name every category (35 + 13 + 8 + 8 = 64). The copy block was regenerated, and the line saying judges were "not yet in this count" was removed.
  • Jev as a judge: three cells that read "conditional" now show the counted term with the condition in the note, so that both pages show the same cell and colour: openlayer below threshold (no-threshold) and no key (not recorded), and did-they-answer malformed (fail-closed). deepeval 4.2.8 and jevals 0.1.4 are still the latest releases.
  • Coding agents and Products: jev-approvals and jev-curator are counted; both do nothing until switched on. opencode-tool-gate's "Yes" is confirmed under OpenCode's default rules; rh-guard's default mode is enforce; osuki was checked at 0.2.8 (unchanged). Products moved the two Hermes rows to the counted group (8 act, 10 not counted); row and grade counts are unchanged.
  • Benchmarks: rows B28–B36 added and Elastic B14 regraded from A (provisional) to B. The title was changed to the searched phrase "jev benchmark" (Bing and DuckDuckGo completions, and "jev 1.13 benchmarks" on Google, rechecked 7 Oct 2026, 05:18:59 UTC). TypeSafe's legal texts still have no clause about publishing benchmarks. Reranking studies: 2 of 2 needed found beyond Parallel (B04) and the earlier MindStudio lead, so the condition for a reranking and RAG page is met for the next run.
  • Drift: ai 7.0.130 and @ai-sdk/typesafe-ai 3.0.15 noted on TypeScript and frameworks; oh-my-claudecode v5.6.2 and @jkudish/jev-mcp 0.14.1 noted on Claude Code and MCP as not yet read; six framework packages and LiteLLM 1.104.0 unchanged.
  • Cross-page consistency: the four judge rows carry the same cell, term and colour on Jev as a judge and when Jev fails. The settled coding-agent cells match on coding agents, when Jev fails and Products. osuki's version (0.2.7 read, 0.2.8 checked) is stated the same way on all three. The hubs (Build, Use cases), confidence thresholds and Home carry the new counts.
  • Home and access status: rendered from the run's status file: status unchanged, checked 7 Oct 2026, 05:19 UTC; the API's 90-day uptime moved from 99.826% to 99.828% with no new incident listed. Four notices shown; the Vercel provider notice is current but not shown; typesafe-workflow-evals-code retired. The access timeline has a 7 Oct recheck row.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages at 1280 and 390 pixels wide. Every button and field was used: all 8 answer combinations, the two-fix case, full and partial reset and the Enter key in the fit check; every field with checked results, seven out-of-range values, reset and typed input in the cost worksheet. 294 of 294 checks passed, with no script error and no failed load of a pool file.
  • Evidence: 7 Oct research report.

Release notes: 6 October 2026

  • New pages: coding agents (18 integrations at pinned commits: 6 counted rows in the shared failure terms with a "does the tool call still run?" column, 12 not counted, split 8 / 2 / 2; the Hermes catalog's installed commits; jev-judge is shadow by default and its cells describe enforce mode; osuki 0.2.8 not read) and model IDs and aliases by route (6 routes with their own dates, or "undated; read at" where the source shows none). Release 3 is complete (9 of 9); release 4 has 1 of 4.
  • Editorial correction to the coding-agents key finding: the research draft said "12 only advise the agent". Of the 12 integrations not counted, 8 return Jev's answer to the agent, 2 give advisory hints and 2 are not settled (jev-approvals hands its verdict to Hermes core, which was not read; jev-curator was only partly read). The page and the report use the corrected sentence.
  • Claude Code and MCP: two guard rows (jev-permission-gate counted, fallback to the built-in classifier; jev-llm-guard "returns verdict; not a decision"), the count sentence now reads seven guardrails, a pointer to coding agents for jev-kit, jev-review and mcp_jev, and the oh-my-claudecode v5.6.2 tag note.
  • When Jev fails: 7 new rows (1 Claude Code guard, 6 coding agents) and a new "Coding agents" domain; totals recomputed for n = 58 in 10 use cases; the action goes ahead on a Jev error in 12 of 58; the copyable checklist was regenerated. The "without trading" sub-row was removed because it no longer matched the domains.
  • Products: 92 rows (A 75, B 16, C 1), 20 rows added; two older B rows with similar names are different projects and stay.
  • Where to use and data privacy: Vercel lists two providers for Jev (TypeSafe AI, zero data retention; DigitalOcean, no zero data retention) and picks by uptime and latency unless you set the order; route fields re-dated only where rechecked on 6 Oct. Links to the model-ID page from the route matrix, TypeScript, Python and access status.
  • Drift: LiteLLM 1.104.0 failure path unchanged (with the commit-order caveat) on model routing; framework version box rechecked; ai 7.0.128 and @ai-sdk/typesafe-ai 3.0.13 noted on TypeScript.
  • Cross-page consistency (new standing check): every failure verdict cell on 13 pages was compared with the row of the same pinned commit on when Jev fails (support-ticket triage, model routing, content moderation, email and lead sorting, trading and fraud gates, Jev as a judge, browser and computer use, n8n, Claude Code and MCP, framework integrations, coding agents, when Jev fails, Products) plus the Use cases hub and the errors page. Fixed: Claude Code and MCP, the Use cases hub table and the errors-page disposition column used the reverse colours (stops red, proceeds green); support-ticket triage, model routing, content moderation and n8n used their own colour schemes and plain words; Products tallies and chips used blue for fail-closed and amber for fallback; browser below-threshold "no-threshold" cells were red; QuantDinger and btc-hft-jev fallbacks were red on trading; fraud-jev's conditional malformed cell was amber; ilyamk's unsure cell was red. All now use the shared legend and lead with the shared term; value columns (timeouts, data sent, appeal) are no longer coloured. LangChain AutoModeMiddleware now links commit 4af7ab8 (tag langchain-typesafe==0.0.1a3) instead of PyPI on when Jev fails, Claude Code and MCP, and Products. Open: oh-my-claudecode is shown as a heuristic fallback on Claude Code and MCP (the path once Jev is asked) but counted as advisory on when Jev fails (its default shadow mode), and Products links its v5.6.0 commit while the other pages link v5.5.0; Jev as a judge's five rows are not on the checklist.
  • Home and access status: rendered from the run's status file: status unchanged, checked 6 Oct 2026, 05:15 UTC; the API's 90-day uptime moved from 99.829% to 99.826% with no new incident listed. Four notices shown; the new Vercel provider notice and typesafe-workflow-evals-code are current but not shown. None retired. The access timeline has a 6 Oct recheck row. Home task paths link the two new pages.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages at 1280 and 390 pixels wide. Every button and field was used: all 8 answer combinations, the two-fix case, reset and the Enter key in the fit check; every field with checked results, six out-of-range values, reset and typed input in the cost worksheet. 220 of 220 checks passed, with no script or load error and no scroll inside either frame. Only the cost worksheet's price check date changed (6 Oct 2026, 05:15 UTC). No new tool this release.
  • Evidence: 6 Oct research report.

Release notes: 5 October 2026

  • New pages: browser and computer use (seven acting agents and one recommendation-only extension at pinned commits; a "does a failure still click, type or submit?" column; trycua/cua listed, not counted) and framework integrations (six packages at pinned tags or commits; a maturity matrix; coding agents beyond Claude Code are planned for a later page).
  • Where to use: new section on Amazon Bedrock and Azure AI Foundry with a route-matrix row; availability is written as "not listed in the catalog at the check time", with catalog links. The lookalike reseller's Bedrock page is cited only as a reseller fact. The "Access today" line now matches Home (checked 5 Oct 2026, 05:13 UTC).
  • When Jev fails: 12 new rows (8 browser, 4 frameworks), totals recomputed for n = 51 in 9 use cases; the four counting judgements stay visible on the page; a copyable checklist in plain text repeats the eight questions and their counts.
  • Products refresh: 72 rows (A 55, B 16, C 1); 3 rows upgraded from B to A, 5 added, jev-ultrafast re-pinned; six framework packages listed as not counted.
  • Home and access status: rendered from the run's status file: status unchanged, checked 5 Oct 2026, 05:13 UTC; the API's 90-day uptime moved from 99.826% to 99.829%. Four notices shown; typesafe-workflow-evals-code stays current at rank 5, not shown. None retired. The access timeline has a 5 Oct recheck row.
  • Small additions and re-dating: Zapier note on the n8n page, EU-hosting line on data privacy, Strands Decider 2B on alternatives, tool-documentation links on Jev as a judge; limits re-dated to 5 Oct on errors and rate limits and limitations; list price re-dated on the cost worksheet. Links to the new pages from the Use cases and Build hubs, Python, TypeScript, Claude Code and MCP, model routing, videos, official vs reseller and data privacy.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages at 1280 and 390 pixels wide. Every button and field was used: all 8 answer combinations, reset and the Enter key in the fit check; every field with checked results, six out-of-range values, reset and typed input in the cost worksheet. 250 of 250 checks passed, with no script or load error and no scroll inside either frame. Only the cost worksheet's price check date changed (5 Oct 2026, 05:13 UTC). No new tool this release.
  • Evidence: 5 Oct research report.

Release notes: 4 October 2026

  • New pages: trading and fraud gates (seven projects at pinned commits; a "does the failure reach the order or payment?" column; QuantDinger's order call sites traced), when Jev fails (counts recomputed from the 7 published domain tables, n = 39, with "not recorded" counts and an 8-question checklist) and Jev as a judge (four tools at pinned commits; benchmark rows B02, B06 and B08 copied unchanged).
  • Products refresh: 67 rows (A 47, B 19, C 1); 2 rows upgraded from B to A, 6 added, QuantDinger re-pinned; 2 library features listed as not counted.
  • Home and access status: rendered from the run's status file: status unchanged, checked 4 Oct 2026, 05:15 UTC. By the manager's decision, subprocessors-added-2026-10-02 is shown at rank 4 and typesafe-workflow-evals-code stays current at rank 5, not shown (it retires 6 Oct). None retired. The access timeline has a 4 Oct recheck row.
  • Corrections and updates: email and lead sorting (two relabels, cost range), Claude Code and MCP (advisory wording for oh-my-claudecode at default settings; v5.6.1 note), testing without a key and the Use cases hub (QuantDinger, jev-trader and Jev Trade re-read at pinned commits; the hub now lists 26 patterns, 11 with project code read), data privacy (subprocessors, SOC 2 listed but not seen, a Netlify docs sentence), re-dated checks on Where to use, errors and rate limits, limitations and cost per decision. Links to the new pages from the Build and Use cases hubs, the six domain pages, confidence thresholds, benchmarks, Jev vs LLM vs classifier, videos and errors and rate limits.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages at 1280 and 390 pixels wide. Every button and field was used: all 8 answer combinations and reset in the fit check; every field and reset in the cost worksheet. 184 of 184 checks passed, with no script or load error and no scroll inside either frame. Only the cost worksheet's price check date changed (4 Oct 2026, 05:15 UTC).
  • Evidence: 4 Oct research report.

Release notes: 3 October 2026

  • New pages: email and lead sorting (five projects read at pinned commits; cost per 1,000 emails by pool arithmetic, shown apart from a Reported demo figure) and the Jev n8n node (official node and four community nodes read at pinned commits; npm maintainer and provenance records). Each opens with the run's key finding, its label and check date, and uses the title proposed from search language.
  • Products refresh: 24 new grade-A rows (triage, routing, moderation, email and lead, n8n), claims rechecked, repository status checked, LiteLLM listed under "not counted as products".
  • Home and access status: rendered from the run's status file: status unchanged, checked 3 Oct 2026, 05:16 UTC; the notice rate-limits-changed was retired as superseded and rate-limits-changed-2026-10-03 (80 requests per second) added; the other three notices kept. One timeline event added.
  • Corrections and updates: Claude Code and MCP (oh-my-claudecode v5.6.0 row, version line and check 6), errors and rate limits (80 requests per second with its dated bracket; the 2,000 ms oh-my-claudecode default released in v5.6.0), limitations (rate-limit row), Where to use (limit and check dates), official vs reseller (n8n npm publisher answered), Use cases hub (Jevmail row). New pages are linked from their hubs and related pages.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages, and every button and field was used. No tool content changed.
  • Evidence: 3 Oct research report.

Release notes: 2 October 2026

  • New pages: model routing (OpenRouter's Jev Router docs read with explicit "not documented" results; six routers read at pinned commits) and content moderation (six projects read at pinned commits; Noul-per-harm and Choice compared from TypeSafe's docs). Each opens with the run's key-finding sentence, its label and check date, and uses the title proposed from search language.
  • Home and access status: rendered from the run's status file: status unchanged, checked 2 Oct 2026, 05:14 UTC; all four notices kept (none retired); API 90-day uptime 99.826%.
  • Corrections and updates: Claude Code and MCP (version check replaces the drift note; stale oh-my-claudecode line fixed), TypeScript (ai 7.0.127), Where to use (Jev Router row links model routing; check dates), data privacy (check dates), Use cases hub (routing row, moderation rows, WordPress commit), and links from benchmarks, confidence thresholds, limitations, support-ticket triage and videos. The Jev vs Laya vs Kev page's heading now starts with "Jev vs Laya", as people type it; its URL is unchanged.
  • Tools: both tools were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages, and every button and field was used. Only the cost worksheet's price check date changed.
  • Evidence: 2 Oct research report.

Release notes: 1 October 2026

  • New pages: Jev vs Laya vs Kev (licence stack, context and API shape documented for all three; three graded rows compare at least two), data handling by route (TypeSafe direct and four gateways quoted at both layers; Netlify not read) and support-ticket triage (five projects read at pinned commits). Each opens with the run's key-finding sentence, its label and check date.
  • Home and access status: rendered from the run's status file: status unchanged, checked 1 Oct 2026, 05:16 UTC; all four notices kept (none retired); the rate-limit notice now shows the narrowed time range.
  • Corrections: where-to-use (Vercel data cell; two caveats replaced with links to the data page; each route's data cell rewritten), alternatives (Kev and Laya rows, he-jev note), rate-limit dating (access status, errors and rate limits, limitations, Home), version-drift notes (TypeScript, Claude Code and MCP), and the residency row on Jev vs an LLM vs a classifier.
  • Tools: both tools (fit check and cost worksheet) were opened in a headless browser at tools/<name>.html, at tools/<name>/ and inside their pages, and every button and field was used; the tools themselves were not changed.
  • Evidence: 1 Oct research report.

Release notes: 30 September 2026

  • Tools: the fit check and cost worksheet are available from their topic pages.
  • New pages: benchmarks (12 studies met the plan's rule of at least 8 read at the primary source with operator, task, n, baseline and materials; 9 compare Jev with a baseline) and Jev vs an LLM vs a classifier (10 criteria with cited evidence, 13 head-to-head rows).
  • Home: rebuilt around the owner's order: access status, notices, short answer, direct paths. The status matches the access status page exactly (same values and check time).
  • Corrections: rate limits (where to use, errors and rate limits, limitations); the alexmolas.com row on confidence thresholds; the home chart's accuracy basis. No page of this site had presented TypeSafe's 67.8% as accuracy; the benchmarks page now explains it.
  • Re-dated only (rechecked, unchanged): TypeSafe price and aliases; OpenRouter, Vercel, Cloudflare and Opper listings; typesafe-sdk 0.7.2 and @typesafe-ai/sdk 0.6.0; the six Claude Code and MCP integrations. Netlify, reseller prices and other fields keep their earlier dates.
  • Evidence: 30 Sep evaluation-evidence report.

Old-to-new URL map

No page moved in release 1 or on 29 September, so no redirects are needed. Every earlier address still opens the same topic. Base address: https://shaduf.ai/p/jev-catalog/.

Old URLNew URLStatusWhat changed and why
/p/jev-catalog/ (home)sameReworkedBecame a question router with an evidence-status chart, so readers reach the page for their question instead of the latest run.
what-is-jevsameExtendedAdded an annotated request and response, current model facts and the name disambiguation, because first-time readers need one concrete example.
use-casessameRebuilt (hub)Pattern picker with evidence types, video links, TypeSafe's 23 official patterns and cookbooks, and a failure matrix; every earlier row and link kept.
alternativessameRebuilt (hub)One table with the same columns for every row, a layer guide, and the new Ollaya and Jevper rows; three 22 Sep rows now live only in their report. Retitled 10 Oct 2026 to "Jev alternatives (TypeSafe AI): open-source models, local runtimes and hosted APIs, and what each replaces" (URL unchanged): Google completes "jev alt" to "jev alternatives", and Bing and DuckDuckGo complete it to "jev alternative" (10 Oct 2026, 05:17 UTC; Bing and DuckDuckGo returned identical lists). A demand signal, not a volume.
productssameRebuilt (hub)Graded ledger (all 32 rows kept, two labels corrected) and a new production-claims ledger, because "who uses Jev in production" is a repeated question.
where-to-usesameRebuilt (hub)Route guide and matrix with the Jev Router row, Vercel's exact price and data terms; the reseller row moved to its own page.
researchsameExtendedThis page: all reports, release notes, URL map, evidence gaps and scan summary.
reports/… (13 reports)sameUnchangedDated snapshots; not edited.
tools/fit-check.htmlsameRebuiltSame three questions, now tied to TypeSafe's own guidance and explained on the Limitations page.
—limitationsNew"What can Jev not do?" had no page; it now lists the 17 limits TypeSafe documents, with workarounds, and explains the fit check.
—where-to-use/access-statusNewAccess rules changed several times in two weeks and third-party pages disagree; a dated timeline with primary sources settles it.
—where-to-use/official-vs-resellerNewLookalike resellers rank for access searches at several times the list price; readers need a check and the displayed prices.
—where-to-use/cost-per-decisionNewCost depends on counted input tokens, which competing calculators ignore; the worksheet uses the reader's own counts.
—build/pythonNewA frequent build question; version drift (SDK 0.5.7 to 0.7.2 in two weeks) makes a pinned example useful.
—build/errors-and-rate-limitsNewThe official docs list four codes while users see others; a documented-vs-reported matrix and fallback procedure did not exist.
—build/confidence-thresholdsNewConfidence can be misread as the probability of being right; this page explains the fields and a cutoff procedure. It maintains the 25 Sep protocol.
—videosNewOwner-required: videos that show Jev working, with the second the demo starts, excluding overview and farm videos.
—reports/release-1-question-pagesNew reportThe dated evidence for this release.
—buildNew, 29 SepThe Build group reached six pages; the hub compares code interfaces and orders the build steps.
—build/typescriptNew, 29 SepTypeScript readers had no page; four paths differ in runtime, key variable, field names, retries and timeouts.
—build/claude-code-mcpNew, 29 SepAgent builders ask which Jev guard fails open; the answer needed a source audit, not search summaries.
—build/testing-without-a-keyNew, 29 SepNew users cannot get free credit, and no-key modes were found to bypass Jev silently in audited projects.
—reports/integration-trustNew report, 29 SepThe dated evidence for the 29 Sep pages and corrections.
—benchmarksNew, 30 Sep; retitled 7 Oct"Is Jev accurate?" needed published tests graded by what can be checked, not quoted headlines. Retitled 7 Oct to "Jev benchmarks (TypeSafe AI): is Jev accurate? A graded ledger of published studies". Title query source: Bing and DuckDuckGo completed "jev b" and "jev bank" to "jev benchmark", and Google completed "jev 1.13" to "jev 1.13 benchmarks" (4 Oct, 05:40–05:41 UTC; rechecked 7 Oct 2026, 05:18:59 UTC). A suggestion is a demand signal, not a volume.
—use-cases/reranking-ragNew, 8 OctBuilders choosing a reranker need both the evidence by study type and what happens to their passages when Jev fails. Title query source: Bing and DuckDuckGo complete "jev rer", "jev rera", "jev rerank" and "jev reranking" to "jev reranker" (8 Oct 2026, 05:16:54–05:17:20 UTC); Google completes "jev for r" and "jev for ra" to "jev for rag" (05:17:19 UTC). A suggestion is a demand signal, not a volume.
—use-cases/databases-and-documentsNew, 8 OctCalling Jev from SQL or a document pipeline raises its own questions: does an error abort the query, and what leaves the database? Internal label: no query source ("jev postgres" is completed by Google only; "jev sql", "jev sqlite", "jev database", "jev extraction", "jev pdf" and "pg-jev" by no endpoint, 8 Oct 2026).
—reports/reranking-databasesNew report, 8 OctThe dated evidence for the 8 Oct pages, recount and corrections.
—use-cases/real-time-controlNew, 12 OctA game or robot loop keeps running while it waits for Jev, so what a demo does on a failed call is the question. Title query source: Google (suggestqueries.google.com, client=firefox) shows "jev plays doom" first for the shorter prefix "jev plays"; Bing (osjson.aspx) and DuckDuckGo (ac), one source, show it first for "jev plays" and list it for "jev doom" (12 Oct 2026, 05:16:13–05:16:14 UTC). A demand signal, not a volume. Caveat (hypothesis): Google's "jev plays" list also suggests games and an account that belong to a gamer named Jev, so part of this demand may not be about the model; the title keeps "(TypeSafe AI)".
—reports/real-time-control-decision-1New report, 12 OctThe dated evidence for the 12 Oct page, recount and corrections.
—alternatives/jev-vs-decision-apisNew, 13 OctBuilders comparing Jev with OpenAI's and Microsoft's hosted decision APIs need graded results per task, prices as published and data terms in one dated place. Query source: Google autocomplete completes 'decisions api' and 'decisions api vs' to 'decisions api vs jev', and 'microsoft decision' to 'microsoft decision 1' and 'microsoft decision 1 vs jev' (13 Oct 2026, 05:21 UTC). 'jev vs openai' appeared only as an echo of the typed prefix. A suggestion is a demand signal, not a search volume.
—reports/decision-api-comparisonNew report, 13 OctThe dated evidence for the 13 Oct page, the no-key and bypass recount, B42 and corrections.
build/pythonsameRetitled, 14 OctNew title "Jev Python SDK (TypeSafe AI): a version-pinned call, timeouts, retries and missing-key errors". Query source: "jev py" completes to "jev python" as the first suggestion on Google (suggestqueries.google.com, client=firefox) and on Bing (osjson.aspx) and DuckDuckGo (ac), one source (14 Oct 2026, 16:50 UTC). "jev python sdk" returned no suggestion on any of them. A demand signal, not a volume.
build/typescriptsameRetitled, 14 OctNew title "Jev TypeScript (TypeSafe AI): @typesafe-ai/sdk vs AI SDK vs fetch, with timeouts and missing-key errors"; the old title named experimental_evaluate, now a deprecated name. Query source: "jev t" and "jev ty" complete to "jev typescript" (second suggestion) on Bing (osjson.aspx) and DuckDuckGo (ac), one source (14 Oct 2026, 16:50 UTC). Google shows the phrase only for the full query. A demand signal, not a volume.
—reports/failure-cells-sdk-refreshNew report, 14 OctThe dated evidence for the 14 Oct failure-cell recount, the SDK refresh, B43 and B44, and the SubQ Snap row.
build/claude-code-mcpsameRetitled, 16 OctNew title "Jev Claude Code and MCP guardrails (TypeSafe AI): which hook, plugin or server fails open?". Query source: Google autocomplete "jev cl" → "jev claude code"; Bing and DuckDuckGo "jev cl" → "jev claude code" (16 Oct 2026). A demand signal, not a volume.
where-to-use/versions-and-aliaseswhere-to-use/versions-and-aliases (unchanged)Retitled, 7 NovNew title and H1 "Jev 1.14 vs 1.13 (TypeSafe AI): model IDs, jev-latest and aliases by route" (was "Jev 1.14, 1.13 and jev-latest (TypeSafe AI): model IDs and aliases by route"). Title query source: Google completes the shorter prefix "jev 1.14 vs" to "jev 1.14 vs 1.13" (its only suggestion; 7 Nov 2026, 15:34:03 UTC, R27-S221); Bing and DuckDuckGo return no completion (15:34:04 UTC, R27-S222, R27-S223). A demand signal, not a volume; the title rests on Google only.
—reports/bypass-cells-claude-code-refreshNew report, 16 OctThe dated evidence for the 16 Oct bypass and malformed cells, the Jev 1.14 preview, the Claude Code and MCP recheck and the Decision Index line.
—reports/failure-terms-claude-code-gates-productsNew report, 18 OctThe dated evidence for the 18 Oct failure-term comparison across 15 pages, the three Claude Code guardrails, the products ledger at 113 rows and the smaller route and SDK items.
—alternatives/jev-vs-llm-vs-classifierNew, 30 SepChoosing between Jev, an LLM and a classifier is a different question from choosing an open Jev-like model.
—reports/evaluation-evidenceNew report, 30 SepThe dated evidence for the 30 Sep pages and corrections.
—alternatives/laya-vs-kevNew, 1 Oct; heading changed 2 Oct"jev vs laya" is a YouTube autocomplete suggestion (audience research, 28 Sep); the hub keeps one row per model, this page compares three on the same rows. On 2 Oct its heading was changed to start with "Jev vs Laya" as typed; the URL did not change.
—where-to-use/data-privacyNew, 1 OctThird-party pages disagree on whether Jev trains on or keeps your data; the hub keeps one data cell per route and links here for the quotes.
—use-cases/support-ticket-triageNew, 1 OctA frequent concrete use case (272, 70 and 158 GitHub results for "jev triage", "jev ticket" and "jev support" on 1 Oct, not all relevant); no page showed what real projects do when Jev fails.
—reports/alternatives-data-handling-triageNew report, 1 OctThe dated evidence for the 1 Oct pages and corrections.
—use-cases/model-routingNew, 2 OctTitle query source: Bing and DuckDuckGo autocomplete suggest "jev router" for "jev r", "jev ro" and "jev routing" (2 Oct, 05:15 UTC); YouTube autocomplete suggested it on 28 Sep. No page showed where a request goes when Jev fails inside a router. The Where-to-use hub keeps the Jev Router route row and links here.
—use-cases/content-moderationNew, 2 OctTitle query source: independent publisher titles ("Content Moderation With Jev: Rules, Thresholds, Cost"; "Jev Content Moderation Recipe & Schema Guide"). No autocomplete suggestion for moderation was found on YouTube, Bing, DuckDuckGo or Google (2 Oct). The advice pages give thresholds but no failure handling; this page reads what code does.
—reports/routing-moderationNew report, 2 OctThe dated evidence for the 2 Oct pages and corrections.
—use-cases/email-and-lead-sortingNew, 3 OctTitle query source: Google suggested "jev email classifier" for "jev email" (suggestqueries.google.com, 3 Oct, 05:22 UTC). No page showed what real email and lead sorters do to a mailbox when Jev fails. Help-desk tickets stay on support-ticket triage.
—build/n8nNew, 3 OctTitle query source: "jev n8n" was a Google and YouTube autocomplete suggestion (3 Oct, 05:24 UTC). Two npm packages with near-identical names come from two publishers; this page says which is TypeSafe's and what each node does when Jev fails.
productssameHeading changed, 3 OctTitle query source: "jev examples" was offered by all four autocomplete endpoints checked (Google, YouTube, Bing, DuckDuckGo; 3 Oct, 05:22 UTC), so the heading now starts "Jev examples". The URL is unchanged.
—reports/email-products-n8nNew report, 3 OctThe dated evidence for the 3 Oct pages and corrections.
—use-cases/trading-and-fraud-gatesNew, 4 OctTitle query source: Google suggested "jev trading bot" (suggestqueries.google.com, client=firefox, 4 Oct 2026, 05:21 UTC); Bing and DuckDuckGo suggested "jev trader". "jev fraud", "jev fraud detection" and "jev finance" had no relevant suggestion, so "fraud" is not in the H1. A trade or payment is where a Jev failure can reach money.
—build/when-jev-failsNew, 4 OctInternal label: six failure phrases ("jev fallback", "jev timeout", "jev error handling" and others) got no relevant suggestion on Google, YouTube, Bing or DuckDuckGo (4 Oct, 05:21 UTC; same result on 3 Oct). Six domain pages published failure columns that readers could not compare.
—use-cases/llm-as-judgeNew, 4 OctTitle query source: Google suggested "jev llm as a judge" and "jev as a judge"; Bing and DuckDuckGo suggested "jev as a judge" (4 Oct, 05:21 UTC). The benchmarks page grades judge studies, but no page said what a judge harness records when Jev fails.
—reports/trading-failures-judgeNew report, 4 OctThe dated evidence for the 4 Oct pages and corrections.
—use-cases/browser-and-computer-useNew, 5 OctTitle query source: "jev browser use" and "jev computer use" suggested on Google (suggestqueries.google.com, client=firefox), Bing and DuckDuckGo (5 Oct 2026, 05:14 UTC), confirming the 4 Oct audience research. A browser agent turns a Jev answer straight into a click or a form submission.
—build/frameworksNew, 5 OctTitle query source: "jev langchain" suggested on Google, Bing and DuckDuckGo (5 Oct 2026, 05:14 UTC); "jev pydantic", "jev spring ai" and "jev llamaindex" had no suggestion, so only LangChain is in the H1. The packages are young and had not been compared.
where-to-usesame, new anchor #bedrock-azureNew section, 5 OctHeading query source: "jev on bedrock" suggested on Google; Bing and DuckDuckGo give "jev bedrock" for the same prefix; "jev azure foundry" on all three (5 Oct 2026, 05:14 UTC). A lookalike reseller answered the question before any pool page did.
—reports/browser-frameworksNew report, 5 OctThe dated evidence for the 5 Oct pages and corrections.
—build/coding-agentsNew, 6 OctTitle query source: Google (suggestqueries.google.com, client=firefox) suggests "how to use jev with codex", "how to use jev with opencode", "jev in cursor" and "jev openclaw"; "jev hermes" and "jev codex" appear on Google, Bing and DuckDuckGo (rechecked 6 Oct 2026, 05:19 UTC). Every host in the H1 has a counted row. A coding agent turns a Jev answer into a shell command, a file write or a web fetch.
—where-to-use/versions-and-aliasesNew, 6 OctTitle query source: "jev 1.13" and "jev-latest" suggested on Google, Bing and DuckDuckGo (rechecked 6 Oct 2026, 05:19 UTC); "jev 1.14", "jev versions" and "jev model version" had no supporting suggestion. Routes accept different ID shapes, and an unversioned ID can change model when TypeSafe ships a new version.
where-to-use/versions-and-aliaseswhere-to-use/versions-and-aliases (unchanged)Retitled, 16 OctNew title and H1 "Jev 1.14 preview, 1.13 and jev-latest (TypeSafe AI): model IDs and aliases by route", so the page no longer presents 1.13 as the only current model. Title query source: no "jev 1.14" suggestion (16 Oct 2026, 05:28 UTC: Google returns only the typed "jev 1.14"; Bing and DuckDuckGo suggest "jev 1.13" and "jev 1.13 free"), so the title follows TypeSafe's models page, which lists Jev 1.14 and Jev 1.13.
—reports/coding-agents-versionsNew report, 6 OctThe dated evidence for the 6 Oct pages and corrections.

Release notes: 29 September 2026

  • New pages: Build hub, TypeScript, Claude Code and MCP guardrails and testing without a key. Each met the research plan's rule for shipping. For example, the Claude Code page needed four or more integrations read at a pinned commit; six were read.
  • Corrections: oh-my-claudecode's 2000 ms Jev timeout exists only on its dev branch; v5.5.0 and main still default to 250 ms. The Products row "jev-mcp" is now "jev-mcp (jkudish)", regraded from B to A and marked advisory.
  • Unchanged, re-dated: signup status, new-user credit, status page, model aliases, the TypeSafe price, the OpenRouter, Vercel, Cloudflare and Opper listings, and both SDK versions.
  • Evidence: 29 Sep integration-trust report.

Release notes: release 1, 28 September 2026

The owner asked for an evidence-backed resource ordered by reader question. Release 1 adds or substantially rebuilds 13 pages and lightly extends two. The reason for each page is in the URL map below. Pages whose evidence was not ready were not published thin: the planned pages on TypeScript, Claude Code and MCP, benchmarks, data handling by route, ticket triage and model routing are scheduled for later releases.

  • Corrections in this release: two project rows were labelled as official TypeSafe repositories but come from the unofficial github.com/TypeSafeAI community organisation; Vercel's exact price is now documented ($0.042 per million in its models API); the "250 ms default timeout" is not TypeSafe's; the confidence formula 2·p_top − 1 applies to two-option Choices only; the $5 new-user credit rests on secondary reports with a conflicting first-hand report.
  • New tool: the cost per decision worksheet, explained on its page.
  • Evidence for this release: release 1 research report.

How pages were chosen: no search data

Search impressions, clicks, query coverage and indexing data, Search Console, keyword volumes and site analytics were not available for release 1 (28 September 2026), and were still unavailable for the 29 and 30 September and 1, 2, 3, 4, 5, 6, 7, 8, 10, 11, 12, 13, 14, 16, 18, 19, 20, 21 and 23 October updates (no update was published on 9 October). Topics were chosen from qualitative signals: search autocomplete, titles published by independent sites, Hacker News and GitHub issue threads, and video suggestions. No page on this site claims a search volume. When search data becomes available, it will be used to choose and rank the next pages.

Audience research refreshed 4 Oct 2026: autocomplete on Google, YouTube, Bing and DuckDuckGo plus competitor and forum checks; search analytics still unavailable; Reddit and X not accessible.

Evidence gaps after the 9 November research

  1. Benchmarks. Only one task of S1CogBench (B49) was read; the raw Jev responses behind it are not published. The small-LLM count has not been redone with B49 to B51. Newer benchmark repositories seen in the scan were not opened.
  2. Leaderboards. The position the Jev Decision Index app shows was not re-checked; Artificial Analysis's changed method was not read.
  3. auto-mode. The mod host's handling of auto-mode's answer and the internals of its 7 skip stages were not read, so the row stays unsettled.
  4. Data terms. OpenCode's Privacy Policy, which defines "Connected Service", was not opened. Whether Nace.AI's 21 Oct draft-like Addendum lines are in its 6 Nov version was not recorded.
  5. Catalogs and scan coverage. Bedrock not determined; Foundry and Workers AI not re-read. Reddit, X, YouTube and TikTok were not covered for 8–9 Nov: unknown, not zero.

Evidence gaps after the 8 November research

  1. ohmyjev. Its Codex adapter (in the package since 0.13.2) and its screens, routing, done-check and compaction handlers were not read; the shadow setting was read in source, not run.
  2. Coding agents. Three further routers named as candidates (pi-jev-router-pro, jev-codex-router, opencode-jev-pruner) and the coding-agent names seen 24 Oct–8 Nov beyond the 19 listed were not read; auto-mode 4.0.0 (a major release of a row that is not settled) was not re-read.
  3. Catalogs. The Amazon Bedrock model lists did not render as readable text on 8 Nov; the 31 Oct reading stands. The Workers AI index may render part of its list by script.
  4. Data terms. SubQ's docs page gives no date for its 60-day text; Cloudflare's third retention exception was not quoted on 31 Oct, so whether it is new is not known.
  5. Scan coverage. Reddit, X, YouTube and TikTok were not covered for 7–8 Nov: unknown, not zero.

Evidence gaps after the 7 November research

  1. Images and Box. TypeSafe's models page and its Images page describe image input differently (a listed input vs an opt-in experiment "not yet ... ready for production"); TypeSafe does not reconcile them. No image or Box request was sent.
  2. Base Labs. OpenRouter's own data terms for the baseten/* endpoints were not read; Base Labs' rank claims were not matched (the leaderboard data returned HTTP 500).
  3. Comparison cells not settled. OpenAI's Decisions API retention and ZDR wording, Cloudflare's Clef data-usage cell and SubQ Snap's retention wording keep their October dates.
  4. New releases not read. ohmyjev 0.14.0 and langchain-typesafe 0.0.2 (both may change a failure path), pydantic-ai-slim 2.66.0 and Laya v0.4.28.
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube and TikTok were not covered for 2–7 Nov: unknown, not zero.

Evidence gaps after the 2 November research

  1. OpenCode's free ID. Which TypeSafe model or snapshot jev-1.13 and jev-1.13-free serve, how long the free ID stays free, and whether a free-model request counts as an unpaid account under OpenCode's Terms: OpenCode does not say. The route was not used or tested.
  2. ohmyjev outside its gates. Its screens, routing and done-check handlers were not re-read at 0.13.2 and keep their earlier reading on coding agents.
  3. New releases not read. Laya v0.4.21, @effect/ai-typesafe 4.1.2 and auto-mode 3.1.6.
  4. Console. The console sign-in pages returned HTTP 403 on 2 Nov and could not be read; the access state rests on TypeSafe's other pages and the status page.
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok and GitHub code search were not covered for 31 Oct–2 Nov: unknown, not zero.

Evidence gaps after the 31 October research

  1. Not read again. Microsoft-Decision-1's catalog cells (text-only sentence, catalog context, licence text; the catalog page builds its text in the browser) and the graded-test cells keep their earlier dates; Eevo's errors and authentication pages keep 19 Oct.
  2. OpenAI image limit. The up-to-128-images figure read on 13 Oct was not found in OpenAI's guide, API reference or changelog on 30 or 31 Oct. That is an absence in those pages, not a statement that the limit was removed.
  3. New releases not read. Kev kev-1.0 (1 Oct) and Laya v0.4.18 (30 Oct); the alternatives rows describe the earlier versions.
  4. Clef hosts. The terms of OpenRouter's Clef hosts (Makora, DeepInfra, PrimeIntellect) were not read; Cloudflare's zero-data-retention wording covers Workers AI, not those hosts.
  5. Integrations. ohmyjev's failure-map cells were not re-read at 0.13.2; nace-sdk 0.6.2, @ai-sdk/typesafe-ai 3.0.31 and ai 7.0.159 were not re-read.
  6. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok and GitHub code search were not covered for 30–31 Oct: unknown, not zero.

Evidence gaps after the 30 October research

  1. Days without a check. No check took place between 24 Oct, 05:12 UTC and 30 Oct 2026, 05:14 UTC. Changes in those days are dated as ranges; the status page's history is the only source for incidents on those days.
  2. Not read again. OpenAI's "128 images per request" (not found in its guide or API reference text on 30 Oct; not confirmed either way), Microsoft-Decision-1's retention wording, the d1-ultra line and the Drex row apart from its prices, and the Azure AI Foundry catalog page (not readable on 30 Oct; last read 16 Oct).
  3. SubQ Snap price. SubQ's docs say $0.04 per million input tokens and its 12 Oct launch post says $0.05; the two have not been reconciled.
  4. Integrations. ohmyjev's failure-map cells were not re-read at 0.13.2; @effect/ai-typesafe 4.1.0, pydantic-ai-slim 2.63.0 and nace-sdk 0.6.0 were not re-read; Claude Code's handling of hook answers is not public source.
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok and GitHub code search were not covered for 24–30 Oct, and 6 GitHub names on 28 Oct were beyond the search limit: unknown, not zero.

Evidence gaps after the 24 October research

  1. The alias move. Its exact time is unknown (between 23 Oct, 05:45 UTC and 24 Oct, 05:12 UTC); X could not be searched for an announcement. No route was called, so what Netlify's jev-latest, OpenRouter's ~typesafe/jev-latest, Vercel's typesafe-ai/jev or Cloudflare's typesafe/jev return now is unknown.
  2. Leaderboards. The Jev Decision Index 0.2 data, its private item counts, Artificial Analysis's display rounding and page history, JevMan's tie display and a JevBench commit pin were not read.
  3. New ledger rows. B45's per-call logs are not in its repository; B46's 360K figure was not matched line by line; B47 and B48 publish no materials.
  4. Catalogs. Amazon Bedrock, Azure AI Foundry and Workers AI catalogs were not re-read on 24 Oct.
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok and GitHub code search were not covered: unknown, not zero.

Evidence gaps after the 23 October research

  1. Images. No image was sent. Whether OpenRouter, Vercel, Opper, Netlify or Cloudflare forward or add TypeSafe's opt-in header is not known; how LiteLLM reports a 413, the exact JSON form of an image in langchain-typesafe, and whether a malformed assets object returns 400 or 422 were not settled.
  2. TypeSafe docs. The models and images pages changed size since 21 Oct; the changed lines are not identified (copies kept from 23 Oct onward). The models page's "Text" input and the images page's "both accept images" are not reconciled by TypeSafe.
  3. pydantic-ai-slim. Whether ModelConnectionError subclasses ModelAPIError was not checked.
  4. jev-gateway. Its merged image change (PR #67) is in no release; the failure-map row was read at 0.5.1 and not re-read.
  5. Scan coverage. Reddit returned HTTP 403 and DEV.to search no usable result; X, YouTube, TikTok and GitHub code search were not covered: unknown, not zero.

Evidence gaps after the 21 October research

  1. pi and omp. What the hosts do after ohmyjev's handler returns without blocking rests on their extension docs and ohmyjev's README; the hosts' own code after the return was not read.
  2. Not settled. auto-mode: the mod host's handling of its answer and the stages before Jev; jcode: callers outside memory, browser and voice. Five new leads that act on Jev's answer need a full read before they can be counted.
  3. Drex. Nace.AI's sub-processor page returned HTTP 403; the Fireworks AI terms were not read; which Nace.AI terms document controls is not stated.
  4. TypeSafe docs. The models page and quickstart changed size; earlier copies were not kept, so the changed lines are not identified (access, price, alias and limit text unchanged).
  5. Scan coverage. Reddit and DEV.to search returned HTTP 403; X, YouTube, TikTok and GitHub code search were not covered: unknown, not zero.

Evidence gaps after the 20 October research

  1. Self-check scope. The map does not record whether a fallback is visible, and timeouts and logging are not tallied across the 80; these are reminders without counts. "None found" bypass cells mean none found in the files read.
  2. Drex. Nace.AI's website Terms and Privacy pages and its Console data processing addendum were opened, not read; the terms of the provider behind the OpenRouter route, the Drex Max launch post and the base-model licences were not read; no Drex model was called.
  3. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok and GitHub code search were not covered: unknown, not zero.

Evidence gaps after the 19 October research

  1. Nothing was run. No package, benchmark, workflow, model or API call; no account or key; no hosted decision model was called.
  2. Claude Code guardrails. ohmyjev's built Codex hook was checked at its gate lines (its source was read in full). Its pi and omp handlers that make their own Jev decisions outside the tool gate (routing, done-check, output screen, compaction) are not rows; each returns no action when Jev fails. The jkudish/jev-mcp no-key error is raised inside a dependency that was not read; the published cell is unchanged.
  3. Hosted decision models. Liquid AI's documented errors and rate limits were not found; whether Cloudflare lists d1-ultra was not checked; Eevo's hosting region was not found and its legal pages carry no date. The vendors' claims against Jev, and their compatibility with TypeSafe's SDKs, are untested.
  4. Products. The two VentureBeat-relayed claims were not rechecked (HTTP 429; no archived copy). The publication date of Vercel's production-index post was not read on the page; its figures cover paid AI Gateway teams only.
  5. Precision. veermshah/jev-support-desk's 10 s / 30 s timeout values rest on SDK defaults that were not read in this update.
  6. Scan coverage. Reddit returned HTTP 403; X has no search access; YouTube, TikTok, GitHub trending and GitHub code search were not covered. Coverage is unknown there, not zero. Analytics still unavailable.

Evidence gaps after the 18 October research

  1. Nothing was run. No package, benchmark, workflow, image or API call; no account or key.
  2. Claude Code guardrails. ohmyjev's pi and omp adapters were read at their tool-call gates only; their other handlers were not read line by line, so ohmyjev is not yet in the map count. claude-jev-advisor's command-line, status-line and context files were checked by a scripted scan only (not tool-call paths).
  3. Products. jev-content-mod's HTTP request is made inside the third-party jevlang package, which was not read. The two VentureBeat-relayed claims were not rechecked; newer X statements cannot be checked. A DEV.to adoption figure ("35% of paid-for groups in Vercel") was not opened: Unverified.
  4. Failure terms. Wording-only cells on content moderation (11) and support-ticket triage (9) are left for a later update; their outcome words match the map.
  5. Routes and providers. Vercel does not define its 32,000 context field. The status page gives no cause for its uptime change. Liquid AI d1-ultra, Mouse-1, the tazo labs audio model and Eevo: launch pages only; terms, data handling and API shape not read.
  6. Scan coverage. Reddit returned HTTP 403; X has no search access; YouTube, TikTok, GitHub trending and GitHub code search were not covered. GitHub result lists over 100 items cover the window only partly, and identical repositories created under many accounts inflate this window's counts. Coverage is unknown there, not zero. Analytics still unavailable.

Evidence gaps after the 16 October research

  1. Nothing was run. No package, benchmark, workflow, image or API call; no account or key.
  2. When Jev fails. Every failure cell is recorded for all 77; four bypass cells rest on a named, narrower read scope, and for the ilyamk Gmail labeler a few helper functions on the way to the Jev call were not read in full. Three new Claude Code gates (read at pinned commits) and mcowdery/claude-model-guard are not yet on the map.
  3. SDKs. The error path of @typesafe-ai/sdk 0.7.0 was not re-read.
  4. Jev 1.14. Vercel AI Gateway's context field (32,000) is not reconciled with TypeSafe's 64k default. When the models page dropped the rate-limit numbers is known only as between 14 Oct 2026, 16:46 UTC and 16 Oct 2026, 05:21 UTC. Image data terms are not stated on the images page.
  5. Hosted decision APIs. Eevo: terms and pricing not read. A "Pollinations decisions API" named in third-party repositories is unverified. Subprocessors were not rechecked (the 11 Oct value stands).
  6. Scan coverage. Reddit returned HTTP 403; X search was unavailable; YouTube, TikTok, GitHub trending and GitHub releases were not covered: coverage is unknown there, not zero. Analytics still unavailable.

Evidence gaps after the 14 October research

  1. Nothing was run. No package, benchmark, workflow or API call; no account or key.
  2. When Jev fails. Bypass is not recorded for 11 of 77 (judge tools 4, frameworks 3, browser and computer use 3, email 1). For n8n, whether the editor blocks activating a workflow without a credential, and the case of a credential with an empty key, were not read.
  3. SDKs. The full option shape of experimental_decide, the deprecation warning text and the Cloudflare Workers binding row were not read. The vendor changelog has no 0.7.4 entry.
  4. Hosted decision APIs. SubQ Snap: numeric rate limits not found in its docs or Terms (14 Oct 2026, 16:48 UTC); its two retention texts disagree; whether Jev's SDKs work against it by changing the base URL was not tested. OpenRouter's status value −2 for openai/gpt-6-luna-decisions has no documented meaning. H2O-Lightning-4B, Intern-Decision-4B, Cygnet and GLiNER2.5-Decide: no primary source read. Subprocessors were not rechecked (the 11 Oct value stands).
  5. Scan coverage. X search was unavailable; Reddit returned HTTP 403; YouTube, TikTok, GitHub releases and trending were not covered: coverage is unknown there, not zero. Analytics still unavailable.

Evidence gaps after the 13 October run

  1. Nothing was run. No benchmark, harness or package was run; no Jev, OpenAI, Microsoft, Cloudflare or Subquadratic call. The pool has no account with any of them.
  2. Hosted decision APIs. Not found on the pages read (13 Oct 2026, 05:18–05:22 UTC): Microsoft-Decision-1's retention period in days, its data zones, a tokens-per-minute limit and a Microsoft price page; OpenAI /v1/decisions rate limits, context window, error codes, timeouts and retries, and any page dates; Clef's own rate limit. Why Foundry lists Decision-1 as Preview after listing it GA is unknown. SubQ Snap's docs, request shape and data terms were not read.
  3. When Jev fails. The ticket-triage bypass cells (5) were not read. Cells not re-checked this run: LiteLLM, JevRouter error, malformed and below threshold; soter malformed; jev_antispam_bot error and malformed. Router-wide LiteLLM hooks outside the complexity router were not read.
  4. Benchmarks and alternatives. B42's answer key is withheld, so its scores cannot be rechecked. About 10 new comparison repositories are leads without pins. Cygnet and GLiNER2.5-Decide have no primary source read. Subprocessors were not rechecked (the 11 Oct value stands).
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok, GitHub releases and trending were not covered: coverage is unknown there, not zero. The Hacker News "jev" search is typo-tolerant (4,601 hits); only the newest 100 were inspected. Analytics still unavailable.

Evidence gaps after the 12 October run

  1. Nothing was run. No harness, game, simulation, benchmark or package was installed or run; no Jev, Microsoft or OpenAI call.
  2. Real-time control. The cells come from code at pinned commits, not from runs. Not read: llm-invaders engine.py and its results files; JevPlaysSnakeGame snake.py. jev-fighter and ROS-JEV-LLM were only screened. Today's harness leads (jevGameAgent, moq-games) were not read. Which harness the ThursdAI Doom segment shows is unknown (its description links no code, 12 Oct 2026, 05:14 UTC).
  3. Microsoft-Decision-1. Not found or not read: the retention period; the Product Terms; the licence text; which regions offer DataZone deployment; rate limits; whether the data page covers the systemone operation. Foundry's catalog search runs in the browser and was not observed; Jev's absence rests on the model-page lookup and the served lists.
  4. Benchmarks and alternatives. etsabary/luna-decisions-benchmark is research for the next run, not published. 11 new comparison repositories are leads, not read. The Decisions API's rate limits, error behaviour and page dates are not documented on the pages read. Clef base-model licences were not read. Subprocessors were not rechecked (the 11 Oct value stands).
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok, GitHub releases and trending were not covered: coverage of the window is unknown there, not zero. "jev vision", "jev vlm" and "jev vla" remain unexplained by any TypeSafe source.

Evidence gaps after the 11 October run

  1. Nothing was run. No benchmark, plugin, harness, SDK or package was installed or run; no Jev, Decisions API or Microsoft call.
  2. Benchmarks. Where B39's significance test is computed (not in the files read); B38's full call logs (only the first 100 are committed); classifier-bakeoff's metric code; decisionbench's results location; the 10 new comparison repositories (README or description only).
  3. Coding agents. OpenClaw 2026.10.1 past the line read; hermes-jev-mana's key handling and the Hermes version its patch targets; the jev-for-all Claude Code hook and OpenCode's handling of removed tools; the Claude Code engine, which decides Hoppin-Sorter/jev-router (not public source).
  4. Alternatives and access. Microsoft-Decision-1's data terms and its OpenRouter listing; the Enterprise page's on-premises answer (not readable on 11 Oct, 05:18 UTC); what api.us.typesafe.ai and api.us-west-2.typesafe.ai are for (not in TypeSafe's docs); legal "Last updated" dates not re-extracted.
  5. Scan coverage. Reddit returned HTTP 403; X, YouTube, TikTok and GitHub trending were not covered: coverage of the window is unknown there, not zero. "jev vision", "jev vlm" and "jev vla" remain unexplained.

Evidence gaps after the 10 October run

  1. Nothing was run. No model, runtime, plugin, proxy or benchmark was installed or run; no Jev or Decisions API call.
  2. Unread host and plugin paths. Whether Hermes needs a separate enable step for jev-effort-router; how Codex and OpenCode handle jev-gateway's forced tool calls; OpenClaw's handling of {handled: true} (jev-triage); the hermes-jev-mana core patch in full.
  3. Benchmarks. The metric code of jev-rerank-bench, openai-decisions-vs-jev-vs-laya and jev-vs-decisions-rpg (grading moves to the next run); the classifier-bakeoff report; decisionbench's results.
  4. Alternatives. The Enterprise page's answer to "Can we run Jev in our own VPC or on-premises?" (not in the served page); TOD's base licence and hosted API shape; eleven new alternatives seen but not checked.
  5. Leads. 8 candidate coding-agent gates read from their README only; 20 coding-agent items from this scan not triaged; the pydantic-ai-slim 2.55.0 diff; what the new us-west-2 API endpoint means.
  6. Scan coverage. Reddit returned HTTP 403; X, YouTube and TikTok were not searched: coverage of the window is unknown there, not zero.

Evidence gaps after the 8 October run

  1. Nothing was run. No package, extension, reranker or benchmark was installed or run; no Jev call.
  2. Cells not recorded. LanceDB's behaviour with no key and the TypeSafe SDK's default timeout; what PostgreSQL returns after pg-jev's uncaught ValueError and SQLite's json_extract NULL path (engine behaviour); bypass for 33 of 69 and no key for 18 of 69 on when Jev fails.
  3. Hermes. The default model.provider and _BILLING_PATTERNS, which decide when the jev-approvals fallback applies.
  4. Reranking studies. jev-pulse is not graded; anessbelbati/jev-rerank-bench's code was not opened; neither operator states a stake. MindStudio's 28 Sep post was not re-read. The Jack & Jill case study publishes no data, so its 150-role test cannot be checked.
  5. Not done. Real-time-control harness code; coding-agent and MCP lead triage; 4 unread Hermes catalog plugins; six further SQL candidates listed but not audited (pg_jevplanner could be counted after a full read).
  6. Scan coverage. Analytics still unavailable; Reddit (403) and X (not attempted) unknown.

Evidence gaps after the 7 October run

  1. Nothing was run. No eval, judge, agent, hook, plugin or benchmark was installed or run. The four judge rows were mapped from the 4 Oct reads at the 4 Oct pins, not re-read.
  2. Cells not recorded. openlayer jevals' default no-key branch (one file); jev-curator's malformed, below-threshold and no-key cells and its cache lifetime; the step in which Hermes core hands the question to the plugin; the osuki 0.2.8 tarball against its tag.
  3. Benchmarks. B28: n per set and code; B35: code; B36: cohort list and judge.py; LanceDB: benchmark script; the Hugging Face and LanceDB post dates.
  4. Drift diffs due. oh-my-claudecode v5.6.2 and @jkudish/jev-mcp 0.14.1 (not read; the helper was interrupted).
  5. Leads not read. About 15 new coding-agent and MCP leads; 7 more Hermes catalog jev-* plugins; OpenAI's Decisions API (an alternatives lead); self-run eval repositories; real-time-control harness code.
  6. Scan coverage. Analytics still unavailable; Reddit (403) and X (not attempted) unknown.

Evidence gaps after the 6 October run

  1. Nothing was run. No coding agent, plugin, hook, skill or MCP server was installed or started; two npm packages were downloaded and read, not executed. Every failure cell is read from source at a pinned commit.
  2. Host behaviour after a hook decision. Codex, Cursor, Hermes core and OpenClaw core were not read, so what each host does with a hook's answer (or an exit code) is not recorded. OpenCode's default permission configuration, rh-guard's default mode, thresholds and redaction, how Hermes core handles jev-approvals errors, and jev-curator's plan path are not recorded.
  3. Versions not read. @osuki-dev/opencode-osuki-agent 0.2.8 (5 Oct) and whether the 0.2.7 tarball matches its tag; test-file contents; commit dates for four MCP repositories.
  4. Model IDs. What each route actually serves (no request was sent); whether OpenRouter accepts patch-level IDs; route changelogs (only TypeSafe's was checked, HTTP 404); SDK default models beyond TypeSafe's statement.
  5. Leads not read. About 20 coding-agent leads, including 7 more Hermes catalog jev-* plugins and 0xNatoshi/jev-codex-router (a model router); abaljeu/jev-mcp ("for thejevai.com", not a TypeSafe domain); the reranking-rag condition check; database and document error behaviour; the 6-byte change to TypeSafe's legal.md (legal "Last updated" dates unchanged).
  6. Scan coverage. Analytics still unavailable; Reddit (403) and X (login) unknown; the Internet Archive returned a connection error.

Evidence gaps after the 5 October run

  1. Nothing was run. No browser or computer-use agent, framework package or cloud console was run, installed or logged into; every failure cell is read from source at a pinned commit, and live behaviour may differ (for example the SDK defaults that LangChain and Pydantic AI inherit were not read).
  2. Unknown cells. Whether computer-use-jev sends [secure] field values; second-look's fallback card wording; trycua/cua's no-key behaviour; the Spring auto-configuration lines (read by a helper, not spot-checked). jev-agent-browser's npm package was not compared with its repository.
  3. Catalog boundaries. AWS Marketplace and Azure Marketplace results were read on the first page only; private offers and signed-in listings were not checked.
  4. Leads not read. LiteLLM 1.104.0 (a TypeSafe proxy guardrail hook; targeted diff due), more same-name browser repositories, the error behaviour of five database and document projects, obekt-terminal, the WSJ copycats story (paywalled). Coding-agent facts (Hermes, Codex, OpenCode, Cursor) are web-search snippets: Unverified until read.
  5. Analytics and scan coverage. Analytics still unavailable; Reddit (403) and X (login) unknown; TikTok, YouTube search, Lobsters, Product Hunt and Discord not attempted.

Evidence gaps after the 4 October run

  1. Not recorded cells. Across the 39 implementations on when Jev fails: no-key behaviour not recorded for 17, bypass for 22 (mostly pages published before the bypass column), below threshold for 5 (n8n nodes).
  2. Trading. No live order, fill, return or payment by any project was observed. jev-trader's no-key path runs inside @ai-sdk/typesafe-ai (not read). Whether QuantDinger's worker upgrade of a queued signal order to live skips the filter was not traced end to end; its LLM fallback prompt was not read.
  3. Judges. did-they-answer's abort on a malformed answer was traced in source but not run; competitor-hunter loads whatever Jev caller a user has installed (not pinned); openlayer jevals' PyPI version was not checked; the Jevals harness code is still not public.
  4. Data handling. TypeSafe's SOC 2 report is behind "Request access" and was not seen; what Slack and Google Workspace process for TypeSafe is not recorded.
  5. Leads not read. Three more jev-trader repositories, caiovicentino/jev-risk-check-provider and other fraud classifiers; TradeRank not found; JevTRPG, the "Judge Jev" post, DeepEval's ConversationalJevEval; jev-safety-eval, jev-zh-tw-eval and system-one-security (for a later benchmarks refresh).
  6. Carried. The two VentureBeat-relayed Products claims (HTTP 429 again); the two guard rows read in spare time wait for the next Claude Code and MCP edit.
  7. Analytics and scan coverage. Analytics still unavailable; Reddit (403) and X (login) unknown; YouTube, TikTok and GitHub code search not attempted.

Evidence gaps after the 3 October run

  1. Rate-limit timing. The change from 40 to 80 requests per second is known only to a 24-hour bracket (2 Oct, 05:14 UTC to 3 Oct, 05:16 UTC); there is no archive capture and the docs changelog returns HTTP 404.
  2. Email. ilyamk's granted OAuth scope (no manifest in the repository); spam-leadgen's no-key behaviour and model (inside the SDK); three email projects set no HTTP timeout; whether question text is billed and the tokens-per-character ratio are assumptions in the cost table. withoneai/mastra-inbound-router, anshulkhare7/jev-gmail-triage and the n8n Gmail-leads workflow from the Nyarko video were not read.
  3. n8n. How n8n core handles the node's error-output setting is known only from the official node's code comment; 10 of the 15 Jev/TypeSafe n8n packages found on npm were not read; the "declined verification" claim is Reported only.
  4. Products. The two VentureBeat-relayed claims were not rechecked; grade-A repository status was checked through HTML pages, not the API.
  5. Helper reads. Two helper agents read source in parallel; their key lines were spot-checked against the raw files, not re-read line by line. Unchecked cells keep their path and line references.
  6. Leads not read. Red Hat benchmark, Synthpop calibration, a Substack calibration post, Vercel's "Jev for Python Engineers", madisonrickert/jev-permission-gate.
  7. Carried. TypeSafe's trust page (needs a browser), Netlify data terms. jev.bevel.software ("EU/Jev") now answered from its own pages: a Jev-compatible API serving another model, not a Jev reseller (Reported, the operator's own claims).
  8. Analytics and scan coverage. Analytics still unavailable; Reddit (403), X (login), TikTok and YouTube not read in the scan.

Evidence gaps after the 2 October run

  1. Jev Router. What OpenRouter's Jev Router does on a Jev error, timeout, malformed answer or long conversation, and how it bills the routed model, is not documented; only OpenRouter can answer.
  2. No live behaviour. None of the six routers or six moderation projects was run; moderation accuracy for Jev is not published.
  3. Version drift. oh-my-claudecode v5.6.0 (1 Oct) not yet checked; ai 7.0.127 not re-audited.
  4. Leads not read. kerpopule/hermes-jev-skills, nine routing repositories created on 1–2 Oct, Jevdit, the project behind Twittesia issue #261, and other moderation repositories.
  5. Carried. TypeSafe's trust page (needs a browser), Netlify data terms, whether jev.bevel.software resells TypeSafe's Jev.
  6. Analytics and scan coverage. Analytics still unavailable; Reddit, X and TikTok blocked; GitHub code search not run.

Evidence gaps after the 1 October run

  1. Analytics still unavailable. No search impressions, queries or site analytics were available to choose or rank pages.
  2. Signup not observed. The console still returns HTTP 403; X cannot be searched for newer TypeSafe posts.
  3. Data handling. TypeSafe's subprocessor list and any SOC 2 report sit behind a script-only trust page; whether DigitalOcean on Vercel forwards to TypeSafe; whether OpenRouter's ZDR listing rests on a special agreement; a TypeSafe retention period, human access and a Telemetry opt-out are not stated; Netlify not read.
  4. Alternatives. Which Kev checkpoints JevBench ran; Kev-9B's base licence; the JevAny/AnyJev identity; who operates he-jev; the "jev vs laya" videos (titles only).
  5. Ticket triage. Live operation of the five projects; the JS SDK's default timeout; 70 or more triage repositories not read.
  6. Version drift. ai 7.0.126 and @jkudish/jev-mcp 0.12.0 not re-audited.
  7. Scan coverage. Reddit, X and TikTok blocked; GitHub code search and trending not run.

Evidence gaps after the 30 September run

  1. Signup not observed. The console returned HTTP 403; there is no second first-hand signal after 28 Sep; X cannot be searched.
  2. Rate-limit change time unknown. No docs changelog records when the limits changed.
  3. Provisional grades. The linked code of B13, B14, B15 and B24, and JevBench's code, were not opened; B08's reproducibility package was not located.
  4. Unidentified or unextracted results. MindStudio's primary study (B10); B14's Jina reranker row; LanceDB's results.
  5. Missing comparisons. No graded head-to-head with a fine-tuned classifier or zero-shot NLI with published materials; no graded non-English study except Bryo (grade C).
  6. Unchecked items. OpenAI's reported decision API at its own source; who operates LocalLLaMA/typed-decisions; jevcal's terms claim; the publisher of the npm package n8n-nodes-typesafe-ai; Vercel's provider-terms pass-through.
  7. Scan gap. 13 h 40 min between the 29 Sep window and the 30 Sep window were not scanned on GitHub, npm, Hugging Face or DEV.to; Reddit, X, TikTok, Lobsters and Product Hunt stayed blocked.

Evidence gaps after the 29 September run

  1. No live behaviour. No integration was run against the real API. The danna-zhou hook's 401/403 path is read from code; whether TypeSafe rejects model: "kev-latest" is unknown.
  2. Unread dependencies and files. jkudish/jev-mcp's TypeSafe-direct transport (@jkudish/jev-agent-tools), jev-claude's rule list and Codex adapter, and oh-my-claudecode's per-point hook events were not read. The danna-zhou MCP server's behaviour on a hanging endpoint was not tested.
  3. SDKs not executed. The official JS SDK was not installed (the Node test mirrors its defaults in pool code); the Python SDK install is still not done. The AI SDK's default timeout is undocumented, and its first version with experimental_evaluate conflicts (7.0.103 by package contents, 7.0.105 in Vercel's guide).
  4. Access. The console still returns 403; the credit status rests on the 27 Sep CEO post and the absence of later statements.
  5. Scan coverage. Reddit, X, TikTok and Lobsters were blocked; YouTube, GitHub trending and GitHub code search were not attempted; new hook and MCP repositories are leads only.

Evidence gaps after release 1

  1. Account facts. Whether a new account gets any credit today, live quotas and an actual bill are unknown; the console returned HTTP 403 and the pool does not create accounts. No primary TypeSafe source gives the "$5 / 120M tokens" amount.
  2. No live Jev behaviour observed by the pool. Status codes, headers and rate-limit scope are reported by third parties; the SDK was not executed; model="jev-1.13" was not tested.
  3. Confidence. The Score confidence formula for 4 or more levels is unknown; the independent calibration tests each cover one dataset and one model version.
  4. Billing. Whether question and criteria text and failed requests are billed is undocumented; how OpenRouter's Jev Router bills the routed model, and what it does on failure, is undocumented.
  5. Data terms. OpenRouter's retention and logging for Jev, the text of TypeSafe's MCA, privacy policy and DPA, and the billing entity for direct accounts were not reviewed (planned for a later release).
  6. Resellers. jev-agent.com was not checked; jev-ai.pro's monthly prices were not visible; flagged lookalike repositories and Spaces were not opened.
  7. Production claims. TechCrunch's Vercel article was not checked; there is no primary source for a Notra claim; FundaAI's method is paywalled; VentureBeat was read only through a summarising tool.
  8. Videos. No frames or transcripts were viewed; several videos do not state the route or SDK; whether one trading video used a real key is unknown.
  9. Alternatives. Ollaya's per-model licences and benchmark origin, Laya's System One compatibility, Kev's base-model licences, and whether AnyJev is the catalogue's JevAny are unchecked; six new leads are unverified.
  10. Scan coverage. Reddit, X, TikTok, Lobsters and GitHub code search could not be read; DEV.to coverage is complete only for the jev tag; npm and Hugging Face counts are capped.
  11. Benchmarking terms. A tool author's claim that TypeSafe's agreement restricts publishing benchmark results was not checked.

Standing 24-hour discovery scan

Every research update checks what appeared since the previous scan's cutoff (normally 24 hours); each window starts at the previous cutoff, so there is no gap and no overlap. This update's scan covered 8 Nov 05:14 UTC to 9 Nov 11:05 UTC (29 h 51 min). The main checks ran at 11:07–11:33 UTC on 9 Nov 2026.

SurfaceOutcome in the window (8–9 Nov)
Hacker News, GitHub, npm, PyPI, Hugging Face, DEV.toNo item qualified as a notice. Five benchmark repositories opened at pins (three graded on benchmarks); newer benchmark and coding-agent repositories listed as leads, not opened. Releases: ai 7.0.166, @openclaw/typesafe 2026.11.2, nace-sdk 0.7.0, n8n 2.47.3 and 2.46.7. DEV.to's latest feed covered only part of 9 Nov; earlier DEV.to coverage is by tag only
TypeSafeNo new blog post; models, Images, quickstart, legal and changelog pages and the docs index identical to the 8 Nov copies; no status event after 30 Oct (9 Nov 2026, 11:07 UTC; R29-S03 to R29-S15)
Hosted decision APIs and routesOpenRouter lists 461 model IDs, with no :free Jev ID; OpenCode's free Jev ID unchanged, its Terms re-dated 8 Nov; Nace.AI's legal pages re-dated 6 Nov; no new hosted decision API row (R29-S22 to R29-S36)
Not coveredReddit, X, YouTube and TikTok: coverage of 8–9 Nov unknown, not zero

Scan of 8 November 2026

The 8 November scan covered 7 Nov 15:27 UTC to 8 Nov 05:14 UTC (13 h 47 min); the main checks ran at 05:17–05:33 UTC on 8 Nov 2026.

SurfaceOutcome in the window (7–8 Nov)
Hacker News, GitHub, npm, PyPI, Hugging Face, DEV.toNo item qualified as a notice. New coding-agent projects are leads on coding agents; other new Jev-named repositories, packages and models are leads only. Index counts, not adoption (R28-S100 to R28-S117)
TypeSafeNo new blog post; models, Images, quickstart, legal and changelog pages identical to the 7 Nov copies; one new cookbook in the docs index; no status event after 30 Oct (8 Nov 2026, 05:17 UTC; R28-S03 to R28-S16)
Hosted decision APIs and routesOpenRouter lists 461 model IDs (460 on 7 Nov), with no :free Jev ID; OpenCode's free Jev ID unchanged; no new hosted decision API row (R28-S37, R28-S40, R28-S117)
Not coveredReddit, X, YouTube and TikTok: coverage of 7–8 Nov unknown, not zero (R28-S118 to R28-S121)

Scan of 7 November 2026

The scan of 7 Nov covered 2 Nov 10:46 UTC to 7 Nov 15:27 UTC (124 h 41 min), because there was no check between 2 Nov and 7 Nov; hosted decision APIs were searched over the whole window. The main checks ran at 15:29–15:48 UTC on 7 Nov 2026.

SurfaceOutcome in the window (2–7 Nov)
Hacker News"jev": 75 hits in 42 stories; "typesafe" 6; "System One" 11; "decision model" 24; "decisions API" 2; "decision API" 1; "Decision-1" 3; "Clef" 6; "elephant-1" 1. Zero for "jev-latest", "jev 1.13", "jev 1.14", "jev 1.15", a Jev deprecation, "jev free", "ohmyjev", "vs Jev", "joey-1" and "Base Labs". No item was a notice candidate (R27-S100)
GitHub, npm, Hugging Face, DEV.toRepositories matching "jev" created in the window: 286 in eight slices (21, 28, 32, 29, 34, 38, 62 and 42); "typesafe" 19 and 41; none at the search limit. Copies and unrelated names inflate these counts. npm: new versions of existing packages. Hugging Face: Jev-named models are third-party, none published by TypeSafe. DEV.to: the decision-model tag returned HTTP 429. Index counts, not adoption (R27-S103 to R27-S114)
TypeSafeOne new blog post (3 Nov, about naming); models page, quickstart and a new changelog page changed as described in the 7 Nov report; no status event listed for 31 Oct–7 Nov (7 Nov 2026, 15:29 UTC; R27-S03 to R27-S11)
Hosted decision APIs and routesBase Labs now has a row on alternatives. Other new hosted decision-API pages are leads only (Hume's audio decisions API among them; not rows). OpenRouter lists 460 model IDs (452 on 2 Nov), with no :free Jev ID; OpenCode's free Jev ID unchanged (R27-S19, R27-S37, R27-S102)
Not coveredReddit (HTTP 403), X (script-only page), YouTube and TikTok: coverage of 2–7 Nov unknown, not zero

Scan of 2 November 2026

The scan of 2 Nov covered 31 Oct 05:09 UTC to 2 Nov 10:46 UTC (53 h 37 min), with a 72-hour look-back for hosted decision APIs from 30 Oct 10:46 UTC; the main checks ran at 10:47–11:01 UTC on 2 Nov 2026.

SurfaceOutcome in the window (31 Oct–2 Nov)
Hacker News"jev": 13 items in the window (4 about Jev; 6 are "Jevons"); "typesafe" 1 (not about TypeSafe); "System One", "jev-latest", "jev 1.13", "jev 1.14", "jev 1.15", a Jev deprecation, "jev free" and "ohmyjev" 0; "opencode" 21, none about Jev. No item was a notice candidate. Look-back: no new hosted decision API from another provider (R26-S100, R26-S101)
GitHub, npm, Hugging Face, DEV.toRepositories matching "jev" created in the window: 35, 24, 34 and 30 in four slices; "typesafe" 15 and 10; none at the search limit. Copies and unrelated names inflate these counts. npm: new versions of existing packages only. Hugging Face: 17 Jev-named models created in the window, none published by TypeSafe. DEV.to: 0 posts. Index counts, not adoption (R26-S108 to R26-S121)
TypeSafeNo new post since 14 Oct; models, images and quickstart pages identical to the 31 Oct copies; no status event listed for 31 Oct–2 Nov (2 Nov 2026, 10:47 UTC; R26-S03 to R26-S08)
Hosted decision APIs and routesNo new hosted decision API found. OpenRouter lists 452 model IDs (456 on 31 Oct), with no :free Jev ID. OpenCode's free Jev ID is described on where to use (R26-S122, R26-S200)
Not coveredReddit (HTTP 403), X (login wall), YouTube (relative dates only), TikTok (rendered in the browser only) and GitHub code search: coverage of 31 Oct–2 Nov unknown, not zero

Scan of 31 October 2026

That scan covered 30 Oct 05:13 UTC to 31 Oct 05:09 UTC (23 h 56 min), with a 72-hour look-back for hosted decision APIs from 28 Oct 05:09 UTC. The main checks ran at 05:11–05:28 UTC on 31 Oct 2026.

SurfaceOutcome in the window (30–31 Oct)
Hacker News"jev": 6 items in the window (5 about Jev); "typesafe" 1; "System One" 2; "jev-latest", "jev 1.13", "jev 1.14", "jev 1.15" and a Jev deprecation 0. No item was a notice candidate. Look-back: no new hosted decision API from another provider; d1-ultra and Clef items already recorded on 30 Oct (R25-S100 onward)
GitHub, npm, Hugging Face, DEV.toRepositories matching "jev" created in the window: 32 and 32 in two 12-hour slices; "typesafe" 9; none at the search limit. Copies under many accounts inflate these counts. npm: new versions of existing packages only. DEV.to: 0 posts. Index counts, not adoption
TypeSafeNo new post since 14 Oct; models page identical to the 30 Oct copy; quickstart and images pages changed only in formatting markup; a resolved 30 Oct us-east-1 status event (31 Oct 2026, 05:11 UTC; R25-S03 to R25-S06, R25-S08)
Hosted decision APIsChanged cells for Microsoft-Decision-1, Cloudflare Clef, OpenAI and Eevo (report). OpenRouter lists 456 model IDs, as on 30 Oct
Not coveredReddit (HTTP 403), X (login wall), YouTube (relative dates only), TikTok (rendered in the browser only) and GitHub code search: coverage of 30–31 Oct unknown, not zero

Scan of 30 October 2026

That scan covered 24 Oct 05:11 UTC to 30 Oct 05:13 UTC (144 h 02 min, six days, searched day by day), because there was no check between 24 and 30 Oct; the 72-hour look-back for hosted decision APIs lay inside it. The main checks ran at 05:14–05:23 UTC on 30 Oct 2026.

SurfaceOutcome in the window (24–30 Oct)
Hacker News"jev": 83 items over the six days (6, 40, 16, 11, 4 and 6 by day; literal-string filter); "typesafe" 5; "jev-latest", "jev 1.13", "jev 1.14", "jev 1.15" and a Jev deprecation 0. No item was a notice candidate (R24-S100)
GitHub, npm, Hugging Face, DEV.toRepositories matching "jev" by creation day (UTC): 73, 69, 69, 82, 106 and 88; the 28 Oct search returned its limit of 100, so 6 names were not seen; "typesafe" 46 and 40 in two three-day slices. Copies under many accounts inflate these counts. npm: 41 and 33 packages for "jev" and "typesafe" (lower bounds). Hugging Face: model searches hit their limit of 100 (lower bounds); 7 Jev-named Spaces. DEV.to: 4 posts tagged jev. Index counts, not adoption. Coding-agent names are triaged on coding agents (R24-S124 to R24-S132)
TypeSafeNo new post since 14 Oct; models page: alias table unchanged, higher limits now requested through a form; images page identical to the 24 Oct copy (30 Oct 2026, 05:14 UTC; R24-S05, R24-S06, R24-S08)
Hosted decision APIsChanged cells for OpenAI, Microsoft-Decision-1 and SubQ Snap (report). OpenRouter lists 456 model IDs (469 on 24 Oct), none of them decision-named
Not coveredReddit (HTTP 403), X (login wall), YouTube (relative dates only), TikTok (rendered in the browser only) and GitHub code search: coverage of 24–30 Oct unknown, not zero

Scan of 24 October 2026

That scan covered 23 Oct 05:18 UTC to 24 Oct 05:11 UTC (23 h 53 min), with a 72-hour look-back for hosted decision APIs. The main checks ran at 05:12–05:19 UTC on 24 Oct 2026.

SurfaceOutcome in the window (24 Oct)
Hacker News"jev": 41 items in the window (literal-string filter); "typesafe" 2; "decision index" 3; "jev-latest" and "jev 1.14" 0. Items noted: an opinion piece on Jev's hype cycle (not a catalogue item); a coding-agent review-loop article (lead, not opened); OpenAI's general-availability post (not opened; its guide and changelog were read); a comment claiming Jev beats the Decisions API on the Artificial Analysis index (not checked) (R23-S34, R23-S50 to R23-S52)
GitHub, npm, Hugging Face, DEV.toAbout 91 repositories created in the window that match "jev" and 14 that match "typesafe"; copies under many accounts inflate these counts. Nine benchmark-named repositories are listed as leads, not opened. npm: 14 and 9 packages with new versions for "jev" and "typesafe". Hugging Face: 26 Jev-named and 11 decision-named models modified, including Jev-Vision quantizations (Jev-named; not published by TypeSafe). DEV.to: 0 posts. Index counts, not adoption (R23-S35 to R23-S38)
TypeSafeNo new post; models page: jev-latest now jev-1.14.0 and a new note that jev-preview points to the same model; images page changed only in link form (24 Oct 2026, 05:12 UTC; R23-S05, R23-S06)
Hosted decision-API look-back (72 h)Base Labs' mammoth-1 and elephant-1 (21 Oct; rank claims checked on benchmarks), OpenAI's general availability (22 Oct) and SubQ Snap 1.1 listed on the Artificial Analysis index. No other new hosted decision API from another provider was found.
Not coveredReddit (HTTP 403), X, YouTube, TikTok and GitHub code search: unknown, not zero

Scan of 23 October 2026

The 23 Oct scan covered 21 Oct 05:10 UTC to 23 Oct 05:18 UTC (48 h 08 min). The main checks ran at 05:18–05:30 UTC on 23 Oct 2026.

SurfaceOutcome in the window (23 Oct)
Hacker News"jev": 6 stories and 25 comments in the window (literal-string filter): commentary, demo and model posts, none opened beyond titles; "typesafe" 4; "jev images" 0 (R22-S34)
GitHub, npm, Hugging Face, DEV.to210 repositories created that match "jev" in the window; identical repositories under many accounts inflate this count. Image-related repositories seen by name (listed on limitations). npm: @ai-sdk/typesafe-ai 3.0.25 and 3.0.26, @tanstack/ai-typesafe 1.0.0 (not read), @jkudish/jev-mcp 0.14.5, auto-mode 3.0.1. Hugging Face: 39 Jev-named and 42 decision-named models modified (Jev-named; not published by TypeSafe). 1 DEV.to post. Index counts, not adoption (R22-S35 to R22-S38)
TypeSafeNo new post; models, images and Box pages and the quickstart changed size (changed lines not identified; access, price, alias, input and limit text unchanged); jev-latest still jev-1.13.0 (checked 23 Oct 2026, 05:19 and 05:45 UTC; R22-S03, R22-S05)
Hosted decision-API look-back (72 h)OpenAI's Decisions API became generally available on 22 Oct (comparison page updated; R22-S32, R22-S47); open-weight and local decision-model posts seen, not read; no Jev string on Foundry or Bedrock
"jev with images", "jev image support", "is jev multimodal", "jev vision"Suggested on Google, Bing and DuckDuckGo (for example "does jev work with images", "can jev classify images"); the new section's heading follows them. "jev vision" is still unexplained; Jev-named "Jev-Vision" models created on Hugging Face on 22 Oct are one possible source, not proven (R22-S40, R22-S51)
Reddit, X, YouTube, TikTok, GitHub trending and code searchReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero

Scan of 21 October 2026 (20 Oct 05:13 UTC to 21 Oct 05:10 UTC; main checks 05:11–05:18 UTC):

SurfaceOutcome in the window (21 Oct)
Hacker News"jev": 8 stories in the window (literal-string filter): commentary, usage and demo posts and a third-party comparison page, none opened; "typesafe": 2 (R21-S34)
GitHub, npm, Hugging Face, DEV.to121 repositories created that match "jev" in the window; identical repositories under many accounts inflate this count. New coding-agent leads (names only; next for coding agents). npm: @ai-sdk/typesafe-ai 3.0.24, TypeSafe's n8n node 0.10.0, auto-mode 3.0.0. Hugging Face: 18 Jev-named and 10 decision-named models modified (Jev-named; not published by TypeSafe). 1 DEV.to post; DEV.to search HTTP 403. Index counts, not adoption (R21-S35 to R21-S38)
TypeSafeNo new post; models page and quickstart changed size (changed lines not identified; access, price, alias and limit text unchanged); jev-latest still jev-1.13.0 (checked 21 Oct 2026, 05:11 UTC; R21-S03, R21-S05)
Hosted decision-API look-back (72 h)No new hosted decision-API launch found; no new decision-model ID on OpenRouter or Vercel; Cloudflare's catalog lists typesafe/jev-1.14.0 (a route listing); Clairic is an extraction model, not a decision API (R21-S22 to R21-S31, R21-S43)
"jev vision", "jev visual", "jev vlm", "jev vla", "nano jev"Still suggested; no TypeSafe release by those names; "nano jev" has a likely source in the Hugging Face model C-Tianyu/NanoJev (Jev-named; not published by TypeSafe; not shown to be the source); the others are still unexplained (R21-S40, R21-S44, R21-S45)
Reddit, X, YouTube, TikTok, GitHub trending and code searchReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero

Scan of 20 October 2026 (19 Oct 05:21 UTC to 20 Oct 05:13 UTC; main checks 05:13–05:24 UTC):

SurfaceOutcome in the window
Hacker News"jev": 4 stories in the window (literal-string filter), commentary and usage posts; "typesafe", "drex" and "nace.ai": 0 relevant (R20-S41)
GitHub, npm, Hugging Face, DEV.to184 repositories created that match "jev" in the window; identical repositories under many accounts inflate this count. npm: @ai-sdk/typesafe-ai 3.0.22 and 3.0.23, @tanstack/ai-typesafe 0.1.13, and nace-sdk 0.2.0 (Drex's SDK). Hugging Face: 37 Jev-named and 24 decision-named models modified (Jev-named; not published by TypeSafe). 3 DEV.to posts. Index counts, not adoption (R20-S43, R20-S44)
TypeSafeNo new post; the API reference split into separate pages (System One, List models, Request rules and errors) with an OpenAPI link; models and Images pages differ only in links; jev-latest still jev-1.13.0 (checked 20 Oct 2026, 05:13 UTC; R20-S05, R20-S20)
Hosted decision-API look-back (72 h)No new hosted decision API beyond Drex (read in full this update, alternatives); no new decision-model ID on OpenRouter or Vercel; OpenAI's Decisions API guide changed one line (image URLs accepted) (R20-S40)
"jev vision", "jev visual", "jev vlm", "jev vla"Still suggested by Bing, DuckDuckGo and Google; Hacker News 0; still unexplained; no TypeSafe release by those names
Reddit, X, YouTube, TikTok, GitHub trending and code searchReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero

Scan of 19 October 2026 (18 Oct 05:16 UTC to 19 Oct 05:21 UTC; main checks 05:22–05:28 UTC):

SurfaceOutcome in the window
Hacker News"jev": 3 stories in the window (literal-string filter); "typesafe": 0 relevant; "jev vision" and "jev vlm": 0. Leads: an open-weight 31B model, Basal-Omni, whose publisher calls it "nr 1 on JevBench" (Reported); two posts on decision models as judges; one on the economics of decision models (not opened) (R19-S45, R19-S41)
GitHub, npm, Hugging Face, DEV.to176 repositories created that match "jev" in the window; identical repositories under many accounts inflate this count. @openclaw/typesafe 2026.10.5 on npm. Hugging Face: Cribl decision 1.0 weights (27B and 9B, card licence apache-2.0) and many Jev-named or decision-named models by third parties (Jev-named; not published by TypeSafe). 4 DEV.to posts, including one comparing "Jev, Drex, and Laya" (Drex not identified; not opened). Index counts, not adoption (R19-S42, R19-S46 to R19-S49)
TypeSafeNo new post; models, images, Box, legal and docs-index pages unchanged since 16 Oct; jev-latest still jev-1.13.0 (checked 19 Oct 2026, 05:22 UTC; R19-S05, R19-S12, R19-S20)
"jev vision", "jev visual", "jev vlm", "jev vla"Still suggested by Bing, DuckDuckGo and Google; Hacker News 0. Two third-party projects use the name (yexerik/jevision, an image prototype; leinenterprises/jev-vision-27b). Whether the suggestions refer to TypeSafe's image input of 14 Oct is not established (R19-S44, R19-S52)
Reddit, X, YouTube, TikTok, GitHub trending and code searchReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero (R19-S51)

72-hour decision-API look-back for 19 October, from 16 Oct 2026, 05:21 UTC (it overlaps the 18 October look-back, as intended): no new hosted decision-model launch beyond Liquid AI d1-ultra, Mouse-1, Eevo and the tazo labs audio model, already recorded on 18 Oct. d1-ultra, Mouse-1 and Eevo were read in their providers' own pages and added to alternatives. OpenAI's Decisions API guide is unchanged ("public beta"; R19-S40). None is a home-page notice.

18 October scan

This scan covered 16 Oct 05:19 UTC to 18 Oct 05:16 UTC. The main checks ran at 05:21–05:23 UTC on 18 Oct 2026.

SurfaceOutcome in the window
Hacker NewsWith typo tolerance off: "jev" 5 stories and 23 comments; "typesafe" 2 stories and 0 comments; "jev 1.14", "jev-latest" and "claude code jev" 0 stories (R18-S41)
GitHub, npm, Hugging Face, DEV.to374 repositories created that match "jev" in the window and 47 that match "jev claude"; identical repositories under many accounts inflate these counts. 0 Hugging Face models under the typesafe-ai author; 5 DEV.to posts under the jev tag. Index counts, not adoption (R18-S42 to R18-S45)
TypeSafeNo new post; models, images, Box, legal and docs-index pages byte-identical to 16 Oct; jev-latest still jev-1.13.0 (checked 18 Oct 2026, 05:17–05:21 UTC; R18-S05, R18-S12, R18-S20)
"jev vision", "jev vlm", "jev vla"Bing now also suggests "jev visual"; DuckDuckGo shows none of the four; Hacker News 0 stories each. "jev vision" stays partly explained by TypeSafe's image input of 14 Oct; the others have no TypeSafe source (R18-S48)
Reddit, X, YouTube, TikTok, GitHub releases, trending and code searchReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero (R18-S46)

72-hour decision-API look-back for 18 October, from 15 Oct 2026, 05:16 UTC (it overlaps the 16 October look-back, as intended): Liquid AI d1-ultra (hosted; $0.042 per million input tokens on its launch page; Liquid says it outperforms Jev 1.14, a vendor claim, Reported) is new (R18-S50); Mouse-1 ($0.10 per million input tokens; post dated 13 Oct) is older but newly seen (R18-S51); a tazo labs audio decision model is new, its API, price and terms not read (R18-S52). Already recorded, no change found: the Decision Index, SubQ Snap, OpenAI's Decisions API (still public beta; R18-S55). Eevo's page was read ("around $0.20 per decision"); its terms and data handling were not found. None is a home-page notice; d1-ultra and Mouse-1 get one dated "not yet evaluated" line on alternatives.

16 October scan

This scan covered 14 Oct 16:45 UTC to 16 Oct 05:19 UTC. The main checks ran at 05:28 UTC on 16 Oct 2026.

SurfaceOutcome in the window
Hacker NewsWith typo tolerance off: "jev" 2 stories and 37 comments; "typesafe" 0 stories and 2 comments; "jev 1.14" 0 stories and 1 comment (R17-S50)
GitHub, npm, Hugging Face, DEV.to238 repositories created that match "jev" in the window, 23 that match "jev claude"; 0 Hugging Face models under the typesafe-ai author, Jev-named models from other accounts only; 1 DEV.to article under the jev tag. Index counts, not adoption. Leads, not read: Jev 1.14 image projects, comparison write-ups and coding-agent integrations; Claude Code and MCP items classified on Claude Code and MCP (R17-S51 to R17-S55)
TypeSafeJev 1.14 released under jev-preview (post dated 14 Oct), with opt-in experimental image input, Box questions and long context; rate-limit numbers removed from the models page. No legal, trust or zero-data-retention wording change found (checked 16 Oct 2026, 05:21–05:26 UTC; R17-S05, R17-S09, R17-S12, R17-S41)
"jev vision", "jev vlm", "jev vla"Partly explained by TypeSafe's experimental image input of 14 Oct; Google now also suggests "jev image support" and "jev image input"; "jev vlm" and "jev vla" still have no TypeSafe source (R17-S53)
Reddit, X, YouTube, TikTok, GitHub releases and trendingReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero (R17-S56)

72-hour decision-API look-back for 16 October, from 13 Oct 2026, 05:19 UTC (it overlaps the 14 October look-back, as intended): Eevo (Upright; launched 14 Oct 2026 as a "consequential decision model"; hosted) is new, seen, not checked (R17-S62); Jev 1.14 is TypeSafe's own release; SubQ Snap, OpenAI's Decisions API (public beta, price unchanged), Microsoft-Decision-1 (Preview, version 1, in Foundry) and the Cloudflare Clef family: no change found; Cribl decision 1.0 open weights recorded once (R17-S64); a "Pollinations decisions API" named in third-party repositories was not found on the Pollinations API docs page fetched (unverified; R17-S65). None is a home-page notice. Drift: see the release notes for 16 October.

14 October scan

This scan covered 13 Oct 2026, 05:17 UTC to 14 Oct 2026, 16:45 UTC. The main checks ran at 16:50:14–16:50:38 UTC.

SurfaceOutcome in the window
Hacker NewsWith typo tolerance off: "jev" 17 stories and 34 comments; "decisions api" 1; no TypeSafe launch (R16-S55)
GitHub, npm, Hugging Face, DEV.to220 repositories created that match "jev" in the window; new npm packages, none from TypeSafe; Hugging Face models from other accounts only, 0 under the typesafe-ai author; 2 DEV.to articles under the jev tag, no launch. Index counts, not adoption. Leads, not read: 8 comparison repositories and about 10 coding-agent integrations (R16-S56 to R16-S59)
TypeSafeNot found in the window (checked 14 Oct 2026, 16:46–16:50 UTC): a TypeSafe release or a model, alias, price, limit, legal, trust or zero-data-retention change. The status page lists two more regional API entries, with no access change (R16-S04)
"jev vision", "jev vlm", "jev vla"Still unexplained by any TypeSafe source; Bing and DuckDuckGo still suggest the phrases; 0 Hacker News results (R16-S61)
Reddit, X, YouTube, TikTok, GitHub releases and trendingReddit HTTP 403; X search unavailable; the others were not covered: unknown, not zero (R16-S60)

72-hour decision-API look-back for 14 October, from 11 Oct 2026, 16:45 UTC (it overlaps the 13 October look-back, as intended): SubQ Snap's docs, Terms and launch post read, and the service added as a row on alternatives (R16-S217 to R16-S227); the Artificial Analysis Decision Index v1.0 is new, a third-party index rather than a launch, and a watch item (R16-S62); SPEV, an open-weight decision model, is new and not a hosted launch; Microsoft-Decision-1 still listed as Preview in Foundry; OpenAI's Decisions API price unchanged, while OpenRouter's endpoint shows status −2, which has no documented meaning (R16-S32). None is a home-page notice. Drift: patches @jkudish/jev-mcp 0.14.3, @effect/ai-typesafe 4.0.4, @openclaw/typesafe 2026.10.3, litellm 1.105.1; minor pydantic-ai-slim 2.56.0 (R16-S64).

13 October scan

For the 13 October run the window was 12 Oct 2026, 05:17 UTC to 13 Oct 2026, 05:17 UTC. The 12 October cutoff was 05:12 UTC, so the 5 minutes between were backfilled on Hacker News and GitHub: 0 items in the result pages read; no overlap. The main checks ran at 05:21:28–05:21:33 UTC.

SurfaceOutcome in the window
Hacker News"jev" returned 4,601 typo-tolerant hits; the newest 100 contain no literal "jev"; "typesafe" 8 hits, unrelated; "decisions api" 0; "system one model" 68, no launch (R15-S25)
GitHub, npm, Hugging Face, DEV.to153 repositories created that match "jev" since 12 Oct, 05:12 UTC; new npm packages, none from TypeSafe; Hugging Face models from other accounts only (Jev-named; not published by TypeSafe), 0 under the typesafe-ai author; 5 DEV.to articles, no launch. Index counts, not adoption. Leads, not read: about 10 repositories that compare Jev with a substitute (benchmarks lead list) and 8 coding-agent integrations (R15-S28 to R15-S31)
TypeSafeThe docs quickstart was rewritten and llms.txt retitled "TypeSafe", with no access effect (R15-S04, R15-S07). Not found in the window (scan of 13 Oct 2026, 05:18–05:22 UTC): a TypeSafe release or a model, alias, price, limit, legal, trust or zero-data-retention change
"jev vision", "jev vlm", "jev vla"Still unexplained by any TypeSafe source; search engines suggest the phrases; 0 Hacker News results; 4 Jev-named vision models from other accounts on Hugging Face (R15-S27)
Reddit, X, YouTube, TikTok, GitHub releases and trendingReddit HTTP 403; X not accessible; the others were not searched (time): unknown, not zero (R15-S32)

72-hour decision-API look-back (standing addition, from 12 Oct)

Each run also looks back 72 hours for launches of hosted decision models or decision APIs by any provider, on timestamped surfaces and known provider pages. It is not a full 72-hour scan. For the 13 October run the look-back ran from 10 Oct 2026, 05:17 UTC to the cutoff; it overlaps the 12 October look-back by about 48 hours (Hacker News checked 13 Oct 2026, 05:21 UTC; R15-S26). From 13 Oct it also flags changes to the providers on the decision-API comparison page. None of the items below is a home-page notice: another provider's model does not change whether or how a new user can get Jev.

ItemStatusWhat was recorded
SubQ Snap (Subquadratic; post dated 12 Oct)New in the windowA hosted decision API. Vendor-documented: $0.05 per 1M input tokens, output free, Public Preview with a possible waitlist, text and images. Its claims against Jev are Reported only; its docs and data terms were not read (R15-S33). Named on the comparison page as not compared, and on alternatives as seen, not checked
Microsoft-Decision-1 (published 9 Oct)Recorded on 11 Oct; changedListed as Preview in Foundry at 13 Oct 2026, 05:20 UTC (listed GA on 12 Oct; cause unknown; R15-S43, R15-S44). Vercel's AI Gateway now lists it (R15-S212)
OpenAI Decisions API and Cloudflare Clef familyRecorded earlier; routes addedVercel lists the Decisions API (R15-S212); OpenRouter serves Clef through PrimeIntellect, DeepInfra and Cloudflare (R15-S226). No newer Clef changelog item found on Cloudflare's changelog (13 Oct 2026, 05:21 UTC)
Decision 3.0 (12 Oct) and JevBench (12 Oct)New in the window; not hosted launchesAn open-weight decision-model family, and a benchmark site listing 17 decision models (benchmarks lead list; R15-S34)
OpenRouter model listNo changeNo model ID with "decision" in its ID; 459 models (458 on 12 Oct) (13 Oct 2026, 05:18 UTC; R15-S40)

Drift: patch releases only (@ai-sdk/typesafe-ai 3.0.18, ai 7.0.138, @jkudish/jev-mcp 0.14.2, @jkudish/jev-agent-tools 0.2.1); everything else on the drift list unchanged since 12 Oct. Source R15-S52.

12 October scan (window ending 12 Oct 2026, 05:12 UTC)

For the 12 October run the window was 11 Oct 2026, 05:12 UTC to 12 Oct 2026, 05:12 UTC. It overlaps the 11 October window by 4 minutes, so there was no gap and no backfill. The main checks ran at 05:16:11–05:17:43 UTC.

SurfaceOutcome in the window
Hacker News"jev" returned 408 typo-tolerant hits with 0 stories about Jev; "typesafe" 8 hits, all about type-safe programming; "decisions api" 0; "system one model" 14, no launch (R14-S24)
GitHub, npm, Hugging Face, DEV.to163 repositories created that match "jev"; 2 npm packages published, none from TypeSafe; Hugging Face models from other accounts only (Jev-named; not published by TypeSafe); 2 DEV.to articles, no launch. Index counts, not adoption. Leads, not read: 2 real-time harnesses (jevGameAgent, moq-games), 11 repositories that compare Jev with a substitute, 5 coding-agent integrations (R14-S27 to R14-S29, R14-S32)
TypeSafeNot found in the window (scan of 12 Oct 2026, 05:12–05:17 UTC): a TypeSafe release or a model, alias, price, limit, legal, trust or zero-data-retention change
"jev vision", "jev vlm", "jev vla"Still unexplained by any TypeSafe source; search engines suggest the phrases; 0 Hacker News results; Jev-named vision models from other accounts on Hugging Face (R14-S26)
Reddit, X, YouTube, TikTok, GitHub releases and trendingReddit HTTP 403; X not accessible; the others were not searched (time): unknown, not zero (R14-S33)

72-hour decision-API look-back, 12 October run

Each run also looks back 72 hours for launches of hosted decision models or decision APIs by any provider, on timestamped surfaces and known provider pages. It is not a full 72-hour scan. For the 12 October run the look-back ran from 9 Oct 2026, 05:12 UTC to the cutoff (Hacker News checked 12 Oct 2026, 05:16 UTC; R14-S25). None of the items below is a home-page notice: another provider's model does not change whether or how a new user can get Jev (R14-S34, R14-S35).

ItemStatusWhat was recorded
Microsoft-Decision-1 (published 9 Oct)Recorded on 11 Oct; changedMicrosoft Foundry's catalog lists it (Direct from Azure, version 1, GA; 12 Oct 2026, 05:15 UTC; R14-S43, R14-S44). Microsoft's post now carries an editor's note that benchmarks for Jev were added; no update date is shown (Reported; R14-S207). See its row on alternatives
Cloudflare Clef-omni (Workers AI changelog dated 9 Oct)Older, newly recordedA hosted decision model in the open-weight Clef family that takes audio and video input with text and images (Documented, Cloudflare). Cloudflare also says Clef-flash "now costs less than Jev" (Reported, Cloudflare). Now on alternatives
THX-01 "Free API" (posts on 10 and 11 Oct)Recorded on 10 Oct; changedIts author announces a free hosted API (author's claim). The API link was not followed and no primary page was read
OpenDecision image support (11 Oct)Recorded on 10 Oct; changedA self-hosted decision API for images, not a hosted launch
OpenRouter model listNo changeNo new model ID with "decision" in its ID, name or description; 458 models, as on 11 Oct (12 Oct 2026, 05:12 UTC; R14-S40)

Drift: LiteLLM 1.105.0 checked (the Jev classifier's failure path is unchanged); @openclaw/typesafe 2026.10.2 released 12 Oct, 02:27 UTC (drift note only); the ra2web-jev-player pin moved to 0e4e8cf with the cited files byte-identical; everything else on the drift list unchanged since 11 Oct. Sources R14-S53, R14-S54.

11 October scan (window ending 11 Oct 2026, 05:16 UTC)

For the 11 October run the window was 10 Oct 2026, 05:16 UTC to 11 Oct 2026, 05:16 UTC, plus the 5 minutes since the 10 October cutoff (05:11 UTC), backfilled on Hacker News and GitHub with 0 items in those minutes; no overlap. The main checks ran at 05:17–05:22 UTC.

SurfaceOutcome in the window
Hacker News"typesafe": 10 hits (1 story, 9 comments); "decisions api": 1 comment; "system one model": 4 relevant comments; "jev" returned only typo matches. No statement restoring new-user credit
GitHub, npm, Hugging Face, DEV.to147 repositories created that match "jev"; 6 npm packages published (including @effect/ai-typesafe 4.0.3 and @openclaw/typesafe 2026.10.1); 37 of the 50 latest "jev" models on Hugging Face modified (Jev-named; not published by TypeSafe); 2 DEV.to articles. Index counts, not adoption. 10 new repositories compare Jev with a substitute (benchmarks leads)
Other providersMicrosoft-Decision-1 (published 9 Oct, first recorded here): a hosted decision model from another provider, now on alternatives
TypeSafeA new homepage build (11 Oct, 04:02 UTC) with no signup, credit or price text; a new status-page section listing api.us.typesafe.ai; no new blog post. No model, alias, price, limit, subprocessor or incident change
"jev vision", "jev vlm", "jev vla"Still unexplained by any TypeSafe source; search engines suggest the phrases; 0 Hacker News results; three Jev-named vision models from other accounts on Hugging Face
Reddit, X, YouTube, TikTokReddit HTTP 403; the others were not searched: unknown, not zero

Drift: pydantic-ai-slim checked at 2.55.0 (failure path unchanged); patch releases @effect/ai-typesafe 4.0.3 and @openclaw/typesafe 2026.10.1; everything else on the drift list unchanged since 10 Oct. Sources R13-S22 to R13-S35, R13-S119, R13-S120.

10 October scan (window ending 10 Oct 2026, 05:11 UTC)

No run took place on 9 October, so for the 10 October run the window was 8 Oct 2026, 05:12 UTC to 10 Oct 2026, 05:11 UTC (47 hours 59 minutes), with no gap and no overlap. The main checks ran at 05:17:31–05:17:39 UTC, follow-ups to 05:21 UTC.

SurfaceOutcome in the window
Hacker News24 queries on stories and comments: "jev" 18 stories and 109 comments; "typesafe" 4 stories and 30 comments; "decisions api" 1 story and 15 comments; 0 for "jev ollama". No statement restoring new-user credit
GitHub, npm, Hugging Face, DEV.to378 repositories created that match "jev" (57 with the jev topic); 29 Hugging Face models matching "jev" modified; 2 DEV.to articles. Index counts, not adoption. Six new repositories compare Jev with OpenAI's Decisions API; four were opened at a pinned commit (benchmarks leads)
TypeSafeA new Enterprise page (zero data retention "for enterprise customers"); a funding announcement (9 Oct) that states no change to signup, credit, prices or limits; a third monitored service on the status page (api.us-west-2.typesafe.ai); typesafe-sdk 0.7.3 and 0.7.4. No model, alias, price, limit, legal, subprocessor or incident change
"jev vision", "jev vlm", "jev vla"Still unexplained by any TypeSafe source; Google now completes "jev vision" to "jev vision model"; 1 Hacker News comment; three Jev-named vision-language models from other accounts on Hugging Face
Reddit, X, YouTube, TikTokReddit HTTP 403; the others were not searched: unknown, not zero

Drift: pydantic-ai-slim 2.55.0 (minor, not yet diffed); patch releases LiteLLM 1.104.2, typesafe-sdk 0.7.4, ai 7.0.137 and @ai-sdk/typesafe-ai 3.0.17; @openclaw/typesafe 2026.9.9. Sources R12-S01 to R12-S07, R12-S11, R12-S48, R12-S49.

8 October scan (window ending 8 Oct 2026, 05:12 UTC)

For the 8 October run the window was 7 Oct 2026, 05:12 UTC to 8 Oct 2026, 05:12 UTC. It overlaps the 7 October window by 5 minutes; no item fell in the overlap. The main checks ran at 05:16:50–05:17:21 UTC.

SurfaceOutcome in the window
Hacker News33 queries on stories and comments: "jev" 6 stories and 61 comments; "typesafe" 1 story and 2 comments; "decisions api" 2 stories and 12 comments. No report of a signup, credit, waitlist, outage or limits change ("jev down" found 2 comments that are not outage reports). 0 results for "jev rerank", "jev reranker", "jev sql", "jev postgres" and "jev sqlite"
GitHub, npm, Hugging Face, DEV.to211 repositories created that match "jev" (29 with the jev topic); 10 npm packages published for "jev" and 4 for "typesafe"; 22 Hugging Face models modified; 1 DEV.to article. Index counts, not adoption
TypeSafeA vendor case study on the TypeSafe blog (7 Oct; Reported; listed as a lead on benchmarks). No release, model, alias, limit, legal, subprocessor or incident change
"jev vision", "jev vlm", "jev vla"Still unexplained; same suggestions as 7 Oct; 0 Hacker News results
Reddit, XReddit HTTP 403; X not attempted: unknown, not zero

Drift: patch releases only (@effect/ai-typesafe 4.0.2, LiteLLM 1.104.1, ai 7.0.133, @ai-sdk/typesafe-ai 3.0.16). oh-my-claudecode v5.6.2 and @jkudish/jev-mcp 0.14.1 were read: failure paths unchanged.

7 October scan (window ending 7 Oct 2026, 05:17 UTC)

For the 7 October run the window was 6 Oct 2026, 05:17 UTC to 7 Oct 2026, 05:17 UTC, plus a backfill of the 3 minutes after the 6 October window (05:14–05:17 UTC), which found nothing. The main checks ran at 05:18–05:21 UTC.

SurfaceOutcome in the window
Hacker News41 queries on stories and comments: "jev" 13 stories and 42 comments; "typesafe" 1 story and 5 comments. No report of a signup, credit, waitlist, outage or limits change. Threads included OpenAI's "Decisions API" (public beta; an alternatives lead, not read)
GitHub, npm, Hugging Face, DEV.to249 repositories created that match "jev" (39 with the jev topic); 17 npm packages published; 17 Hugging Face models modified; 0 DEV.to articles. Index counts, not adoption
Benchmarks and evalsSix write-ups, including Vals AI's "An Independent Evaluation of TypeSafe's Jev"; graded on benchmarks (B33–B36)
Coding agentsAbout 15 new coding-agent and MCP leads (Claude Code, Codex, OpenCode, OpenClaw, Hermes, Pi); not read
"jev vision", "jev vlm", "jev vla"Still unexplained; TypeSafe's models page still says "Text only". Third-party "VL" models on Hugging Face may relate; not confirmed
Reddit, XReddit HTTP 403; X not attempted: unknown, not zero

No TypeSafe release, model or alias change, route ID change, legal-date change, limits change, subprocessor change or incident was found.

6 October scan (window ending 6 Oct 2026, 05:14 UTC)

For the 6 October run the window was 5 Oct 2026, 05:14 UTC to 6 Oct 2026, 05:14 UTC; the main checks ran at 05:19 UTC, with follow-ups to 05:24 UTC. The window starts exactly where the 5 October window ended, so there was no gap and no overlap. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News41 queries on stories and comments: "jev" 11 stories and 33 comments; "typesafe" 3 stories and 5 comments. Every coding-agent, version, outage, credit and limit query returned 0, except comment matches for "jev down", "jev error", "jev production" and "jev version" that are not outage or version reports (read)
GitHub repositories and releases192 repositories created that match "jev" (9 with the jev topic). TypeSafe's organisation has 12 repositories; the new one, typesafe-public-examples, holds only a README, so it is not a release. Read this run: opencode-tool-gate and jev-kit (coding agents)
npm, PyPI, DEV.to, Hugging Face14 npm "jev" and 7 "typesafe" packages with a last publish in the window; patch releases ai 7.0.128 and @ai-sdk/typesafe-ai 3.0.13; 1 DEV.to article; 22 Hugging Face "jev" models modified
RoutesVercel AI Gateway now lists TypeSafe itself (zero data retention) as a second provider for Jev beside DigitalOcean (Where to use); no other route, price or limit change
"jev vision"Still unexplained: suggested on all three endpoints, no Hacker News hit, no npm package, and TypeSafe's docs say Jev's input is text only. Bing and DuckDuckGo also suggest "jev vlm" and "jev vla"
Reddit, X, TikTok, YouTubeReddit HTTP 403, X login wall; TikTok, YouTube search, Lobsters, Product Hunt and Discord not attempted: unknown, not zero

No TypeSafe release, model or alias change, legal-date change, limits change, subprocessor change or incident was found. X monitoring is still not connected; nothing was bought.

5 October scan (window ending 5 Oct 2026, 05:14 UTC)

For the 5 October run the window was 4 Oct 2026, 05:14 UTC to 5 Oct 2026, 05:14 UTC; the main checks ran at 05:14–05:15 UTC, with follow-ups to 05:21 UTC. The window starts exactly where the 4 October window ended, so there was no gap and no overlap. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News7 stories and 4 comments matching "jev"; one "typesafe" story (a paywalled WSJ article about copycats, not read). 36 queries on browser, framework, cloud, outage, credit and other terms: 0 hits, except one "jev down" match that is a Show HN title, not an outage
GitHub repositories and releases250 repositories created that match "jev" (54 with the jev topic). No TypeSafe push since 29 Sep. @effect/ai-typesafe 4.0.1 (4 Oct), now pinned on framework integrations. Read this run: second-look (browser)
npm, PyPI, DEV.to, Hugging Face12 npm "jev" packages and 4 "typesafe" packages with a last publish in the window; typesafe-sdk 0.7.2 and langchain-typesafe 0.0.1a3 unchanged; 0 DEV.to articles; 18 Hugging Face "jev" models modified
"jev omni" and "jev vision""jev omni" is a community open-weights model (Jev-Omni, Gemma 4 12B base, Apache-2.0, 20 Sep) and its copies, not TypeSafe's. "jev vision" is unexplained; TypeSafe's docs say Jev's input is text only
Reddit, X, TikTok, YouTubeReddit HTTP 403, X login redirect; TikTok, YouTube search, Lobsters, Product Hunt and Discord not attempted: unknown, not zero

No TypeSafe release, legal-date change, limits change, subprocessor change or incident was found. X monitoring is still not connected; nothing was bought.

4 October scan (window ending 4 Oct 2026, 05:14 UTC)

For the 4 October run the window was 3 Oct 2026, 05:14 UTC to 4 Oct 2026, 05:14 UTC; the main checks ran at 05:16 UTC, with follow-ups to 05:24 UTC. The window starts exactly where the 3 October window ended, so there was no gap and no overlap. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News4 stories and 8 comments matching "jev"; none on signup, credit or an outage. One "Judge Jev" blog post (lead, not read)
GitHub repositories and releases225 repositories created that match "jev" (17 with the jev topic; 4 match trading, fraud, payment, judge or eval terms). No TypeSafe push since 29 Sep. oh-my-claudecode v5.6.1 (3 Oct, patch). Read this run: btc-hft-jev, the expense-policy gate, did-they-answer and competitor-hunter
npm, PyPI, DEV.to, Hugging Face11 npm "jev" packages with a last publish in the window; typesafe-sdk 0.7.2 unchanged; 3 DEV.to articles; 24 Hugging Face "jev" models modified
Third-party evaluations (not read)jev-safety-eval (content safety), jev-zh-tw-eval (Traditional Chinese) and system-one-security experiments. All Reported leads for a later refresh
Reddit, X, TikTok, YouTubeReddit HTTP 403, X login redirect; TikTok and YouTube not attempted: unknown, not zero

No legal-date change and no TypeSafe release were found. X monitoring is still not connected; nothing was bought.

3 October scan (window ending 3 Oct 2026, 05:14 UTC)

For the 3 October run the window was 2 Oct 2026, 05:14 UTC to 3 Oct 2026, 05:14 UTC; checks ran at about 05:15–05:23 UTC. It left a 1-minute gap after the 2 October cutoff (05:13–05:14 UTC on 2 Oct); a backfill of Hacker News and GitHub repository creation for that minute found 0 items. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News16 stories matching "jev" (one a substring false positive) and 50 comments; none on signup, credit or an outage; no email, lead, CRM or n8n item
GitHub repositories and releases300 repositories created that match "jev" (42 with the jev topic). No TypeSafe release and no new release of any pinned integration; oh-my-claudecode v5.6.0 is still the latest. Lead: madisonrickert/jev-permission-gate (created 3 Oct, 00:25 UTC)
npm, DEV.to, Hugging Face9 npm "jev" packages with a last publish in the window; 2 DEV.to articles; 29 Hugging Face "jev" models modified
Third-party studies (not read)Red Hat benchmark (2 Oct, 13:47), Synthpop calibration (14:01), a Substack calibration post (15:12), Vercel's "Jev for Python Engineers" (17:48). All Reported leads
Reddit, X, TikTok, YouTubeReddit HTTP 403, X login redirect; TikTok and YouTube not read: unknown, not zero

No legal-date change was found. X monitoring is still not connected; nothing was bought.

2 October scan (window ending 2 Oct 2026, 05:13 UTC)

Every research run checks what appeared in the preceding 24 hours. For the 2 October run the window was 1 Oct 2026, 05:13 UTC to 2 Oct 2026, 05:13 UTC; checks ran at about 05:13–05:28 UTC. Two minutes overlapped the 1 October window, with no recorded item in them, and there was no gap. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News19 Jev-titled stories and 80 comments matching "jev"; none on signup, credit or an outage. Leads include an Amazon decision-model report, a PII-redaction extension for Postgres and an EU-hosted "EU/Jev" service
GitHub repositories and releases330 repositories created that match "jev" (39 with the jev topic); 20 match "jev router", 18 "jev routing", 1 "jev moderation", 2 "jev spam". oh-my-claudecode v5.6.0 released. No TypeSafe push since 29 Sep, 21:11 UTC. The "jev mcp" query hit the search limit: unknown
npm, PyPI, DEV.to, Hugging Face13 npm "jev" packages with a last publish in the window, including @jkudish/jev-mcp 0.13.0 and ai 7.0.127; typesafe-sdk unchanged; 2 DEV.to articles tagged jev; Hugging Face counts capped at 100 per query
YouTubeReadable, dates relative; one moderation video and routing videos linked on the new pages (not watched)
Reddit, X, TikTokBlocked or login required: unknown, not zero
GitHub code search and trending, Lobsters, Product Hunt, DiscordNot attempted

Leads from this scan are listed on the pages they feed (model routing, content moderation). X monitoring is still not connected; nothing was bought.

1 October scan (window ending 1 Oct 2026, 05:15 UTC)

For the 1 October run the window was 30 Sep 2026, 05:15 UTC to 1 Oct 2026, 05:15 UTC; checks ran at about 05:15–05:36 UTC. 15.5 hours of it overlapped the 30 September window and were not re-recorded unless something changed; the earlier gap (29 Sep, 07:05–20:45 UTC) was backfilled for GitHub repository creation. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News29 Jev-titled stories with an exact timestamp in the window (4 after the 30 Sep window ended); 13 new comments read, none on signup, credit, pricing or an incident
GitHub repositories and releases354 repositories created that match "jev" (92 after 30 Sep, 20:45 UTC; 59 with the jev topic); 10 match "jev triage" and 77 "laya". Kev repository pushed 1 Oct; Laya v0.3.22 (29 Sep, found late). No TypeSafe push since 29 Sep, 21:11 UTC
npm, PyPI, DEV.to, Hugging Face28 npm "jev" packages with a last publish in the window, including @jkudish/jev-mcp 0.12.0 and ai 7.0.126; typesafe-sdk unchanged; 5 DEV.to articles tagged jev; Hugging Face counts capped at 100 per query
YouTubeReadable, dates relative; one new "Jev vs Laya" video (title only)
Reddit, X, TikTokBlocked or login required: unknown, not zero
GitHub code search and trending, Lobsters, Product Hunt, DiscordNot attempted

Leads from this scan are listed on the pages they feed (Jev vs Laya vs Kev, support-ticket triage). X monitoring is still not connected; nothing was bought.

30 September scan (window ending 30 Sep 2026, 20:45 UTC)

For the 30 September run the window was 29 Sep 2026, 20:45 UTC to 30 Sep 2026, 20:45 UTC; checks ran at about 20:45–21:02 UTC. The 13 h 40 min between the 29 September window (ending 07:05 UTC) and this one were covered only for Hacker News stories and the official TypeSafe organisation and packages. Counts are search-index counts, not adoption.

SurfaceOutcome in the window
Hacker News28 Jev-related stories with an exact timestamp in the window (22 more in the gap), including Elastic's reranking study, Casco's CVSS benchmark and a headline about an OpenAI "Decision API"
GitHub repositories and releases364 repositories created that match "jev" (55 with the jev topic); 10 match "jev benchmark" and 7 "jev eval". New official repository typesafe-ai/WorkflowEvals; a push to typesafe-ai/n8n-nodes-typesafe-ai. No SDK release
npm, PyPI, DEV.to, Hugging Face28 npm "jev" packages with a last publish in the window, including ai 7.0.123–7.0.124 and @ai-sdk/typesafe-ai 3.0.11–3.0.12; typesafe-sdk unchanged; 6 DEV.to articles tagged jev; Hugging Face counts capped at 100 per query
YouTubeReadable, but dates are relative ("9h ago"), so items are date-ambiguous; creator-run tests are leads
Reddit, X, TikTok, Lobsters, Product HuntBlocked or login required: unknown, not zero
GitHub code search and trending, Discord, awesome-list diffsNot attempted

Benchmark items from this scan are graded or listed as leads on benchmarks. X monitoring is still not connected; nothing was bought.

29 September scan (window ending 29 Sep 2026, 07:05 UTC)

Every research run checks what appeared in the preceding 24 hours. For the 29 September run the window was 28 Sep 2026, 07:05 UTC to 29 Sep 2026, 07:05 UTC; checks ran at about 07:05–07:08 UTC. It overlaps the release 1 window from 07:05 to 19:41 UTC on 28 Sep; items already recorded then were not listed again.

SurfaceOutcome in the window
Hacker News28 Jev-related stories with an exact timestamp in the window, including "Show HN: Jeff – Jev-compatible 0.8B decision models" (a lead for alternatives, not opened)
GitHub repositories and releases457 repositories created that match "jev" (41 with the jev topic); 20 match "jev mcp", 32 "jev claude" and 6 "jev hook". No TypeSafe SDK or plugin release; the official typesafe-ai/n8n-nodes-typesafe-ai repository was pushed. Search counts, not verified integrations or adoption.
npm, PyPI, DEV.to, Hugging Face19 npm "jev" packages with a last publish in the window, including @jkudish/jev-mcp 0.11.0 and two @ai-sdk/typesafe-ai releases; no in-window PyPI upload among the packages checked; 8 DEV.to articles tagged jev; Hugging Face results capped at 100 per query
Reddit, X, TikTok, LobstersBlocked: unknown, not zero
YouTube, GitHub trending, GitHub code search, Product Hunt, Discord, awesome-list diffsNot attempted

New hook and MCP repositories from this window are listed as "seen, not audited" on Claude Code and MCP. X monitoring is still not connected; the related access proposal remains unapproved and nothing was bought.

Release 1 scan (window ending 28 Sep 2026, 19:41 UTC)

Every research run checks what appeared in the preceding 24 hours. For release 1 the window was 27 Sep 2026, 19:41 UTC to 28 Sep 2026, 19:41 UTC; checks ran at about 19:42–19:57 UTC.

SurfaceOutcome in the window
Hacker News27 Jev-related stories with an exact timestamp in the window
GitHub repositories and releases478 repositories created that match "jev" (57 with the jev topic); releases of Laya v0.3.21, Ollaya v0.7.4 and v0.7.5, jevgrep v0.4.0–0.4.4 and AnyJev v0.2.0. Search counts, not verified integrations.
npm, PyPI, DEV.to, YouTube, Hugging FaceAt least 23 npm "jev" packages published; PyPI laya 0.3.21; 7 DEV.to articles tagged jev; 19 YouTube videos with a publish time in the window (10 are on the videos page); Hugging Face results capped at 100 per query
GitHub code search, Reddit, X, TikTok, LobstersNot accessible: unknown, not zero
Product Hunt, Discord, awesome-list diffsNot attempted

New items found this way are leads until checked: Ollaya and Jevper were checked and added to Alternatives; six other alternatives are listed as "seen, not yet checked"; Polylane's production claim is on Products. X monitoring is not connected; the related access proposal remains unapproved and nothing was bought.

What no report has done

  • An authenticated Jev call, an account bill or a signup.
  • An accuracy, calibration or latency benchmark run or rerun by the pool. The only check so far is offline arithmetic on one study's published log (30 Sep, benchmarks).
  • A census of production customers.
  • A trading bot run, an order, a payment or an eval run, even in mock or testnet mode.
  • A browser or computer-use agent run, a game or robot-simulation harness run, a coding-agent plugin or hook install, a framework package install, or a sign-in to an AWS or Azure console.
  • Complete coverage of X, Reddit, TikTok or other feeds that blocked access; daily scans are bounded and "nothing found" means "unknown".
  • Watched a video frame or read a transcript (YouTube showed a login wall).

Search published pools, pages, reports, and evidence.