Shaduf.Research preview
Jev: Use Cases, Alternatives & Products/Email and lead sorting, oh-my-claudecode v5.6.0, Products refresh and n8n nodes
Research reportRun run:5a50a0ba-e2af-412c-9556-7e4cb232a36b · checks , 05:15–05:33 UTC

Run 6 report: email and lead sorting, oh-my-claudecode v5.6.0, Products refresh and n8n nodes (3 Oct 2026)

This dated report records the evidence behind the release of 3 October 2026. The maintained pages built from it are: email and lead sorting with Jev (new), Jev n8n nodes (new), the refreshed Products hub, the oh-my-claudecode check on Claude Code and MCP guardrails, the rate-limit update on errors and rate limits, and the refreshed home status and access status. Facts on those pages may be rechecked later; this report keeps the 3 October readings.

Run: run:5a50a0ba-e2af-412c-9556-7e4cb232a36b (scheduled regular research run 6) · Pool: pool_jev_catalog · Runner: one runner plus two helper agents that read source in parallel; the runner spot-checked their key lines against the raw files at the same commits (05:25–05:28 UTC) and they matched at every line checked · Checks: 2026-10-03, 05:15 to 05:33 UTC; runner started 05:14 and ended 05:36 UTC. Nothing in S1–S6 slipped.

The run's full notes and its source ledger (62 records, IDs R6-S01 to R6-S77: URL, access time, label, excerpt, outcome) are held in the pool's private record. Key sources are linked below and on the pages each finding feeds.

Rules kept: no live Jev call, account, key, mailbox connection, package install or third-party code run. Nothing in this run is Tested. Every finding below is Documented (official docs, legal text, registry metadata, or source read at a pinned commit), Reported (a named third party's own claim) or Unverified (a lead only). Reddit, X, TikTok and YouTube were not read in the scan.

Previous release: regular-2026-10-02-routing-moderation; report: model routing, the jev-mcp version check and content moderation. All dated reports: Research.

1. Summary answer, as of 3 October 2026

  1. Access is unchanged (checked ). Signups are open with no new-user credit, as since 27 Sep, 22:33 UTC. The API is operational, as since 29 Sep, 22:05 UTC. The console still returns HTTP 403, so the pool did not observe signup. One change: TypeSafe's models page now lists 100K tokens per second and 80 requests per second; it listed 40 requests per second on 2 Oct at 05:14 UTC. Documented
  2. Email and lead sorting: ships. Five projects were read at pinned commits. None of them changes the mailbox when Jev errors. Only 3 of the 5 can write to the mailbox at all, and only 1 of those 3 sends unsure mail to a "Review" label. Selection was by GitHub stars, a reading order, not a sample. Documented
  3. oh-my-claudecode v5.6.0: failure path unchanged, but the published 250 ms claim is now version-bound. The default timeout is 2,000 ms from v5.6.0. A new environment switch, OMC_JEV=<point>:active, lets Jev decide. Documented
  4. Products hub: ships. The ledger grows from 37 to 61 rows; grade A (Jev call read in source) goes from 15 to 39. 5 of the 7 published claims were rechecked and are unchanged; the other 2 were not rechecked. None of the 15 earlier grade-A repositories is gone or archived. Independently verified in production: still 0.
  5. n8n nodes: ships, and the npm question carried since run 3 is resolved. TypeSafe's node is the scoped @typesafe-ai/n8n-nodes-typesafe-ai: npm provenance points to the typesafe-ai repository at tag 0.9.0, and it is the only Jev node verified on n8n. The unscoped n8n-nodes-typesafe-ai belongs to a third party. All 5 nodes read stop the workflow on a Jev error by default. Documented

2. Corrections and updates to published pages, and how they were applied

PagePublished claimCorrection or updateApplied as
Home, where-to-use/access-statusChecked 2 Oct, 05:14 UTC; notice rate-limits-changed (40 requests/s)Checked 3 Oct, 05:16 UTC; listed limit now 80 requests/s. Same state, since-dates and headlineBoth pages re-rendered from home-status.json: rate-limits-changed retired (superseded), rate-limits-changed-2026-10-03 added, check time 3 Oct 2026, 05:16 UTC, and a timeline event for the limit change
build/claude-code-mcpoh-my-claudecode "250 ms default"; "latest release is still v5.5.0 / v5.6.0 not yet checked"v5.6.0 (1 Oct): failure path unchanged; default timeout 2,000 ms (250 ms up to v5.5.0); OMC_JEV=<point>:active makes Jev decideoh-my-claudecode row, version line and checklist item 6 updated
build/errors-and-rate-limitsListed limit 40 requests/s; oh-my-claudecode 2,000 ms default only on devListed limit 100K tokens/s and 80 requests/s (40 from 30 Sep to at least 2 Oct, 05:14 UTC); the 2,000 ms default is released80 requests/s with the dated bracket for the change; oh-my-claudecode 2,000 ms marked as released in v5.6.0 on 1 Oct
limitationsRate-limit row: "100K tokens/s and 40 requests/s (rechecked 1 Oct …)"80 requests/s, rechecked 3 Oct, 05:16 UTCRate-limit row updated
where-to-use hubMentions 40 requests per second; checks dated 2 Oct80 requests per second; rechecked 3 Oct, 05:16 UTCLimit and check dates updated
where-to-use/official-vs-reseller"an npm package named n8n-nodes-typesafe-ai exists at 0.1.1 (21 Sep); whether TypeSafe publishes that package was not checked"TypeSafe publishes the node on npm as @typesafe-ai/n8n-nodes-typesafe-ai 0.9.0 (provenance to typesafe-ai tag 0.9.0); the unscoped n8n-nodes-typesafe-ai is a third party's package (maintainer dommonte). Documented, 3 Octnpm publisher caveat replaced with the answer; links build/n8n
products37 rows, 15 grade A; "Official repository seen but not reviewed: typesafe-ai/n8n-nodes-typesafe-ai"; older Jevmail and oh-my-claudecode rows24 new grade-A rows; on a Jevmail error the email stays unclassified and is retried, repeated 429s pause the run and a 403 stops it; oh-my-claudecode at v5.6.0Products refreshed: 61 rows, 39 grade A; the "seen but not reviewed" line removed; Jevmail and oh-my-claudecode rows updated
use-cases hubJevmail row: "Stops — Retries 429/5xx once, then throws"Jevmail asks only for read-only Gmail access, so nothing in the mailbox changes; on an error the email stays unclassified and is retried on the next run; repeated 429s pause the run. In the shared vocabulary: no-decisionJevmail row reads no-decision and links the email and lead sorting page
Earlier pool notes (internal)Run 5's notes and the 2 Oct plan compared GitHub tag 0.9.0 with npm n8n-nodes-typesafe-ai 0.1.1 as if they were one packageThey are two packages from two publishersNot public; no page change

No earlier claim was found false at its check time: the 40 requests/s and 250 ms readings were correct when made. Older report pages (for example release 1 and evaluation evidence) are dated snapshots and keep their readings.

3. Findings per item and ship verdicts

S1. 24-hour scan

  • Window: 2 Oct 2026, 05:14 UTC to 3 Oct 2026, 05:14 UTC. It leaves a 1-minute gap after run 5's cutoff (05:13 to 05:14 on 2 Oct); the backfill for that minute (Hacker News, GitHub) found 0 items.
  • Index counts: 16 Hacker News "jev" stories (one a substring false positive) and 50 comments; 300 GitHub repositories created matching "jev", 42 with the jev topic; 9 npm "jev" packages; 29 Hugging Face "jev" models modified; 2 DEV.to articles. These are search-index counts, not verified integrations or adoption.
  • No TypeSafe release, no legal-date change, and no new release of any pinned integration. No email, lead, CRM or n8n item on Hacker News.
  • New third-party studies, not read (Reported): a Red Hat benchmark, a Synthpop calibration study, a Substack calibration post, and Vercel's "Jev for Python Engineers". Lead for claude-code-mcp: madisonrickert/jev-permission-gate (Unverified).
  • Reddit (HTTP 403), X (login redirect), TikTok and YouTube were not read: unknown, not zero.

S2. Volatile fields and home status: refreshed

  • Rechecked at 05:16 UTC. Only the published rate limit changed: 40 to 80 requests per second. The change happened between 2 Oct, 05:14 UTC and 3 Oct, 05:16 UTC; there is no archive capture and the docs changelog returns 404.
  • Unchanged, all Documented: price $0.042 per million input tokens with free output; jev-1.13.0 with both aliases; status "All services are online" (API 90-day uptime 99.826%); console 403; OpenRouter needs no separate TypeSafe account; TypeSafe is on OpenRouter's ZDR list; Vercel's endpoints list only DigitalOcean (has_zdr: false); Cloudflare shows "Zero data retention Yes"; SDK versions; legal dates.
  • Headline unchanged: "Available to new users: signups open, but new accounts get no free credit. API operational." Shown as "checked 3 Oct 2026, 05:16 UTC".
  • Notices: kept new-user-credit-disabled, lookalike-resellers (all three named sites still answered HTTP 200) and typesafe-workflow-evals-code (retire 6 Oct, 16:25 UTC). Retired rate-limits-changed (superseded). Added rate-limits-changed-2026-10-03 (event, Documented; retire 10 Oct, 05:16 UTC). One timeline addition: the limit change.

S3. use-cases/email-and-lead-sorting: meets the ship condition

  • Five projects read at pinned commits: Jevmail f6f20af (unchanged since the 28 Sep audit), the ilyamk Gmail labeler ade42bf, the parth-kp IMAP classifier db96194, vynnlee jev-mail 61e346f, and spam-leadgen-pipeline 78a7592 (lead qualification). Selection: GitHub stars in a search with 59 results, plus one lead from S1. A reading order, not a survey.
  • Counts (Documented; 5 projects chosen by stars, a reading order, not a sample): 3 of 5 can write to the mailbox; 0 of 5 have delete or trash code, although vynnlee requests the full https://mail.google.com/ scope and parth-kp has full IMAP access; 0 of 5 change the mailbox on a Jev error. For unsure mail, 1 of the 3 writers labels it "Review" (vynnlee), 1 applies the label anyway (ilyamk), and 1 marks it processed and never re-checks it (parth-kp). 4 of 5 cut the body before sending (1,000 to 5,000 characters); parth-kp sends it uncut.
  • Cost per 1,000 emails (Documented price; pool arithmetic assuming 4 characters per token plus 500 tokens of question text, both assumptions): about $0.004 for very short emails; about $0.04–$0.15 at the four projects' body cuts; up to about $1.34 if every email filled Jev's 32k state. A demo figure of "1,700 emails for $0.18" (Reported, 18 Sep) works out to $0.106 per 1,000.
  • Title query source: Google suggested "jev email classifier" for "jev email" (05:22 UTC).

S4. oh-my-claudecode v5.6.0: version check done

  • Pinned: tag v5.6.0, commit e74b22c. The compare from v5.5.0 shows 165 commits touching 25 Jev paths.
  • Unchanged (Documented): on a timeout, HTTP error or invalid response the heuristic twin decides and the resolver "never throws"; a point stops calling Jev after 3 consecutive failures; nothing is sent unless OMC_JEV lists points; excerpts are cut to 200 characters; points still run in shadow by default.
  • Changed (Documented): default timeout 2,000 ms (was 250); new activation OMC_JEV=<point>:active or all:active, which makes Jev decide and prints a warning; a new script-side judgment channel with the same degrade-never-block rule; the request shape is fixed.

S5. Products hub refresh: meets the ship condition

  • 24 new grade-A rows: triage 5, routers 5, moderation 5, email and lead 4, n8n nodes 5 (1 official). No duplicates and no B-to-A upgrades. Totals on 3 Oct: A 39, B 21, C 1, total 61; independently verified in production 0.
  • Failure behaviour on a Jev error across the 24 new rows (defaults): fail-closed 14, fail-open 4 (all moderation), fallback 4, no-decision 2. Documented
  • Claims: Brizz, Polylane, the Vercel CEO post, the Bryo CTO post and FundaAI are online and unchanged (Reported). The two claims relayed by VentureBeat were not rechecked: the page has no direct link, and VentureBeat returned 429 on 28 Sep.
  • The 14 GitHub grade-A repositories are public and none is archived (HTML check); langchain-typesafe on PyPI is unchanged. LiteLLM was proposed under "not counted as products" (a framework feature); the manager decides.

S6. build/n8n: meets the ship condition

  • Nodes read: the official @typesafe-ai/n8n-nodes-typesafe-ai 0.9.0 at c12537b, and 4 community nodes at pinned commits. 15 Jev or TypeSafe n8n packages were found on npm (1 official, 14 community); 10 were not read. The runner's note said 14; the manager corrected the total to 15 from the note's own list of 5 read and 10 not read.
  • Publisher (Documented): the official package's npm maintainers use typesafe.ai addresses, and its SLSA provenance names typesafe-ai/n8n-nodes-typesafe-ai at tag 0.9.0, commit c12537b. The unscoped n8n-nodes-typesafe-ai 0.1.1 is maintained by dommonte, with provenance to DomMonte's own repository.
  • n8n Cloud (Documented): n8n's docs say unverified community nodes are not available on n8n Cloud. Of 1,765 entries in n8n's public listing, the only Jev or TypeSafe entry is the official node (listed 2 Oct, 11:44 UTC). One maintainer says n8n declined verification of their node in September (Reported).
  • Failure behaviour (Documented): with Continue On Fail off, 5 of 5 nodes stop the workflow. With it on, 3 of 5 can put the failed item on a real branch unless a Fallback, Uncertain or low-confidence output is configured.
  • Title query source: "jev n8n" was a Google and YouTube suggestion (05:24 UTC).

Spare-time queue

  • Failure-checklist preparation: done (internal, for the next run). It maps 32 implementations across 6 domains; on a Jev error: fail-closed 16, fail-open 5, advisory 2, fallback 6, no-decision 3.
  • jev.bevel.software ("EU/Jev"): done. The site states "eu/jev runs its own model", says Jev model IDs return HTTP 400 and says it is "Not affiliated with TypeSafe AI". It claims inference in Germany and prices at €37 per billion input tokens. So it is a Jev-compatible API serving another model: it does not resell Jev and does not claim to host Jev. Reported (the operator's own claims). Not a home notice.
  • trust.typesafe.ai: not done. The page is a 7 KB JavaScript shell and no headless browser was available; nothing was installed.
  • Netlify data terms: not done.

4. Uncertainties and access gaps

  • When the rate limit changed: a 24-hour bracket only.
  • Signup is still not observed directly (console 403). X search is unavailable; Reddit, TikTok and YouTube were not read.
  • Email: ilyamk's granted OAuth scope (no manifest in the repository); spam-leadgen's no-key behaviour and model; HTTP timeouts are not set in three of the email projects; whether question text is billed, and the characters-per-token ratio (both labelled assumptions in the cost table).
  • n8n: how n8n core handles the "error output" setting (only the official node's code comment); the 10 n8n packages not read; the declined-verification claim (Reported only).
  • Products: the two VentureBeat-relayed claims were not rechecked; grade-A status was checked through HTML pages, not the API.
  • Helper-agent reads were spot-checked, not re-read line by line; unchecked cells keep path and line references so they can be verified.
  • Whether any project behaves live as its code reads: nothing was run.

5. Not done

  • Spare-time items trust.typesafe.ai (no headless browser) and Netlify data terms.
  • withoneai/mastra-inbound-router (lead routing) and anshulkhare7/jev-gmail-triage; the n8n Gmail-leads workflow from the Nyarko video (not read as a project).
  • YouTube in the scan; the scan leads (Red Hat, Synthpop, the calibration post, the Vercel tutorial, jev-permission-gate).

6. Key sources

Search Shaduf

Search published pools, pages, reports, and evidence.