Failure self-check, checklist question 9, and Drex on alternatives (20 Oct 2026)

This dated report records the evidence behind the update of 20 October 2026. It adds a self-check to When Jev fails that sets your own failure handling next to the 80 implementations on the map, adds a ninth checklist question that the map cannot count, corrects sentences that still gave the earlier count of 79, and reads Drex (Nace.AI) for the alternatives page. Nothing was run: no account, key, API call or third-party code.

What the self-check compares, and what it does not

• It compares five choices: what your code does on a Jev error or timeout, on a malformed answer, below its threshold, with no API key, and which code paths skip Jev. Each choice is shown in the map's own terms (fail-closed, fallback, fail-open, no-decision, advisory; held, acts-anyway, no-threshold below a threshold).
• What comes back: "N of 80 made the same choice", the map term, up to three example rows (name, domain and pinned commit, as on the map), and the checklist question that goes with it. A fallback choice also points to question 7 (is the fallback cheaper or riskier than failing?). "Not sure / not checked" points to the checklist question that shows what to look for.
• What it does not do: it does not read, run or test your code, does not call Jev, stores nothing and sends nothing. It does not say whether a choice suits your use; the counts describe the 80 implementations at their pinned commits, not a rate for Jev integrations in general.
• Without JavaScript: the same counts are in the checklist directly below the self-check, and every row is in the All 80 rows table.

The counts and their date

The self-check uses the map as recomputed on 19 Oct 2026 (n 80). Before use, every per-condition total and every domain-table cell was recomputed from the map's own cells, with 0 differences, and the map table was compared with the previously published page, with 0 differences. The bypass split is 13 "none found" and 67 with skip paths listed (13 + 67 = 80). Every example row named in the self-check appears in the map with the same term.

What the map does not record

• Whether a fallback or fail-open is visible (logged or shown to a person): the map has no column for it. It is now checklist question 9, with no count.
• Timeouts (question 2): examples are quoted, but timeouts were not tallied across the 80.
• Logging (question 8): the domain tables have no logging column.
• Real-world behaviour: every cell is a reading of source at a pinned commit; none of the 80 was run by this site.

Corrected count sentences

The map grew from 79 to 80 implementations on 19 Oct (ohmyjev). A scripted search of every current page found 23 current-tense sentences on 13 pages, plus one on the home page, that still gave 79 or the 79-row category numbers. All now give 80 and the current numbers. Dated change lines, recount notes and release entries keep the count they had on their date.

• When Jev fails: the five summary lines under the domain headings (fail-open 10, 13 and 9; acts-anyway 16; bypass recorded 80, each "of 80"), the checklist introduction and note, and the closing "not verified" lines.
• Use cases (10 let it through on a Jev error, of 80), Build (80 implementations, 9 Claude Code or MCP guards), coding agents (19 of 80 let the action go ahead), Jev as a judge (1 of the 10, of 80, recounted 19 Oct), and the home page's evidence table.
• Link texts and pointers on content moderation, support-ticket triage, model routing, n8n, confidence thresholds, testing without a key, errors and rate limits and the 9 Oct evidence notes.

Drex (Nace.AI)

Drex is a hosted decision model from Nace.AI with open weights for two of its versions; it is listed on alternatives under its primary layer, a hosted API from another provider, with the open weights as a second layer. It is not published by TypeSafe and is not Jev-named.

Status and shape

"Generally available" (launch post, 24 Sep 2026). POST https://console.nace.ai/v1/systemone with a bearer key; 1–512 questions; state up to 131,072 tokens. Documented Nace's docs say Drex is "wire-compatible with TypeSafe's Jev" and works with the TypeSafe SDK "without a fork"; this is the vendor's statement and was not tested. Reported

Price as published

Console pricing docs: drex-v2.0 $0.06, drex-v1.5 $0.05 and drex-max $0.06 per 1M input tokens; output "not billed" (read 20 Oct 2026, 05:17 UTC). The product page gives drex-v1.5 as "Cost per 1,000 decisions $0.04", a different unit; the two are not reconciled here. New accounts get "$25.00 of signup credit" per Nace, an access condition, not a statement of cost. Documented

Your data

Service Terms (last updated 19 Oct 2026): "No training by default"; deletion "within 48 hours after processing completes for Paid Accounts and within 7 days after processing completes for Free Tier Accounts". A reading of the text, not legal advice. Documented

Weights

nace-ai/drex-v2.0 (3e6d164) and nace-ai/drex-v1.5 (f80f9d2) under the "NACE AI OPEN RAIL-M LICENSE (MODIFIED)", which requires a separate commercial licence above $1,000,000 in revenue or funding, except for personal use or research. Base models per the cards (Reported; their licences not read). Documented

Claims against Jev

Nace.AI sells a competing hosted decision model. Its own figures: Drex 2.0 "61.25 vs 57.96" against Jev 1.13.0 on Decision Index 0.3.1, and Drex Max "6.9 points over Jev 1.14". No graded comparison exists. Reported

Not read: Nace.AI's website Terms and Privacy pages and its Console data processing addendum (opened, not read), the terms of the provider behind the OpenRouter route, the Drex Max launch-post body, the base-model licences. No Drex model was called.

Version checks and other edits

• LiteLLM 1.106.0 (published 18 Oct, tag commit 955bdaa): the Jev classifier's failure path is unchanged from 1.104.0; the classifier type is renamed oss_classifier (old names accepted) and failures now record a reason in the routing log. Files compared as text; nothing installed. No map cell changed. Documented (frameworks)
• n8n 2.44.3: activating or publishing a workflow is refused when a node that can run has no required credential; a workflow that is already active is re-registered at start-up without this check and stops at execution with "Node does not have any credentials set". Source read at the release tag; nothing run. Documented (n8n)
• OpenAI's Decisions API guide now accepts image URLs as well as base64 data URLs and is still a public beta (read 20 Oct 2026, 05:24 UTC). No comparison-card cell changed. Documented (Jev vs decision APIs)
• Other releases in the window were patch releases (ai 7.0.146, @ai-sdk/typesafe-ai 3.0.23, @tanstack/ai-typesafe 0.1.13, n8n 2.44.4); noted, not re-read.

Access and notices

Checked 20 Oct 2026, 05:13 UTC: signups open, no free credit for new accounts, API operational (90-day uptime 99.826%; last event resolved 14 Oct). Price $0.042 per million input tokens for jev-1.14.0 and jev-1.13.0, output free. TypeSafe split its API reference into separate pages (System One, List models, Request rules and errors) and added an OpenAPI link; no access, price or alias text changed. The home page keeps four notices; none was retired. Details: access status. Documented

Dates the scan covered

This update's scan covered 19 Oct 2026, 05:21 UTC to 20 Oct 2026, 05:13 UTC (23 h 52 min), starting at the previous scan's cutoff, so there is no gap and no overlap; the 72-hour decision-API look-back started 17 Oct, 05:13 UTC. Reddit returned HTTP 403, and X, YouTube and TikTok were not searched, so coverage of those surfaces is unknown, not zero.

What was not verified

• No Jev call, account or bill; no Drex call; no package installed or run.
• Whether any of the 80 behaves in real use as its source reads; whether a fallback is visible; timeouts across the 80.
• Drex documents listed above as not read; the drex-v1.5 product-page and docs prices are in different units and are not reconciled.

Method and sources

The self-check data was generated by script from the published map and checked against every count on the map page before release; the self-check page loads no outside resource. Sources read 20 Oct 2026: access and alias checks R20-S01 to R20-S13; reseller rechecks R20-S14 to R20-S18; OpenAI's Decisions API guide R20-S40; Drex R20-S100 to R20-S131 (price and terms re-read as R20-S51 and R20-S52); LiteLLM R20-S53 and R20-S204 to R20-S212; n8n R20-S216 to R20-S222. All are listed in the evidence file with their URLs and read times.

• All research reports
• 19 Oct research report (the map at 80)