Activity log

Append-only. Newest at the bottom. Format: - YYYY-MM-DD — action: what changed (pages).

• 2026-09-17 — ingest: snapshot of docs.typesafe.ai (113 pages via sitemap + llms.txt), api.typesafe.ai/openapi.json (info version 0.2.0), typesafe.ai site (home, manifesto, team, 4 blog posts, legal), evals.typesafe.ai (4 workflows), press coverage, and shallow clones of typesafe-sdk-python @420ef4f, typesafe-sdk-js @66880cc, skills @65a39f3, system-one-adapter-python @0bb819b. Manifest written to raw/MANIFEST.json.
• 2026-09-17 — create: schema (CLAUDE.md) with fixed page inventory; initial pages written by parallel agents for concepts, reference, patterns, cookbooks, guides, entities, syntheses.
• 2026-09-17 — create: 75 content pages written (13 concepts, 18 reference, 5 patterns, 19 cookbooks, 6 guides, 11 entities, 4 syntheses) plus generated index; lint clean (no dangling wikilinks, no orphans).
• 2026-09-17 — fix: The Register article re-captured as a claim summary (raw/site/press-theregister.txt); note added on reference/python-sdk about the upstream model="jev" sample; note on concepts/primitives about the two upstream batching multipliers (11.5x/9.6x vs 12.2x/10.0x).
• 2026-09-17 — deploy: first full deploy to https://jev-wiki.fortibus-group.workers.dev (Cloudflare Workers, static assets).
• 2026-09-17 — lint: cold QA pass against the deployed site (8 agent tasks: 7 complete, 1 partial). Fixes: TypeScript routing samples on patterns/confidence-routing and patterns/intent-routing; full status table on reference/http-api; inline caveats on cookbooks/entity-alignment (model id, placeholder key, TYPESAFE_ENDPOINT, concurrency note); playbook Step 0 wording; console key-URL guidance on guides/quickstart and reference/agent-skill; llms.txt now links the playbook first; /raw/wiki/ without .md is accepted.
• 2026-09-17 — style: human-facing HTML restyled to FakeNerd.ai Brand Standards Vol. I (Space Grotesk, Paper/Ink/Accent, mast with icon + wordmark → fakenerd.ai, skip link, focus rings). Agent surfaces (/raw, llms.txt, llms-full.txt, index.json) unchanged.
• 2026-09-20 — ingest: owner's curated X bookmark folder "Jev" (61 posts, 19 with long-form articles) captured to raw/x/ via scripts/capture_x.py; READMEs and metadata of the 42 GitHub repos they link captured to raw/x-repos/. Raw captures are not published.
• 2026-09-20 — create: community-tier ideas/ section (overview, consult, patterns [25 patterns with fit verdicts], field-reports, community-repos), each under a hard token budget. Build now emits per-page token estimates in llms.txt, index.json and wiki/index.md, a "Route by task" table at the top of llms.txt, a size warning over ~6k tokens, and lints source_tier on ideas/ pages. README gains copy-paste agent prompts. Cross-links added from guides/agent-integration-playbook, concepts/use-case-map, entities/github-repos, guides/testing-and-evaluation.
• 2026-09-20 — lint: cold consult test against the deployed site (sample helpdesk project, 8 decisions). Agent read 4 pages, 12.7k tokens (budget 15k), matched 7/8 items to patterns, and was steered away from all four planted bad uses (dollar threshold, SLA timer, reply drafting, nightly summary). Fixes applied: consult guide now groups decisions by shared state before costing, includes a worked cost example, defines the fit vocabulary, and budgets guides/choosing-a-primitive as a conditional step; patterns Sources block compacted (450 tokens saved); homepage consult prompt points directly at ideas/consult instead of llms.txt.
• 2026-09-20 — style: human-facing HTML restyled to FakeNerd.ai Brand Standards Vol. III (white page, grey cards, teal accent, IBM Plex Mono/Sans, prompt-glyph wordmark in the header, circular mark in each page's section header, 2px radius, no italics, outbound links open in a new tab). The locked hub tagline is available behind SHOW_TAGLINE in scripts/build.mjs and is off. No link to fakenerd.ai anywhere. Agent surfaces unchanged.
• 2026-09-20 — ingest: @k2sbhai "7 Jev skills" article + attached audit-prompt image (raw/x/). All seven use cases already covered by ideas/patterns; added retrofit start order and community-repo vetting checklist to ideas/patterns, community audit prompt to reference/agent-skill and the README. typesafe-ai/skills re-checked: still at 65a39f3, already covered by reference/agent-skill.
• 2026-09-21 — move: public URL is now https://fakenerd.ai/jev-wiki (live proxy of the jev-wiki worker). BASE_URL default updated; HTML root-relative links are prefixed with /jev-wiki and the worker accepts that prefix on any host, so header/home links no longer fall through to the host site's root.
• 2026-09-21 — ingest: 5 X posts (@jerryjliu0 DocJev, @startupideaspod/Ryan Vogel, @AIGuide_ GTM list, @shannholmberg marketing list, @charliejhills repo roundup) and 13 repo READMEs (raw/x-repos now 55). Skipped @cyrilXBT Fastlane promo (no Jev mechanics). Two roundup repos (lahfir/agent-desktop, irfndi/prism-liquidity-agent) never mention Jev and are listed as such.
• 2026-09-21 — restructure: ideas/ split into indexes + domain pages so consult cost stays flat as the section grows. ideas/patterns is now a ~2k-token index of 33 patterns with routing by project type; patterns live in ideas/patterns-agents (P01-P14), ideas/patterns-data (P15-P18, P24-P27; new P26 document classification/packet splitting, P27 moment finding in transcripts) and ideas/patterns-business (P19-P23, new P28-P33 incl. P33 metric-driven ad/budget/timing decisions = poor fit). ideas/community-repos is now a repo index; detail in ideas/repos-agents and ideas/repos-apps. Field reports gain DocJev pilot numbers and Vogel's email run and failures (budget raised to 4,000). Build now fails when an ideas/ page exceeds its budget. Pattern IDs are permanent.
• 2026-09-21 — edit: ideas/patterns-business P33 gains a worked restructure sketch (code computes metric labels, Jev sees labels + semantic parts; request shape checked against reference/http-api), requested by the cold consult test.
• 2026-09-21 — move: public URL is now https://jevwiki.ai (custom domain on the jev-wiki worker, plus www). BASE_URL default updated; links are root-relative again. jev-wiki.fortibus-group.workers.dev stays up (workers_dev: true); the /jev-wiki path prefix is still accepted for old links. The fakenerd.ai/jev-wiki copy is a stale static snapshot in another project and is no longer the canonical address. Rebrand away from FakeNerd styling pending new brand standards.
• 2026-09-21 — style: human-facing HTML restyled to theconsigliere.ai Brand Standards v2.0 (ink ground, IBM Plex Mono throughout, oxblood emphasis, gilt hairlines, terminal pattern: $ prompts, # comments, bracketed statuses, ~/wiki paths, blinking caret; primary lockup in the footer; roundel favicon). FakeNerd marks and styles removed. Mark assets generated by scripts/make_brand_assets.mjs from the master PNG (transparent parchment version replaces the blend-mode knockout). Agent surfaces unchanged.
• 2026-09-21 — ingest: Nate B. Jones, member guide "Find the Jev-shaped problems in your software" (read via his library MCP; used with his permission on condition of credit; digest notes only in raw/nate/, no guide text or prompts reproduced). Added: four-placement routing table on ideas/patterns; new P34 classifier inside an interface element (ideas/patterns-agents); 50-case check and shadow-mode procedure on guides/testing-and-evaluation; two fit-test disqualifiers and a sharper first step on ideas/consult; first-hand measurements and the OpenRouter route on ideas/field-reports; design detail on P18. Public sources he cites captured independently: @nedwize + kyotofin/tax-doc-classifier (P26, repos-apps; 56 repos), @iannuttall (P18), @dabit3 (P34), @gregpr07 (P12). Credit on ideas/overview and the README. Field-reports: second-hand rumour row dropped, replica table condensed to trust notes (detail lives in ideas/repos-apps).
• 2026-09-21 — restructure: ideas/patterns-agents split; browser/computer/voice/UI patterns (P12-P14, P34) now on ideas/patterns-interface. ideas/field-reports split; measured numbers, gateway routes and replica trust notes now on ideas/measurements. Pattern index routes to four domain pages.
• 2026-09-21 — ingest: Diogo Almeida's (@CompleteSkeptic, TypeSafe CEO) public notes on Jev-centred coding agents (raw/community/, labelled proposed/not shipped): KV-cache routing economics on P02, per-query context selection on P07, conditional AGENTS.md and hierarchical tool routing on P08, new P35 security-aware routing and P36 background read-only task triage. Grep.ai AgentRun article (@MiguelriosEN, @_aj): new P37 self-compiling workflow, vendor numbers on ideas/measurements, limitations on ideas/field-reports. Archer Hume's "Jev's architecture unmasked" (10k-call probe): empirical findings (question isolation verified, option-set effects unverified, output_tokens accounting, batching to 5,000 questions, ECE 0.031) on field-reports/measurements; deductions labelled hypotheses. Repos: ryana/jevify (prompt, linked from reference/agent-skill and consult) and kerpopule/hermes-jev-skills; index at 58. @0xCodila cited once as a restatement; @socialwithaayan used only for a star-drift caveat. Consult guide gains jevify's three opportunity kinds and skipped-judgment prompt. Team page notes Diogo's handle.