A weekly pipeline scrapes vendor blogs, benchmarks new models, and forces every shiny release through one question — would this actually improve something we run? Most weeks the honest answer is no. That discipline is free here.
The running scorecard across everything ever tried — platforms, models, connectors. Winners, scars, and the ones dropped on sight. Start here.
VENDOR · MODEL What we tried. What kept working.Vercel & Supabase scars · DigitalOcean & GitHub wins · which models actually connect · the little things nobody tells you. open the ledger →Each week the factory triages everything new — model releases, pricing changes, tool launches, deprecations — and publishes the verdicts. Watch means unproven claims, Adopt means it earned a place in production, Ignore means noise.
WEEK 31 Factory week in review — 2026-W31.30 items judged. 2026 · W31 WEEK 30 Factory week in review — 2026-W30.30 items judged, 26 dropped, 4 kept. 2026 · W30 WEEK 29 Factory week in review — 2026-W29.W29 surfaces two P0 incidents detected on review day: claude-sonnet-4.6 and claude-haiku-4.5 aliases were both re-pointed on 2026-07-19, covering 10 a 2026 · W29 WEEK 28 Factory week in review — 2026-W28.W28 review: 30 items across LlamaIndex blog (15) and model-watch (15). 2026 · W28 WEEK 27 Factory week in review — 2026-W27.W27 review: 30 items processed across three pipelines (inbox: 15, model-watch: 15). 2026 · W27