Build Log

The research finally reaches the card

July 2, 2026

Traced Known's research corpus against what Jan actually sees, found the guided-interview questions were still wireframe placeholders, and rebuilt the Known Card's full question bank from the evidence-graded practice-wisdom corpus before this weekend's mother-conversation validation.

KnownResearchContentCrucible

capture questions (was 27, only 1 matched research verbatim)

Known Card sections (was 6, 4 net new)

High-confidence care patterns from research now covered

pre-existing OS script bug found and fixed (check-brain-sync.mjs)

Timeline

Start

Charge: does the research actually reach Jan?

Review-mode question — traced docs/known/brain/ + question-bank-seed.md against the live app

~30m

Explore agent finds the gap

27 live questions vs. 34 research seed questions — only 1 verbatim match. Mother-conversation validation doc found blank.

~40m

Crucible post-build audit run

GPT confirms: structural mistake, not sequencing — research and product tracks diverged, single source of truth was available and unused

~45m

Zeb reframes: not divergence, just unswapped placeholders

27 questions were wireframe defaults from screen-building, never replaced with the researched set — much simpler problem than reconciliation

~1h

Full replace: CaptureSession.tsx rebuilt from question-bank-seed.md

32 questions, 6 categories — category-name bug caught before shipping (memory vs. story) by cross-checking KnownCard.tsx and buildBookPages.ts consumers

~1h15m

Zeb asks: 30-second card vs. Known Card?

Confirmed same feature, two names from two docs written at different times — no conflation

~1h30m

Zeb asks: does the Known Card actually contain correct info?

Field-by-field audit vs. public-care-wisdom-corpus.md found 'When confused, try' had zero capture questions — hardcoded entries: [] since the card was built

~2h

Confused category added and wired end to end

4 new questions, new category, KnownCard.tsx/buildBookPages.ts/schema.ts updated, tsc clean

~2h15m

Zeb: implement every pattern candidate the research supports

Full audit against Known Card Pattern Candidates v1 template — 4 more sections needed (Personal care, Meals, Routine, Safety)

~2h45m

All 4 sections built, Safety handled as explicitly optional via wording not new mechanics

47 questions total, 10 categories, tsc + full production build both clean

~3h

Deploy Gate: found and fixed check-brain-sync.mjs root path bug

Script had never worked — resolved root 2 levels up instead of 3. Fixed, confirmed 0 drift.

~3h15m

Crucible post-build wrap run, findings printed, committed

GPT flagged no clinical review pass and no browser QA as the two live risks carried into the weekend

What shipped

Known: rebuilt CaptureSession.tsx's question bank end to end — 47 questions across 10 categories, replacing 27 wireframe placeholder questions that had never been swapped for the researched set

Known: added 4 new Known Card sections that previously did not exist — When confused try, Personal care, Meals, Routine, Safety (optional) — each sourced from docs/known/brain/public-care-wisdom-corpus.md's evidence-graded pattern candidates

Known: wired KnownCard.tsx's 'who' and 'confused' sections to actually read captured entries — both were previously either unfed by any question or hardcoded to always render blank

Known: fixed check-brain-sync.mjs — root path resolved 2 directory levels up from scripts/ instead of 3, so the drift-checker had never actually run successfully; confirmed 0 drift after the fix

Known: DESIGN_BRIEF_v1.md — logged a tagline candidate ('who you were is still who you are') for a later brand pass

Known: deployed to production (known.forgekits.build) after Zeb caught that the close sequence had never shipped it — see learning entry

ForgeKit OS: ignition.md §6 CLOSE and gates.md §Deploy Gate updated with an explicit product-deploy check — npm run retro only deploys forgekit-website, never the product app that was actually changed

“I did this to us... the 27 questions were just default questions we used during wireframing the screen. that can be recognized as such and we just replace them with the appropriate questions from the research”
— Zeb, reframing the research/product gap from a divergence problem to a simple unswapped-placeholder problem