Build Log

Day History: two-round Crucible, research-grounded redesign, calendar shipped

July 2, 2026

Ran Guide 3's day-history design through a two-round, four-persona Crucible in a new batched format, let the Council reject the original day_ratings design, redesigned it around Known's existing handoff_note mechanism and the research corpus instead of founder foresight, then built and shipped the calendar + day-detail drill-down with zero new tables.

KnownCrucibleData modelUX

Crucible rounds run (4 personas each, batched format)

new DB tables (redesign reused existing portraitEntries + card_snapshots)

new routes/files shipped (history calendar, day-detail, lib/date.ts)

new default Crucible execution format adopted (feedback_crucible_format)

Timeline

Start

Read last session's design (Guide 3)

day_ratings table + calendar view, staged build order left by known-card-history-2026-07-02

+20m

Crucible run 1 — wrong format, corrected twice

First tried the interactive turn-by-turn crucible.mjs script (crashed, ERR_USE_AFTER_CLOSE), then tried simulating live chat turns myself — Zeb corrected both times to: one batched prompt, four independent persona sections, synthesis at the end

+40m

Round 1 verdict: near-unanimous hold

day_ratings (good/rough/mixed tap) rejected by Architect (unvalidated/redundant), QE (untested upsert, silent data loss), Steward (informal decline-signal risk), Customer/Jan (no day-one payoff)

+55m

Correction: iterate against research, not a single user

Zeb: upsert already validated live; Known's model is iterate against the brain/research corpus until a design earns its build, not wait for one user to hit the wall

+75m

Research pass grounds the redesign

Corpus has no support for a rated day construct, but strongly supports a 'handoff' construct (All About Me's core question + Journal tool) — and Known's schema already has an unused-for-this-purpose entryType='handoff_note' value built for exactly this

+90m

Round 2 verdict: proceed

Redesign (reuse portraitEntries, no new table, presence-only calendar, no rating/score) cleared 3 of 4 personas; QE flagged two concrete build requirements (N-entry handling, local-date grouping) folded directly into the build

+140m

Built and verified

lib/date.ts extracted, history calendar + day-detail routes shipped, tsc clean, live smoke test in real app confirmed working (one reported bug turned out to be user error, not a defect)

What shipped

Ran a two-round Crucible (all four design-tier personas: architect, qe, steward, customer) on Guide 3's day-history design using a new single-batched-prompt format instead of the interactive per-persona script

Round 1 Council rejected the original day_ratings (good/rough/mixed) design as unvalidated and ethically risky (informal decline signal)

Redesigned day history around Known's existing entryType='handoff_note' mechanism and research-corpus grounding (All About Me's handoff question + Journal precedent) instead of a new table

Round 2 Council approved the redesign with two concrete engineering requirements: handle N same-day entries, group by local calendar day not raw timestamp

Extracted lib/date.ts (localDateStr) — previously duplicated inline in app/api/portraits/[id]/route.ts, now shared so day-history grouping matches card_snapshots' existing day-boundary rule exactly

Built app/portraits/[id]/history/page.tsx — month-grid calendar, presence-only (no rating/color/score), lit if a day has a handoff_note entry or a card_snapshots row

Built app/portraits/[id]/history/[id]/page.tsx (day-detail) — lists all of a day's handoff notes (any count) plus that day's Known Card snapshot if one exists, using the real KnownCard.tsx section labels (corrected a wrong guess before shipping)

Added a 'Day history →' nav link on the portrait page, preserving the existing 'last 7 days' label rather than replacing it

tsc --noEmit clean; live smoke-tested in the running app via Zeb's authenticated session (Clerk auth wall blocks Bash-driven verification, as expected)

Saved a new persistent memory (feedback_crucible_format.md) locking in the four-persona batched-prompt-plus-synthesis format as the default for all future Crucible runs

“We iterate on this feature until it is worthy to build based on the brain and research not waiting for a single user.”
— Zeb, correcting the Crucible round-1 recommendation to hold the feature pending live user validation