Build Log

Known's second pillar: from an anecdote to the Behavior Loop

July 2, 2026

Zeb and Court walked through Guide 1 and found the Known Card was the only feature with an obvious reason to return — this session found the second one, first from a hypothetical (rejected) and then from the brain corpus itself (kept).

KnownDesignResearchCrucible

Crucible Council runs (one mechanically broken, one real)

brain corpus documents read in full to find the real signal

design doc written, then substantially rewritten in place

code changed — design-only session

Timeline

Start

The gap named

Zeb + Court walkthrough of Guide 1 found the Known Card is the one clear feature; no second reason to return surfaced

+20m

First Council run misfires

GPT reviewed the Known Card as an artifact instead of answering the a/b/c/d question — persona args weren't reaching the script correctly

+35m

Sharpened Charge, mechanical bug found

npm arg-passing bug identified (personas landed in wrong argv slot, silently fell back to COUNCIL_DEFAULT); fixed by passing personas as a distinct shell argument

+45m

Council lands, output is generic

Correct personas ran but surfaced only boilerplate ("automate data capture") — synthesized manually instead of trusting the raw output

+60m

Guide 4 v1 written

Day-history-as-pattern-recognition, three-tier design with false-positive-prevention machinery, anchored on an invented example ("Sinatra worked 4 of 5 times")

+75m

Zeb redirects

Corrected Court's relationship (wife, not mother-in-law), rejected the single-anecdote anchor, named the brain/research corpus as the actual north star

+95m

Corpus read in full

source-extractions.md, dementia-home-care-corridor.md, public-care-wisdom-corpus.md read end to end — found convergent, source-verified signal on repeated questions/Behavior Loop that the anecdote-driven design never touched

End

Guide 4 rewritten as v2

Behavior Loop established as the real second pillar, reusing existing schema seams (portraitEntries.category already anticipates 'confused'), day-history demoted to a supporting feature

What shipped

docs/known/guides/guide-04-pattern-recognition.md — full design doc for Known's second pillar, written twice in the same session (v1 rejected, v2 kept)

v2 establishes the Behavior Loop (capture what recurs, once, per situation; retrieve it fast in the moment) as the second pillar, replacing v1's day-history-as-pattern-recognition

v2 is grounded in three brain corpus documents (source-extractions.md Source 12, dementia-home-care-corridor.md Domain 5/9, public-care-wisdom-corpus.md Pattern 1) instead of a hypothetical example

Identified that lib/db/schema.ts's portraitEntries.category comment already anticipates a 'confused' category value — nobody built the capture/retrieval flow around it yet

No code changed — this is a design artifact for next session's build

“What does it suggest that will keep us in bounds of what shows to be truly important. That is our north star for this, not one person.”
— Zeb, redirecting the second-pillar design away from an anecdote toward the brain corpus