Session Handoff — Obsidian brain audit (Fable edition) delivered; awaiting fix-batch approval

Where it started

Sondra asked whether to split her Obsidian into separate "brains" (per Nate Herk video youtu.be/hQvwMj7IJe4) and wanted a full audit of the vault + a plan to make it PROACTIVE — "10x–100x my memory." She explicitly wanted the same methodology as the July 2 AIOS context audit (interview → scorecard → map → report BEFORE touching anything → ~10 ranked fixes → approved batch). Answer locked early: ONE vault, segmented inside, never separate vaults (that's what Nate actually does; splitting would blind the AIOS).

Decisions locked + what shipped

• ONE brain, never split — Nate's transcript confirms his "multiple wikis" live inside one project. Decision recorded in PRD.
• Interview answers (drive everything): pain = passive + messy/overwhelming; nothing sacred; moderate aggression; NEVER delete — suggest-only, she approves every removal; goal = proactive recall ("I never hear anything about it... want it to say 'I saw this in a training, so we should do this'").
• PRD written — c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Obsidian Brain Reorg\PRD.md (audit findings, target structure, 5 phases, §8 open decisions, Gap Check, DoD).
• Audit prompt written (July-2 methodology ported to the brain) — c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Obsidian Brain Reorg\FABLE-BUILD-PROMPT.md.
• First audit pass (Opus) delivered, then Sondra switched model and demanded a Fable re-run. Fable report supersedes: c:\Users\srver\AIOS-Workspace\aios-starter-kit\outputs\workspace-audit\brain-audit-2026-07-05-fable.html (first pass kept: ...\brain-audit-2026-07-05.html). Both opened in Edge.
• Fable verdict 3.5/10 utilization. Three new findings vs first pass: (1) 540 of 792 watch-log entries say "Transcript: Not available" while 549 transcripts EXIST in raw/youtube/transcripts/ — backfill scripts never re-linked the labels, and daily capture still writes false labels; (2) NO read path exists — 20 scripts + /prime all write-in or read state (dashboard/open-questions) only; nothing ever reads learning/ or people/; (3) true counts ~2x what the brain believes (792 videos, Nate 77 / Chase 54 / Luke 39 vs profile's "415, Nate 49"); plus stray empty Sondra Brain/wiki/raw/_processed/meetings/ tree.
• Top-10 fixes ranked in the Fable report; ordering doctrine = voice → integrity → volume. Recommended first batch: 1 (recall block at /prime) + 2 (re-link 540 false labels + patch capture scripts) + 3 (rebuild _master-index + self-updating rule) + 6 (goals/ page).
• ZERO vault files changed — both audits read-only. The gate is Sondra approving a fix batch.

Key files for next session

• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Obsidian Brain Reorg\PRD.md — read first; the whole plan
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Obsidian Brain Reorg\FABLE-BUILD-PROMPT.md — Phase E = how to execute the approved batch (git snapshot first, one fix at a time, verify each)
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\outputs\workspace-audit\brain-audit-2026-07-05-fable.html — the authoritative audit + ranked fixes
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\handoffs\workspace-context-audit-2026-07-02.md — the July 2 AIOS audit this mirrors (her preference: run approved fixes one at a time, reporting after each without stopping)
• Memory files touched: none

Running state

• Background processes: none (5 background find/ls shells from this session all completed)
• Dev servers / ports: none
• Open worktrees / branches: none — everything on main; PRD/prompt/reports are uncommitted new files

Verification — how to confirm things still work

• ls "c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Obsidian Brain Reorg" — expect PRD.md + FABLE-BUILD-PROMPT.md
• Open outputs\workspace-audit\brain-audit-2026-07-05-fable.html in Edge — scorecard, D2 map SVG, 10 fix cards render
• git status — vault (Sondra Brain/) shows no changes from this session; only the new project/report files are untracked

Deferred + open questions

• Open (THE GATE): which fix batch to run — Fable's rec 1+2+3+6, all 10, or her own subset. Nothing executes until she says.
• Open: fix 6 (goals/) needs ~15 min with her to confirm 2026 targets in her words — draft from context/strategy + wiki/memory/ikigai-clarity-2026-06.md, she approves.
• Open (suggest-only, her approval): retire dead wiki/canvas/command-center.canvas pointer; remove empty wiki/raw/ tree; archive triaged inbox piles (move to raw/_processed/, never delete).
• Deferred: render PRD as HTML + register report/project on CEO Dashboard (offered twice, unanswered).
• Deferred: commit the new project files + reports to git.
• PRD §8 decisions (rec'd but unconfirmed): sources-log index instead of 549 per-video pages; personal/faith content routed to a personal shelf during digest.

Pick up here

Ask Sondra which fix batch to approve (rec: 1+2+3+6). On her go: git commit the vault snapshot first (Phase E, Phase 0), then execute fixes one at a time per FABLE-BUILD-PROMPT.md, reporting after each without stopping.