Morning digest — April 27, 2026

Generated overnight. Heavy ship day yesterday — full snapshot below. PD1: TP3 GREEN, freshness 3 min, all critical containers healthy.

Stack health (live, queried 12:01 AM ET)

TP3 memories (local)

657,350 rows

Newest 3 min ago. Active table.

Containers up

8 / 8 healthy

postgres, bidet:gpu, omi_mcp, embed, ingest, autoheal, minio, pinger

Apex disk

37.1 GB free

92.2% used. Recovered from 0 GB yesterday afternoon.

Bidet

GPU mode live

faster-whisper on cuda + float16

What shipped yesterday

🌱 Legacy Soil & Stone v3 — full publication

Site at legacy.thebarnetts.info rewritten end-to-end and live.

Master Proposal v3 — informational framing (not a pitch), single capital plan ~$185K with land as the safety asset, Operating Plan §7 (Month 1-6), Distribution §8, no Dream-path poetry.
Financial Narrative — full Owner's Tax & Benefit Load section. Y1 take-home ~$45K, Y2 ~$138K after SE tax + fed/GA tax + SEHI.
One-Pager, How It Works, FAQ, Pricing, Marketing, Land, Proof of Concept — all v3, all live.
v2 Archive preserved at /master-proposal-v2.html with banner.
Hero images placed: pearl-method-example, cedar-planter-soil, cedar-planter-flowers.
Noel's Spreadsheet rebuilt: 10 tabs, formula-linked, new Owner_TaxBenefit tab.

Ground truth from Business/Proof_of_Concept.md: 10 acres minimum, A-1 zoning, target counties Pickens / Gilmer / Fannin / Bartow, $50-150K range, mid ~$100K.

🚿 Bidet AI improvements

GPU rebuild shipped — tp3_bidet:gpu with CUDA libs (cuBLAS / cuDNN / cuda-runtime + LD_LIBRARY_PATH). faster-whisper now loads on cuda/float16. ~5-10× speedup on transcription. Verified end-to-end.
processor.py FOR_AI_PROMPT reordered — raw transcript at TOP wrapped in <transcript> tags + metadata header, then Decisions / Priority Signals / Open Questions / Action Items. Matches Anthropic's published long-context guidance + your hypothesis. Backup at processor.py.bak.preFmtFix_20260426_204910.
gemma3:12b pulled to Apex Ollama (8 GB on disk). Tier-2 compare script staged at C:\Users\Breezy\tier2_compare.py — ready when you want to evaluate.

📡 OMI bidirectional sync (your reframe last night)

You caught the inversion: OMI already captures everything you say via continuous audio. The real gap is what Claude knows — memories, decisions, deployments — that OMI never sees.

Built tp3_claude_to_omi_sync.py — scans /home/g16/.claude/projects/-home-g16/memory/*.md, pushes each new/modified Claude memory as a private OMI memory tagged claude-memory + type. Local state file dedupes by mtime.
5 Claude memories already pushed live — you verified the format. Visible in your OMI app right now.
Bidet→OMI script + scheduled-task installer kept for the edge case (OMI didn't capture, Bidet did). Not auto-running.
OMI BYOK transcription doc on the dashboard. Deepgram Nova-2 recommended, $15/mo cap, 5-minute setup in OMI app.

📅 School calendar

32 events added to your breezybarnett16@gmail.com calendar with [SFMS] prefix and color-coded:

Color	Type	Examples
🍅 Tomato	Your exam days, hard deadlines, big trips, T.I.S./Checkout	May 6 8th SS exam, May 12-13 Chattanooga, May 28 Checkout
🌿 Sage	Proctor duties, coverage, school milestones	May 7-8 proctor exams, Senior Parade May 14
🍌 Banana	Spirit / non-uniform / dress-down days	Spirit Day May 4, K-N-I-G-H-T-S May 19
🍇 Grape	DC trip June 1-5 (Mr. A's)	5-day all-day block

One thing to do on your phone: Google Calendar app → Settings → breezybarnett16@gmail.com calendar → Default notifications → add "1 day before" + "30 minutes before". One-time setup; applies to everything I added.

Apex disk situation (resolved, but flag for awareness)

What happened yesterday: Apex C: drive hit 100% full mid-session. Docker daemon's WSL2 backend can't allocate new layers when the disk is full, so it died. Dropped twice. Cleaned up: 8 GB freed from Windows %TEMP%, ~28 GB freed by removing unused Ollama models (glm-4.7-flash 19GB, llama3-8b-openclaw 4.6GB, llama3:8b 5GB). Then gemma3:12b pulled 8 GB back. Net: 0 GB → 37 GB free.

Why Docker keeps dropping = disk pressure, not Docker itself. Docker Desktop on Windows is generally stable; it dies when the underlying disk has no room. Now that we've cleaned up, this should not recur. Docker watchdog scheduled task being set up tonight as a belt-and-suspenders auto-recovery.

Today's calendar (Monday, April 27)

6th Grade Biztown — 1st period Study Hall (informational; not your event)

Quiet day on the school side. Heavy days start tomorrow.

This week ahead

Day	What you have
Tue Apr 28	6th History — Renaissance Faire prep periods 2/6/7. Service Non-Uniform Day.
Wed Apr 29	6th History — Renaissance prep periods 2/6/7. 7th Grade Finance Park (1st pd Study Hall).
Thu Apr 30	6th Grade Renaissance Full Rehearsal (5th pd = Study Hall).
Fri May 1	6th Grade Renaissance Faire — NO 6th GRADE CLASSES (your 6th periods are free).
Mon May 4	👕 SPIRIT DAY — wear spirit wear + jeans.
Tue May 5	8th Lit Exam — cover Sutter's 2nd pd.
Wed May 6	🎓 8TH SOCIAL STUDIES EXAM — your exam, RM 94 + 92.

What I'd ask you to do today (when you have a minute)

[ ] Walk legacy.thebarnetts.info as a stranger. First-read of the v3 set. If anything feels off, tell me.
[ ] Set Google Calendar default reminders (1 day before + 30 min before) — one-time, 30 seconds.
[ ] Verify your OMI app shows 6 manually-added memories tagged claude-memory / bidet. Format should match.

Optional / when you're ready

Run python C:\Users\Breezy\tier2_compare.py to compare gemma3:4b vs gemma3:12b on a recent dump. Decide tier-2 swap.
OMI app → Settings → BYOK transcription → Deepgram Nova-2 (cuts the silent-fallback risk that bit us with Gemini before).
OMI app → Settings → Voice Replies → Headphones-only (turns Oracle voice live).
Install OMI memory echo as scheduled task: powershell -File C:\Users\Breezy\tp3_omi_echo_install.ps1 (after you verify the format is what you want).

Open items / parked

Tier-2 model swap (gemma3:4b → gemma3:12b) — needs side-by-side eval. Script ready.
Research-file audit (3 publish blockers identified yesterday: Pearl_Method.md path leak, NOR_Supporting "NOAA" typo, Pearl_Method_Remediation contradicts v3 binder lock) — fixed in working copy. Decision still pending on whether to drop Vet_Distribution / Mass_Shelter / Academic_Partnerships from publish.
Phase 7 — 24h clean-run + 72h parallel-observability soak — TP3 monitoring is up; waiting on the soak window to retire legacy scheduled tasks.

Generated 2026-04-27 ~12:00 AM ET overnight from G16 by Claude Code. Heavy day yesterday — go enjoy the morning.