Morning digest — July 27, 2026

Generated by tp3_morning_digest_web.py · 2026-07-27

Who is Mark? 3 / 9 sections complete

Pick a section, brain-dump into Bidet AI, ~5–10 min each. The agent structures your answers and ticks the pill.

✓1. Typical day ✓2. Work rhythms ○3. Health & energy ○4. People in your life ○5. Money & risk ○6. Tech preferences ○7. Goals (1m / 6m / 1y / 5y) ✓8. Working with me (meta) ○9. Privacy lines

What's coming up today

⚠️ Agenda is 37h old — may be stale. Refresh it before trusting.

Today — Thursday, May 7, 2026

🪨 Legacy Soil — current state

Production paths are narrowed to finalists "Path 3", "Path 5", and "Path 6", while "Path 1" remains the baseline. The production binder is locked as "colloidal silica," superseding "water glass" to ensure long-term archival stability. Research identifies "nerikomi-inspired" layering as the valid artisan lineage for the line. The current model targets a flat price of $495–$795 for a single "soothe stone" heirloom.

Open:

Selection of a final production path from the finalists
Decision on "bundle SKU" to return both finished stones and unaltered pearls
Green-light for porting V2 deliverables into the "Legacy Soil V3" Drive structure

🧠 TP3 Neural Stack

As of 2026-05-07, the tp3_memories_local table contains 47,138 rows and tp3_memories contains 10,867 rows. The stack is operational with TP3_USE_LOCAL_EMBED and TP3_USE_LOCAL_LLM set to 1 to utilize local Ollama processing. Conflict: Memory from 2026-05-04 cited 668,000+ rows for tp3_memories_local, contradicting the 47,138 rows verified on 2026-05-07. The "TP3 Takeout ingestion plan" is designed and awaiting authorization to proceed.

Open:

Location of the Google Takeout archive on Apex
Gemini budget allocation for the high-signal lane (estimated $50-150)
PII and family-name routing rules for private table storage
Data scope for ingestion (Drive, Gmail, MyActivity, Gemini, and Calendar)

🎙️ Bidet AI

The project is currently in a pre-build phase for the "bidet-phone" Android app centered on a "brain-dump → AI input" pipeline. The architecture defines a "four-tab brain-dump UX (RAW / CLEAN / ANALYSIS / FORAI)" with the "FORAI (anchor)" tab acting as the primary structured output. Conflict: Documentation conflicts on using "Gemma 4 multimodal" versus "Whisper-tiny via whisper.cpp" for transcription after tests on the Gemma audio path returned "Please provide the speech segment." The "v2 DRAFT" of the Cursor brief is currently "awaiting prompt lock + audio-path verification" before development begins.

Open:

sign off or push back on the architecture
iterate the four prompts in the playground using real OMI test transcripts

🔮 Oracle

The "I don't know" trigger is refined to require a measurable pause and a specific follow-up question. Oracle is converging with Bidet AI and OMI into a single system for ambient and conscious voice input. Execution is approved for the "Path B" permanent daemon and "Oracle revive (not rebuild)" efforts. Current infrastructure is in a restart loop on Apex due to port-bind failures.

Conflict: One source marks Oracle as retired while the most recent refinements treat it as an active system to be revived.

Open:

Choice between Meta or Aura glasses
Telegram vs. Slack for notification priority
Confirmation of ag.thebarnetts.info as the unified dashboard URL

📅 Calendar

*(Calendar wiring deferred — Google Calendar MCP integration pending.)*

AI this week (2026-06-29)

Three things worth your attention this week: MCP goes stateless July 28 (your twin-memory + OMI MCP servers need an audit, not a calendar reminder — start now), NLW’s weekly brief on frontier “licensing” rollouts (opaque access regimes for Mythos-class models — relevant to how Claude Code picks models), and your OMI webhook is live again after the June 22 app update silently killed Developer Settings toggles. On embeddings: we’re shadow-testing qwen3-embedding:0.6b vs nomic-embed-text on Apex with a warmup run + timed runs — no mass re-embed of 161K rows until retrieval quality says yes.

Full AI Radar →

TP3 brain status

Gemini-embedded (tp3_memories): 11,222
Ollama-embedded (tp3_memories_local): 215,131 (intentionally frozen since 2026-04-04)
Shared memory (tp3_shared_memory): 8,161
Combined brain: 234,514 rows

Effectiveness checks

🔴 freshness: freshness: newest ingest is 23021 min old (limit 720) (overnight threshold)
🔴 throughput: throughput: only 0 rows ingested in last 24h (min 5)
🔴 poller: poller: last successful run 22431 min ago (limit 15)
🟠 source_truth: source_truth: Omi API HTTP 403

Generated automatically

This page is built daily at 6:30 AM ET by tp3_morning_digest_web.py on Apex and deployed via wrangler to Cloudflare Pages. Hand-edits to this file are overwritten by the next run — to add new sections, edit the generator template.