Morning digest — April 28, 2026

Generated by tp3_morning_digest_web.py · 2026-04-28

What's coming up today

Today — Monday, April 27, 2026

🪨 Legacy Soil — current state

The "Yushunxin SXYZ-600" is the identified hardware pick for the pearling process, estimated at "~$1,500 landed" via Alibaba Trade Assurance. Production finalists are narrowed to "Path 1", "Path 3", "Path 5", and "Path 6", with "colloidal silica" confirmed as the primary binder. "Path 6" is projected to reach an "~80mm" diameter through a "layered core-build" of "bimodal" bone meal. Finishing is standardized as a "Paraloid B-72" consolidation dip followed by a "tung oil + beeswax" hand-rub.

Open:

• Spending go-ahead for the "Yushunxin SXYZ-600" granulator
• Choice of first-batch intent: single-color plain pearls vs. pigmented pearls
• Decision on the "blended option, a hybrid option" combining Path 1 and Path 6
• Choice between cremains-loaded resin or cremains-concrete for the worry-stone base

🧠 TP3 Neural Stack

Current row counts as of 2026-04-28 are 10,867 for "tp3_memories" and 657,494 for "tp3_memories_local". Metadata extraction is currently routed through Ollama llama3:8b following the resolution of Gemini quota exhaustion. The "Oracle" spoke system is in a revival phase after an audit identified port-bind failures at :8002 and :8000. Google Drive backfill remains paused.

Open:

• Relationship between Oracle and Bidet AI (rename, new surface, or Bidet evolution)
• Location of the 100GB Google Takeout archive on Apex
• Gemini budget allocation for Takeout ingestion (estimated $50–$150)
• PII and family-mention filtering rules for Takeout processing
• Selection of Takeout data scope regarding Photos, Maps, and YouTube

🎙️ Bidet AI

Mark decided to fork the personal honest-answers codebase into a separate "public app fork" for friends at bidetai.app. The project transitioned to a "browser-only state" using IndexedDB to store keys and "custom AI prompt tabs" locally. This distribution uses a "Bring-Your-Own-Key + optional ~$2 buy-me-a-coffee" model and is "NOT a business venture." The personal version at bidetai.thebarnetts.info remains frozen. Conflict: Memory lists the domain registrar as both Namecheap and Cloudflare Registrar.

Open:

• Sign-off on proposed hosting via Cloudflare Pages, Google Cloud Run free tier, and Cloudflare R2
• Confirm if friends require a "hybrid free tier" with server-side keys before merging PR 3
• Choose a pronoun for the "It/I/We cleans up your mess" tagline

🔮 Oracle

Recent audits confirmed Oracle is in a severe restart loop with zero functional throughput due to port-bind failures on :8002, :9292, and :8000. Mark selected "Path (c) — audit code first, then decide" and approved the "Oracle revive (not rebuild)" plan to restore the hub-and-spoke architecture. The "I don't know" trigger is refined to require a specific pause and follow-up question to prevent over-answering. The system targets a Phase 3 end-state using Meta or Aura glasses for the reply channel.

Open:

• Clarification on whether Oracle is a rename of Bidet, a new surface, or Bidet's evolution
• Selection between Meta and Aura glasses (deferred until hardware release)

📅 Calendar

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

TP3 brain status

• Gemini-embedded (tp3_memories): 10,867 (legacy; no new writes since 4/22)
• Ollama-embedded (tp3_memories_local): 659,590 (active — all new ingest lands here)
• Shared memory (tp3_shared_memory): 67
• Combined brain: 670,524 rows

Effectiveness checks

• ✅ freshness: freshness: newest ingest 42 min ago (overnight threshold)
• ✅ throughput: throughput: 2109 rows in last 24h
• ✅ poller: poller: last run 1 min ago
• 🔴 source_truth: source_truth: Omi has memories 11441 min newer than TP3 (limit 30)

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.