TP3 Private Vault + People Wiki

Saturday, 2026-05-23 · private architecture + first build slice

Recommendation: build the Karpathy/Matt Wolfe wiki pattern on top of TP3, not instead of TP3. Your life streams in; TP3 preserves the raw truth; private wiki and people pages compile from it; only sanitized outputs leave the vault.

This is not an Obsidian project. It is a TP3 layer.

Decision

GO: Add a private compiled layer with people pages, topic/wiki pages, media provenance, and share policies.

GO: Student entities are first-name-only, private-only, and never linked to last names.

GO: Dashboard should show a status/link tile, not raw student content.

HOLD: Do not publish raw OMI/Bidet classroom transcripts. Use sanitized outputs only.

What This Means

Layer	Purpose	Mark-facing result
Raw vault	Immutable OMI, Bidet, mail, SMS, audio, calendar, health, files.	Truth source. Private. Queryable.
Compiled wiki	Stable summaries by topic, project, classroom unit, decision, recurring question.	"What do we know about X?" without rereading 50 raw transcripts.
People pages	Private first-class records for people. Student lane is first-name-only.	"What have I noticed about Alex this quarter?" with sources, privacy labels, and no public exposure.
Media registry	Audio/video/image/document index with transcript and provenance.	No more "where did that WAV go?"
Sanitized outputs	Clean class notes, report-card drafts, student-facing notes, public-safe summaries.	Useful outputs that do not leak the raw vault.

First Build Slice

1. Tables

Create private TP3 tables for people, mentions, wiki pages, media assets, compile runs, and share policies. Keep raw content in tp3_memories_local; link to raw rows by ID.

privatefirst-name student laneno last-name field

2. Compiler

Process a chosen window of OMI/Bidet content. Default can be last 24 hours, but a real teaching day is better for the student/person slice.

Output private drafts: student/person mentions, wiki topic updates, media references, and a run report.

3. Dashboard Link

Add a dashboard tile: "Private Vault" with status, last compile time, counts, warnings, and a link to the private vault page. No raw student content on the dashboard.

4. Sanitizer

Before anything leaves private mode, run a sanitizer: strip raw student quotes, remove source timestamps, avoid comparative/judgy language, and keep first-name-only references.

Privacy Stance

Your point is right: AI can infer identities from enough context. "First name only" is not magic cryptography. The real protection is layered:

Usability: first names stay usable for you.
No bridge: no first-last mapping table, no imported roster with full names.
Private vault: raw student content stays inside TP3/Bidet/OMI private systems.
Sanitized exits: reports, class notes, and public writing use filtered output, not raw memory.
Dashboard restraint: dashboard gets status and links, not student details.

Reports Waiting To Read

Tangent Consolidation — current master loose-end report.
Bidet AI Marketing — positioning and claims guardrails.
Instabidet Vision — relevant to the combined Bidet Quick + Instabidet direction.
Dashboard Bento v2 — current dashboard surface.

Decision Queue

Teaching-day sample: pick one recent real teaching day for the first compiler test, or use last 24 hours as a smoke test.

Student page format: first-name person page, class-period page, or both. Recommendation: both.

Default output: private teacher notes first, sanitized class notes second, report-card helper third.

Dashboard rule: status/link only for student lane. Recommendation: locked.

Next Action

Build the database layer and a tiny compiler run. Use a fixed window, write a private run report, and show counts instead of raw student content. Then decide whether the extraction quality is good enough to expand.

Sources consulted: Matt Wolfe/Future Tools summary coverage, Andrew Karpathy's LLM knowledge base gist, TP3 live schema/source counts, and existing Mark memory rules for classroom/student privacy.