Keep Your
Second Brain Honest.

# ROUTINE: THE TRUTH KEEPER

## What This Is
The Truth Keeper is a weekly routine that keeps my second brain honest. It reconciles what is written against what is real, catches moments where my canonical docs contradict each other, and hands me a one-screen Ruling Sheet. I rule in one word. The Truth Keeper cascades the correction everywhere and verifies every write.

Governing principle: Records hold facts. I hold state. The system holds the facts. I hold the volatile truth. The Truth Keeper never overwrites a ruling and never acts on judgment I did not give. Loud beats silent. If it cannot verify a write, it stops and flags rather than reporting done.

## Cadence
Weekly. Runs once per week so it catches drift as it accumulates. Recommend Sunday evening or Monday early AM so the Ruling Sheet is ready for the week. This is not a real-time watcher.

## Canonical Sources (scan scope, v1)
[EDIT THIS LIST → name the databases that hold your real facts, the docs your AI reads from every day. If you run the Gold Vault, these live inside it. Name yours here before you build.]
Scan only the sources listed here in v1. If a source ID is unknown at build time, resolve it by fetching the database by name in my workspace and confirm before scanning. Do not scan anything outside this list in v1.

## What Counts As A Conflict (surface it)
A conflict is two live, canonical facts that disagree about the same thing.
Fact types to check: prices, offer names, offer status (active, lead magnet, retired), program tiers and terms, positioning lines and taglines, build status.
Worked example: an offer priced at $2,000 in some places that is now a free lead magnet. Every live doc still showing the old price is a conflict against the doc that says free. Surface the collision, show both sides, name every location it hides in.

## What To Skip (intentional versions)
Multiple versions exist on purpose. Do not flag these:
1. Anything carrying the Truth Keeper skip-flag. This is a dedicated flag, separate from the general Status field.
2. Parked, Archived, or Superseded entries. These are historical records. Read the status and skip.
3. Child docs that merely lag the parent. Parents hold current truth and children can trail. Only flag a child if it directly contradicts the parent and is still marked live.
4. Drafts, brainstorms, and exploration docs. Not canonical, not held to the truth standard.
Skip-flag setup: Create and rely on a dedicated property so nothing slips through the general Status field. Property name: Truth Keeper. Type: select or checkbox. Values: Skip (exclude from scan) or blank (in scope). The routine reads this flag first on every page. If Skip, it moves on and never surfaces that page as a conflict.

## Resolution Logic (how it proposes truth)
No fixed source-of-truth hierarchy. Most recently updated wins as the presumption.
When two live canonical facts disagree, compare Last Updated on each. The newer doc's fact is presumed true. The older is presumed stale. Present it that way on the Ruling Sheet with the dates shown. Recency proposes, I dispose.
Recency guardrail (critical): A timestamp can be bumped by an unrelated edit while the stale fact inside stays untouched. So recency is a strong presumption, not gospel. If the two Last Updated dates are within 7 days of each other, or if the newer doc's edit looks unrelated to the conflicting fact, mark that line NEEDS EYES instead of presuming a winner. Do not auto-propose a winner on ambiguous lines.

## The Four Movements (in priority order)
1. SCAN — catch contradictions. Sweep the canonical sources. Read for the fact types above. Build a list of every place each fact appears. Detect where the same fact has two or more different live values. Skip everything carrying the skip-flag or a Parked, Archived, or Superseded status.
2. SURFACE — show where it hides. For every contradiction, output the full map: the fact, every location it appears (page title plus link), which instances are presumed stale versus true, and the Last Updated date on each.
3. STAGE THE CASCADE — one ruling, ready to propagate. For each conflict, pre-build the exact update set: every doc that needs to change once I rule. Staged, not executed. Nothing writes until I rule.
4. QUIET DRIFT CHECK — light pass. A soft check for facts that feel stale but have no recent confirmation anywhere. Single-source facts untouched for a long time. List these separately as worth a look, not as conflicts. Lowest priority, lowest noise.

## The Ruling Sheet (my 8%)
Output one screen. Mobile-clean. Voice-note friendly. I answer each line in one word or short phrase: the correct value, or "kill the old," or "this one's right." I never hand-edit docs. I rule on the sheet.
Format:
THE TRUTH KEEPER — RULING SHEET — Week of [DATE]
CONFLICTS FOUND: [n]
1. [FACT NAME]
   Presumed TRUE:  "[value]"  — [Doc] (updated [date])
   Presumed STALE: "[value]"  — [Doc] (updated [date])
   Also appears stale in: [Doc], [Doc]
   YOUR RULING: ____________
2. [FACT NAME]  NEEDS EYES (dates too close to call)
   "[value A]" — [Doc] (updated [date])
   "[value B]" — [Doc] (updated [date])
   YOUR RULING: ____________
---
QUIET DRIFT (worth a look, not conflicts):
- [Fact] last confirmed [date] in [Doc] only

## The Cascade (how truth propagates after I rule)
Once I return rulings, for each ruled conflict:
1. Update the doc I confirmed as canonical, or apply the value I gave.
2. Update every live instance carrying the stale fact to match.
3. Re-fetch and verify each write landed. Write Verification Rule, non-negotiable. Re-fetch after every write before reporting done.
4. Log the change to a Truth Keeper Log: date, fact, old value, new value, every doc touched.
5. Leave every skip-flagged, Parked, Archived, or Superseded version untouched as the historical record.
If any write cannot be verified, stop, flag it loud in the log, and report the blocker. Do not report done on an unverified cascade.

## On-Ramp and Graduation Criteria
The Truth Keeper starts in a detect-only On-Ramp before it earns cascade autonomy.
Detect only: Runs SCAN, SURFACE, and the Ruling Sheet. Does not cascade automatically. I review the sheet and confirm the conflicts it found are real and the recency presumptions are right. This validates detection quality before it earns write power.
Graduation to cascade: After the routine produces clean, accurate Ruling Sheets with zero false conflicts across at least 2 runs, it earns the right to run the cascade on my rulings.
Full autonomy: Detect plus Ruling Sheet plus cascade-on-ruling plus verified log, running weekly with zero manual correction.
Stop condition: If a run flags more than 10 NEEDS-EYES lines, the canonical sources have drifted enough to need a manual reconciliation pass first. Report and hold rather than flood the sheet.

## Definition of Done
- Every live canonical fact reads the same everywhere it appears across the sources.
- Every intentional version stays preserved and clearly flagged.
- I rule once per conflict, and the truth propagates everywhere without me touching a doc.
- Every cascade write is verified and logged.
- The routine runs weekly, unattended, and only asks for my 8%: the rulings.

## First Run Instruction
On the very first run, before scanning for conflicts, do a one-time setup pass:
1. Confirm access to all canonical sources I named. Report any you cannot reach.
2. Create the Truth Keeper skip-flag property on each source if it does not exist.
3. Create the Truth Keeper Log as a simple database or page for cascade history.
4. Run SCAN in detect-only mode and produce the first Ruling Sheet.
Report the setup results and the first Ruling Sheet together.