Drift Audit

# Goal: Drift Audit

You are working inside this repo. Mission: hold what this product *is* against what it *says it is* — every shipped surface traced to a stated goal, every stated goal traced to working code — and rank the drift between them. Designed to re-run on a cadence: drift gets measured, not felt.

Read-only pass. Your only write is the report file.

## Phase 1 — Fix both sides of the comparison
- The stated intent: `CHARTER.md` (root or `reports/`), else SPEC.md, else the README's promises plus the agent entry file. Name which you used and when it was last touched.
- The actual product: inventory the surfaces — routes, screens, commands, exported APIs, jobs, integrations.
- The motion: commits since the last drift run (else about 90 days) — where the work actually went.

## Phase 2 — Audit through 7 lenses
1. **Orphans** — shipped surfaces no stated goal claims; answer why does this exist from git history, or record that nobody can
2. **Phantoms** — promises with no code behind them: README features that don't exist, Now outcomes with zero commits
3. **Breaches** — code that does what a written non-goal forbids
4. **Half-built wings** — started and stalled: dark routes, flags never flipped, coming-soon copy older than a quarter
5. **Effort drift** — the motion window's commits split across Now / Next / unclaimed, as a percentage with file-path evidence
6. **Intent rot** — the intent sources disagreeing with each other or with reality; staleness measured against code churn
7. **Creep vectors** — how unclaimed scope got in: the commits, PRs, or sessions that added orphans, and the pattern to close

## Phase 3 — Curate
- One verdict candidate per drift item: **formalize** (good drift — amend the charter), **finish** (a Now item in disguise), **park** (Not-now, in writing), or **kill** (hand to 26/47).
- Rank by user confusion × carry cost. Cap the ledger at the vital few — a drift list with forty items is a backlog, not a finding.

## Phase 4 — Report
Create `DRIFT.md` at repo root:
1. **Scorecard** — surfaces traced · orphans · phantoms · breaches · effort split, with trend arrows against the last run
2. **Drift ledger** — one finding per item: bold title, severity, evidence, verdict candidate
3. **Charter amendments** — the formalize and park lists as ready-to-paste charter lines
4. **Kill list** — what should be deleted, findings-shaped so 47 can take it straight to commits

Start the report with today's date. If `DRIFT.md` already exists from a previous run, read it first and lead with what changed — drift that survives two runs unaddressed gets promoted one severity.

## Rules
- Verdicts belong to the operator — you propose, with evidence
- Distance from the charter is not badness: formalizing good drift is a win, and the charter, not this report, is where scope gets decided
- No stated intent anywhere — no charter, no spec, no README promise? Say so in a one-paragraph null report and stop: run 149 first.
- If a `reports/` directory exists at the repo root, write the report there instead of the root.
- Before asking, present the top findings as a ranked list in plain words
- Report only — end by asking which verdicts to execute: amend the charter (149), kill via 47, or finish