| id | GUIDE-AUDIT-PATH-001 | ||||
|---|---|---|---|---|---|
| title | Audit Path (15 Minutes) | ||||
| doc_type | guide | ||||
| status | published | ||||
| audience | auditors, reviewers, platform-engineers | ||||
| tags |
|
||||
| stories |
|
||||
| requirements |
|
||||
| acs |
|
||||
| adrs | |||||
| last_updated | 2026-01-07 |
This repo optimizes for trusted change: large diffs are normal; review happens via bounded scope + receipts, not line-by-line diff reading.
The local gate is canonical. GitHub Actions are optional/advisory.
nix develop
cargo xtask selftestIf selftest is green, the governance contracts are intact. That's the first trust anchor.
cargo xtask check # fmt, clippy, unit tests
cargo xtask docs-check # version alignment, doc_index syncThese validate that documentation matches code and specs match implementation.
Start with the curated exhibits in:
docs/audit/CASEBOOK.md— Best examples of governed changedocs/audit/FAILURE_MODES.md— What went wrong and how we hardened
Every PR should have a Cover Sheet section with:
- Scope map — what changed (by directory/domain)
- Receipts links — paths to gate outputs, test results, evidence
- Known limits / errata — what's incomplete or was wrong
- Reproduce command — how to verify locally
See docs/audit/PR_COVER_SHEET.md for the canonical format.
This repo follows explicit rules about what can be claimed:
A PR body can state:
- Absolute measurements + links to receipts
- Gate outcomes + links to receipts
- Bounded known limits
A PR body cannot state:
- Multipliers ("10x faster")
- "Enterprise-grade" / "production-ready" without a receipt
- Quality claims without falsifiable evidence
If something was wrong, we record:
- What — the incorrect claim or behavior
- How detected — gate, reviewer, receipt that caught it
- Fix — commit/PR link
- Prevention — new gate/contract/test added
See the Errata section in PR cover sheets.
| Receipt Type | Location | Purpose |
|---|---|---|
| Gate outputs | .runs/pr/<n>/<run-id>/receipts/gate.json |
Gate pass/fail evidence |
| Economics | .runs/pr/<n>/<run-id>/receipts/economics.json |
DevLT + compute tracking |
| Dossiers | .runs/pr/<n>/<run-id>/receipts/dossier.json |
Structured PR analysis |
| Exhibits | docs/audit/EXHIBITS/PR-<n>.md |
Version-controlled cover sheets |
.runs/ is ephemeral (gitignored). docs/audit/ is durable (committed).
# Environment health
cargo xtask doctor
# Quick smoke test (validates template baseline)
cargo xtask kernel-smoke
# AC coverage (what's tested)
cargo xtask ac-status
# Full governance (12 steps)
cargo xtask selftestLocal gate is canonical. We do not rely on GitHub UI checkmarks.
cargo xtask selftestis the final arbiter- Receipts generated by xtask are the evidence
- PR cover sheets link to receipts, not prose claims
If you see "checks passed" language without receipt links, that's a red flag.
Mini-walkthrough for generating and publishing PR cover sheets:
# 1. Run gates and generate gate receipt
cargo xtask receipts-gate --pr 123
# 2. Record economics (time, compute, iterations)
cargo xtask receipts-economics --pr 123 \
--author-minutes 30 --author-confidence estimated \
--compute-usd 5.00 --compute-confidence estimated
# 3. Generate cover sheet from receipts (preview)
cargo xtask pr-cover --pr 123
# 4. Update PR body and save exhibit to docs/audit/EXHIBITS/
cargo xtask pr-update --pr 123 --save-exhibit
# Or dry-run first to see what would change
cargo xtask pr-update --pr 123 --dry-runflowchart LR
PR[PR Work] --> RG[receipts-gate]
RG --> gate[receipts/gate.json]
PR --> RE[receipts-economics]
RE --> econ[receipts/economics.json]
gate --> PC[pr-cover]
econ --> PC
PC --> cover[Cover Sheet MD]
cover --> PU[pr-update]
PU --> exhibit[exhibits/PR-N.md]
exhibit --> CB[CASEBOOK.md]
CB --> backlog[Backlog/Issues]
Key points:
receipts-gateruns validation gates and emitsreceipts/gate.jsonreceipts-economicsrecords DevLT + compute inreceipts/economics.jsonpr-covergenerates a cover sheet from receipts (stdout or file)pr-update --save-exhibitupdates the PR body AND saves todocs/audit/EXHIBITS/- Exhibits feed into the casebook and inform future work
- Review a PR: Look for the Cover Sheet block and receipt links
- Audit the factory: Read
FAILURE_MODES.mdto see what's been hardened - Validate claims: Run the reproduce commands yourself