Run this audit before doubling volume, launching, or handing an operation to someone else. It scores the seven things that must hold at 10× load. Score each 0 (missing), 1 (partial/promised), or 2 (working/proven). Aim for 12+ of 14 with no zeros — a zero anywhere is the thing that breaks first.

The seven checks

1. Operation map (84.1–84.2). Trigger, inputs, steps, owner, and next constraint written down and current. Ask: could a newcomer follow it without asking you anything? Check OPERATION-MAP.md date — older than a quarter without review scores 1 at best.

2. Contract + human gates (84.3–84.4, 88.4). AUTOMATION-CONTRACT.md names trigger, input schema, bounded actions, all five failure cases, and green/amber/red steps. Red-line decisions (89.4) are technically enforced — missing scopes or gates, not just written promises. Simulate one unexpected reply: does it reach a human with context?

3. Quality standard (89.1). QUALITY-STANDARD.md has checkable gates, a sampling rate, a named auditor who isn't the producer, and a rejection log with causes. Evidence: last month's audits completed on schedule with at least one corrective action recorded. Zero rejections forever means nobody's checking.

4. Cost guardrails (89.2). Measured unit cost per completed job, per-workflow caps, 50/80% alerts routed to an owner, rate limits, and a tested kill switch. Multiply by 10×: does the result fit the budget? If unit cost is a guess, score 0.

5. Permissions (88.3). Every actor on least privilege with dedicated credentials, test/prod split, vault locations documented, revoke rehearsed. Test: "if this credential leaked tonight, what's the worst it could do?" Anything outside the job description fails.

6. Runbook + bus factor (89.3). One-page runbook exists and a backup has passed the away-test in the last 90 days: found the logs, judged health, located the stop switch in under five minutes. Bus factor 2+ on every critical flow.

7. Review cadence (88.5, 89.5). A dated OPERATING-REVIEW.md or AUTOMATION-REVIEW.md within the last 30 days with a verdict and exactly one owned experiment. No review, no scale — drift compounds faster than volume.

Scoring and verdict

| Area | 0 / 1 / 2 | Evidence (file + date) |

|---|---|---|

| Map | | |

| Contract + gates | | |

| Quality | | |

| Cost | | |

| Permissions | | |

| Runbook | | |

| Review | | |

| Total /14 | | |

• 12–14, no zeros: ready to scale. Proceed with one experiment per cycle.
• 8–11 or any zero: fix zeros first, then lowest scores. Re-audit in 30 days.
• Below 8: do not scale. Narrow scope or pause (Lesson 88.5's verdicts) until the foundations hold.

File the scored sheet with the decision log, and keep a copy in the OPERATIONS-PLAYBOOK/ folder beside the operation map, contract, scorecard, and review agenda. The audit's value isn't the number — it's the named next fix with an owner and a date, which becomes next week's single experiment.