factory-acceptance-test

Factory Acceptance Test Skill

A FAT is the last moment a defect is the factory's problem instead of yours. This skill writes the plan (or the report) with the parts that actually get argued over at the line: which spec clause each test covers, how many units get pulled and at what AQL, what exactly fails a lot, and who signs — so acceptance is a decision with a name on it, not a vibe at the end of a factory visit.

What This Skill Produces

• A test coverage matrix mapping every test to the spec requirement it verifies
• A sampling plan (ANSI/ASQ Z1.4) with lot size, inspection level, and AQL per defect class
• Unambiguous pass/fail criteria and lot accept/reject rules
• Golden-sample handling procedure
• A deviation log format and a sign-off block with authority levels

Required Inputs

Ask for these if not provided; if the spec is thin, draft the matrix from the product description and mark rows [spec clause to confirm]:

• Product and spec — the requirements document or at minimum a feature/claims list
• Lot size — units in the lot(s) under acceptance
• Product risk profile — safety-relevant? battery? affects AQL choice
• Prior quality history — first article vs mature product (allows tightened/reduced inspection)
• Who signs — customer QE, factory QA, third-party inspector?

Sampling & Classification Framework

Defect classes and default AQLs (ANSI/ASQ Z1.4, General Inspection Level II, normal inspection — the consumer-electronics defaults; adjust with reason):

| Class | Definition | Default AQL |

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

| Critical | Safety hazard, regulatory violation, data loss | 0 — any occurrence rejects the lot |

| Major | Product fails to function, or defect the user will certainly notice and return | 1.0 (tighten to 0.65 for premium/first lots) |

| Minor | Cosmetic or workmanship issue within limit samples' tolerance but noted | 2.5 (relax to 4.0 for bulk/industrial) |

From lot size + Level II, derive the sample-size code letter and accept/reject numbers from the Z1.4 tables; state them explicitly in the plan (e.g. "Lot 3,000 → code K → n=125; Major Ac=3/Re=4"). Switch to tightened inspection after 2 of 5 consecutive lots rejected; reduced only with sustained history.

Golden samples. Two signed, serialised golden samples minimum — one held at the factory line, one at the buyer. Sealed, dated, with an expiry/refresh rule (refresh on any ECO that changes fit/finish/function). Cosmetic judgement is against the limit-sample boundary set, never against memory.

Deviations. Any test performed differently than planned, any borderline judgement, and any use-as-is decision goes in the deviation log — numbered, with disposition and approver.

Output Format

FAT [plan | report]: [product] — lot [ID]

1. Scope & lot definition — product rev, lot size, factory, date
2. Test coverage matrix — table: test #, spec clause covered, method/equipment, sample basis (100% / AQL sample), pass criterion
3. Sampling plan — lot size, code letter, n, and Ac/Re per defect class
4. Pass/fail & lot disposition rules — accept / reject / rework-and-rescreen conditions
5. Golden & limit sample register — serials, locations, seal dates, refresh trigger
6. Deviation log — #, description, class, disposition, approver
7. Results (report only) — defects found per class vs Ac/Re, failure detail with photos referenced
8. Sign-off block — lot disposition, name, role, authority, date; rejection escalation path

Quality Checks

• [ ] Every spec requirement appears in the coverage matrix, or its exclusion is stated
• [ ] Sample sizes and Ac/Re numbers are explicit — not "per AQL 1.0"
• [ ] Critical defects have Ac=0 and an immediate-stop instruction
• [ ] Cosmetic criteria reference physical limit samples, not adjectives
• [ ] The report states defects per class against Ac/Re, and the disposition follows arithmetically
• [ ] Sign-off names a role with authority to reject the lot

Anti-Patterns

• [ ] Do not write "inspect to AQL 1.0" without the sample size and accept/reject numbers — that phrase settles nothing
• [ ] Do not let critical defects carry a non-zero acceptance number
• [ ] Do not judge cosmetics against memory or photos alone — physical limit samples or it will be relitigated every lot
• [ ] Do not allow rework-and-reinspect of a rejected lot without 100% rescreen of the reworked defect mode
• [ ] Do not accept a lot with an unsigned deviation in the log
• [ ] Do not let golden samples go stale across an ECO — refresh or they certify the wrong product