The Second Source That Scored 87 Out of 100

Step 1 - Convert Cpk to a sigma distance

Cpk = 0.94 means the nearer specification limit sits 0.94 × 3 = 2.82 standard deviations from the process mean.

Signal width: USL − LSL = 25.10 − 24.90 = 0.20 mm, half-width 0.10 mm.

If the process were centered, σ = 0.10 ÷ (3 × 0.94) = 0.0355 mm.

Step 2 - Convert to defect rate

• Centered case (Cp = Cpk = 0.94): 2 × Φ(−2.82) = 2 × 0.00240 = 4,800 PPM
• Off-center case: the near limit alone gives 2,400 PPM; the far limit adds little. Total is between 2,400 and 4,800 PPM.

At 900,000 parts a year: 2,160 to 4,320 nonconforming parts a year, on a safety-critical characteristic.

Step 3 - What a normal threshold requires

Cpk 1.33 is the common minimum for a critical characteristic:

• Required σ = 0.10 ÷ (3 × 1.33) = 0.0251 mm
• Reduction needed from 0.0355 to 0.0251 = 29%
• Expected defect rate at Cpk 1.33: 2 × Φ(−3.99) ≈ 66 PPM → about 60 parts a year

The gap between 0.94 and 1.33 is a factor of 36 to 73 in defect rate. That is not a marginal shortfall; it is a different process.

Step 4 - The complication

A Cpk of 0.94 without a gauge R&R study is not a measurement, it is an opinion. If the gauge contributes a meaningful share of the observed variation, true capability could be better or worse than reported. For a 0.20 mm tolerance band, an unvalidated gauge is the first thing to fix - you cannot tighten a process you cannot measure.

Step 5 - What it means commercially

The process will produce defects. The only question is whether they are caught at the supplier, at incoming inspection, or in the field. On a safety-critical part, the third outcome is not available. That cost belongs in the qualification decision, not in a later quality discussion.

Recommendation: conditional approval for a controlled pilot, not full production release.

The 87/100 is a false comfort. A weighted average let a critical-characteristic failure be offset by strengths in unrelated areas. The score is not the decision; the decision is whether the process can make the part.

What blocks release today

1. Cpk 0.94 on a safety-critical bore - 2,160–4,320 nonconforming parts a year expected.
2. No gauge R&R - capability is unproven in both directions.
3. Open traceability corrective actions (2 of 9) - without heat-lot traceability, a defect cannot be contained to a lot. This is a recall-scope problem, not a paperwork problem.
4. Three of five lots with CoC deviations - the quality system is producing documentation that does not match the product. That is the same failure mode as the traceability gap.

Conditions for controlled pilot (weeks 1–10)

• Gauge R&R completed and accepted (%GRR under 10%, 10–30% conditional) before any capability data is used.
• A process improvement plan with a dated path to Cpk ≥ 1.33, reviewed every two weeks, with an interim gate at Cpk ≥ 1.10.
• Both traceability corrective actions closed and verified, not just marked closed.
• 100% inspection of the critical bore for all pilot volume, at the supplier's cost, with SPC charts submitted per lot.
• Containment agreement: any nonconformance is quarantined at the supplier, with notification within 24 hours.
• Certification plan for the applicable medical device standard with a dated milestone.

Conditions for volume release

• Cpk ≥ 1.33 demonstrated on three consecutive lots of at least 30 pieces each, after gauge acceptance.
• Zero CoC deviations over the pilot.
• Traceability demonstrated by an actual mock recall, not a procedure review.

On the ten-week pressure

The ten weeks will pass regardless. The question is whether at the end of them the buyer has a qualified second source or an unqualified one running at full volume. Split the volume: release no more than 15–20% to the candidate while conditions are met, keep the incumbent at capacity, and hold the remainder as the incentive. If the candidate cannot meet the interim gate, the program has ten weeks of warning rather than ten weeks of false assurance.

The uncomfortable alternative to put on the table: if the launch cannot tolerate a phased qualification, the honest answer is that this supplier cannot be the second source for this launch, and the capacity constraint must be solved another way - a third source, a design change, or a schedule change. Those are program decisions, and the buyer's job is to make the trade-off visible, not to absorb it by lowering the standard.

The defect in the current method

An additive, weighted-average score assumes deficiencies are substitutable. They are not. A supplier with excellent document control and a process that cannot hold a critical tolerance is not "87% good" - it is unqualified. Averaging converts a pass/fail question into an arithmetic one, and arithmetic can always be argued.

Restructure in three layers

Layer 1 - Knock-out criteria (pass/fail, no score)

Any single failure stops the evaluation regardless of everything else:

• Capability below the minimum on any critical characteristic (Cpk < 1.33, or < 1.10 with a dated plan).
• No valid gauge R&R on critical characteristics.
• Open corrective actions touching traceability, safety, or regulatory compliance.
• Missing regulatory certification where the commodity requires it.
• Any falsified or nonconforming certificate of conformance.

This supplier would have been stopped at the first and second knock-outs. That is the correct outcome, and it took minutes rather than a two-day audit.

Layer 2 - Scored criteria (weighted, and only if Layer 1 passes)

Reserve the weighted score for genuinely substitutable dimensions: capacity headroom, financial stability, systems maturity, responsiveness, continuous improvement culture, geographic and logistics fit. These are the things where being strong in one area legitimately offsets weakness in another.

Layer 3 - Commodity-specific overlay

Weighting should differ by what is being bought. For a safety-critical machined component, process capability and traceability dominate. For MRO or packaging, systems and responsiveness dominate. A single universal scorecard applied to every commodity is how a critical-characteristic failure gets diluted by a strong score on document control.

Additional fixes

1. Score capability from data, not from the audit. Capability studies and gauge studies are submitted in advance and validated by the buyer, not observed on the day.
2. Separate the auditor from the sponsor. If the person who needs the second source in ten weeks also owns the audit conclusion, the conclusion will bend. The audit should report to quality, with sourcing as a stakeholder.
3. Define consequences in advance. Each open corrective action carries a dated closure requirement and a stated consequence. "Open" without a date is a permanent state.
4. Re-audit on a schedule tied to risk, not to the calendar. A supplier with an open traceability finding gets a follow-up in 60 days, not at the next three-year cycle.

Test of the fix: would the restructured framework have reached the same conclusion on this supplier, and would it have reached it faster? It should - and the fact that it produces a clear "no" rather than a debatable "87" is the point.