The Baseline That Was Too Good to Be True

The 2.1% figure is not a defect rate - it is a reporting artifact. Three defects are visible:

1. Self-inspection without verification. Operators inspecting their own parts introduces both bias and inconsistency; there is no evidence of gauge R&R or attribute agreement.
2. Incomplete defect definition. The quality system appears to count only certain defect categories. Scrap (4.3%) is logged separately and rework is not tracked at all, so the 'defect rate' excludes two of the three ways a part can fail to meet spec.
3. Coverage gap. The audit covered only day shift, so the 5.1% estimate may not represent the line. Shift-to-shift variation in stamping is common (setup, tooling wear, staffing).

Recommendation: Do not use 2.1% as the baseline. Use a defined, measured baseline built on a clear operational definition of a defect (scrap + rework + escape), collected across all three shifts, with an attribute agreement study on the inspection method. The 5.1% audit figure is a better estimate but is not yet a baseline - it is a two-day, single-shift sample.

Practical path: run a two-week data collection across all shifts using the agreed definition; validate the inspection method with an attribute agreement study; then set the baseline. If time pressure requires a number now, use the audit figure as a provisional baseline and flag it as provisional in the charter.

Separate the two issues: what is reported to corporate and what the project uses to measure improvement. They serve different purposes and can coexist, but conflating them corrupts the project.

1. Acknowledge the constraint. The plant manager has a real reporting obligation. The Belt should not ask them to falsify a corporate report.
2. Explain the project consequence. If the baseline is 2.1% and the true rate is ~5%, a '50% reduction' target is unachievable in reality but appears achievable on paper - or, worse, the team hits 1.05% on the old metric while true defects barely move. Either way the project produces no real improvement and the plant loses credibility later.
3. Propose a dual-track approach. Keep the corporate report as-is for the current period, but define the project baseline on the corrected definition and disclose the measurement change explicitly. Show the plant manager how the corrected baseline protects them: when the project delivers, the improvement is real and defensible.
4. Escalate with the plant manager, not against them. Bring the measurement gap to the sponsor/champion as a finding ('our reported defect rate excludes scrap and rework'), framed as an opportunity to improve reporting accuracy, not as an accusation.
5. Document the decision. Whatever is decided, record it in the charter so the baseline definition and any known bias are transparent to future reviewers.

The principle: a Belt must never build a project on a number they know is wrong. The reporting question is the plant manager's to own; the measurement integrity of the project is the Belt's.

Risks of the single-shift, two-day audit:

• Sampling bias by shift. Setup, tooling wear, staffing, and supervision differ by shift; day shift may be systematically better or worse. The 5.1% may not generalize.
• Short window. Two days may miss weekly patterns (Monday startup, Friday rush) and special causes.
• Inspection effect. Operators may behave differently when audited (Hawthorne effect), inflating or deflating the observed rate.
• Small sample. 400 parts gives a 95% confidence interval of roughly ±2 percentage points around 5.1% - wide enough to matter for a 50% reduction target.
• Definition drift. If the audit used a different defect definition than the quality system, the two numbers are not comparable.

Data collection design:

1. Stratify by shift - collect across all three shifts, with roughly equal sample sizes, so shift effects are visible rather than averaged away.
2. Extend the window to at least two full weeks to capture weekly patterns and enough parts for a stable estimate.
3. Fix the definition first - write an operational definition of defect (scrap, rework, escape) and train all inspectors to it.
4. Validate the measurement system - run an attribute agreement study (≥30 parts, ≥3 appraisers, known-good/known-bad) before trusting any rate.
5. Use independent inspection for the audit where feasible, or at minimum blind the inspectors to which parts are audited.
6. Log by shift and by day so the baseline can be checked for stability (run chart) rather than reported as a single average.

The goal is a baseline that is defensible, not merely available.