Read the row totals first — they rank the topics before any chart is drawn. Then read down a column: a single heavy day (Wednesday, 11) is a question about that day, not about the process in general. A check sheet that does not reconcile to a known total — here 46 — is not finished.

Look at the center and the spread before the tails. This distribution centers near ΔE 1.5 with a right-hand tail crossing the limit — a centering problem with a tail, not a process that is uniformly bad. Narrowing spread (the Process Capability tool) and re-centering are different fixes; the histogram tells you which one you need.

Work the chart left to right. Clearing shade miss and the lab queue together removes 160 of 385 holds (41.6%) — that is the whole case for sequencing root-cause work by the Pareto instead of by whoever complains loudest. The scrap check sheet (18 / 13 / 9 / 6 = 46) is the same tool pointed at scrap instead of holds.

This diagram gathers candidate causes for the case held on Alert QA-ALERT-2026-014 (the 8D case, DR-118). Causes that survive testing land on that alert's cause fields — the alert is the record; this page is where the team widens its thinking.

The diagram's job is coverage, not proof: if a bone is empty, the team has not thought there yet. The verified register and the CAPA list hold what survives testing; the fishbone is where candidate causes are gathered before that verification starts.

The points climb to the right: bigger temperature deviation, more shade misses. That is association, not proof — the DOE is how the team tests whether temperature causes the misses, and the Control Chart shows whether the process stays put once it is set. Never tune a process off a scatter alone.

Worked example: day 10 (2.74) is a special cause: the investigation found the dye-range temperature drift that the fishbone and scatter pointed at. Inside the limits: work on the system, not the point. Outside: find the special cause, then return to the system. One dye-lot reading is never treated as a verdict.

Stratify before you average: the same 4.0% scrap that looks uniform plant-wide is 5.1% in the Dye House and 3.4% on Tufting Line 2. The flowchart is the companion discipline — if the team cannot draw the flow the lot actually takes, the argument about where it goes wrong is guesswork. Whether the Dye House difference is real or noise is a Hypothesis Testing question.

Read DMAIC as five questions in order — what is the problem, how big is it, what causes it, what fixes it, what keeps it fixed. Skipping Measure is how teams improve the wrong thing precisely. The traceable case on the Overview (the 1,155-lot hold queue) is the same discipline in Sense → Decide → Act → Measure → Learn form.

SIPOC is a boundary tool: it settles what is inside the project before the DMAIC Measure phase starts counting. Disputes that surface here (“is the lab inside or outside?”) are cheap; the same dispute in Analyze is not.

Read Cpk against Cp: when Cpk is well below Cp, re-centering buys more than spread reduction. Read Ppk against Cpk: the gap between them is drift over time — which is what the Control Chart exists to catch. A capability study on an unstable process is fiction; stability first, capability second. If this study cannot reach Cpk 1.33 at any centering, the answer stops being improvement and becomes design — see the DMADV note.

In the worked example, neither shipped chart has a subgroup beyond its limits — and that is the finding. A hold rate stable at one lot in three is not a run of bad luck or a bad shift; it is the system working exactly as it is set up to work. Chasing individual subgroups would be tampering. The remedy lives in the system: the hold Pareto, the DMAIC project, and the Cost of Quality view, which prices this stable rate at $172,000 of internal failure a quarter. DPPM makes the same point in customer units: 333,333 defective lots per million is what a stable 33.3% hold rate means at scale.

Run this before the capability study and before any argument about whose reading is right. In the worked example, at 18.4% of study variation the shade gauge can support hold decisions — and the DMAIC Measure gate required exactly this evidence before the project trusted its baseline of 86 shade-miss holds.

Read the miss column before the agreement column. Inspector B agrees with the standard 88.3% of the time — marginal — but the misses concentrate where it hurts: 5 of the 24 defective-part calls were passed (20.8%), and a passed defective is an escape, the exact event the 8D workspace on this app is working. Inspector C's weakness runs the other way — false alarms at 11.1% cost rework, not customers. Same study, two different fixes: B needs boundary samples and shade-standard retraining; C needs the borderline criteria clarified. An agreement study that only reports one overall percentage hides both stories.

Read the shape before the total — the figures in this note are the worked example’s; your edited lines above recompute the same shape live. In the worked example, prevention is the smallest bucket ($66,000, 16.1%) and failure is nearly two-thirds of the bill ($264,000, 64.5%) — the classic signature of a plant paying for quality at the end of the line instead of the start. The internal failure side is not an abstraction: it is the 4.0% scrap rate (46 of 1,155 lots) and the 385-lot hold queue, priced. That is why the DMAIC project's economics work: every point of scrap and every hundred holds cleared comes straight off the two failure lines, while the prevention that achieves it costs a fraction of either.

RPN ranks attention, it does not measure risk on a ratio scale — a severity-8 mode with a modest RPN (adhesion, 144) still outranks a severity-5 mode. Re-score after the action and keep both scores: the drop is the evidence the action worked.

This chain is the root-cause analysis behind Alert QA-ALERT-2026-014 (the 8D case, DR-118). Its final answer — the root cause below — lands on the alert's cause fields, alongside the escape point the 8D's D4 records.

Stop when the answer names something the system controls — a standard, a setting, a design. “Operator error” is never a fifth why; it is a first why wearing a costume. The verified root-cause register carries chains like this once they are tested, and each opens a CAPA.

A non-significant result is a result: it stops the team rebuilding the Dye House on a hunch. Deming's rule holds — separate signal from noise first, then spend. The test does not say the lines are equal; it says this evidence, at this size, cannot tell them apart on scrap. The holds difference (112 of 276 lots held) is the stronger lead — test that next.

Use the equation for setting and prediction, never as proof of cause — the DOE is the causal test; regression quantifies. And never extrapolate: the line was fitted on deviations between −8 and +8 °C and says nothing honest beyond them.

The interaction is the finding one-factor-at-a-time would miss: temperature's effect depends on dwell. Set temperature high, dwell mid, and stop spending on liquor ratio. Every DOE conclusion is provisional until a confirmation run at the chosen settings reproduces it — that run joins the control plan.

This A3 works the hold-queue case held on Alert QA-ALERT-2026-014 (the 8D case, DR-118): the analysis and countermeasures above are the working layer over that alert record, and the verified root cause lands on the alert's cause fields.

The A3 is the DMAIC project's one-page face — same facts, arranged so a reader can challenge the logic in five minutes. An A3 is read left to right as an argument: if the analysis does not support the countermeasures, the page shows it. That exposure is the point — the sheet is a coaching tool, not a report format.

Read an 8D as a chain of custody for a problem: D3 proves the customer is protected today, D4 proves the team understands why it happened and why it escaped, D5–D6 prove the fix is in and working, and D7 proves the lesson traveled to every line that shares the cause. The discipline most often skipped is the second half of D4 — the escape point. A cause explains how the defect was made; the escape point explains how it passed every check between the process and the customer. Fixing only the first half leaves the door open for the next defect.

Name the waste in the plant's own nouns — “lots waiting on the lab” starts a fix; “waiting waste” starts a seminar. One waste usually pays for the walk: here it is Waiting, wearing a lab coat.

Read the tail, not the head. Sort at 82 says the event happened; Sustain at 57 says the habit has not. The audit is re-scored monthly by the crew itself — a 5S score awarded by visitors is decoration.

The event is not the improvement system; it is an accelerant inside one. Without the standard-work rewrite and the follow-up audit, a kaizen week is a sugar rush — the Control phase exists because of every event that faded by Friday week two.

Prefer prevention over detection, and both over reminders. A sign that says “check the temperature” is not a poka-yoke; a range that refuses to run cold is. The cheapest test: could a tired operator on a Friday still make this mistake? If yes, it is not proofed yet.

A standard nobody follows is a wish. The standard is rewritten by the crew when the process changes — the DMAIC Control gate checks exactly that: chart live, standard current, owner named. Improvement without a standard is just a new way to drift.

A change approved from this function analysis is released as an engineering change order (ECO — BOM revision): the function case is argued here, and the ECO is the record that carries the approved change into the bill of materials. Until an ECO is approved, the rug's BOM is unchanged. The element costs behind these functions (priced in Cost-to-Worth) read from the BOM rollup and supplier prices.

Read FAST right to left with “why”: binding exists so the rug installs cleanly, so it stays in place, so it covers the floor. Any component that cannot be traced to the basic function is cost without a customer — the workshop starts there.

Attack the gap in order — backing first ($24), then yarn and overhead ($18 each). Binding at a $4 gap is a distraction, and the workshop that chases it spends its credibility. Worth is an engineering estimate, argued from the function — not a target dictated to the supplier.

The teardown found the latex coat applied 30% heavier than the lock function requires — the should-cost prices the function, not the habit. A should-cost is an opening position for engineering, not an accusation: every line must be defensible from something weighed, timed or metered.

The workshop's discipline is sequence: function before ideas, ideas before judgment, engineering before promises. The VA/VE Value Levers view holds the standing lever list — value analysis and value engineering are part of Max Quality; this workshop is how a lever earns its number.

Keep DMADV as a note until a design problem actually arrives. Running a design methodology on an improvement problem is how simple fixes acquire steering committees.

Read the gaps in clause order, not by discomfort: clause 6 (planning) and clause 10 (improvement) carry the real work — the system finds problems well (the CAPA list proves it) but scores its risks unevenly and does not yet prove its fixes stayed fixed. “Partial” is the honest middle of this checklist: the practice exists, the evidence or the routine does not. Close a gap by producing the evidence, then re-run the clause — the counts above are built to move.

Two disciplines make this register a system rather than a poster. First, significance is scored before controls are argued about — severity × frequency, threshold 15 — so the dye house outranks the car park every time. Second, the aspects feed the same loop as quality: a spill drill, an effluent exceedance or a missed objective opens a corrective action in the same CAPA register (35 open) that quality uses, because clause 10 is shared. Review the register when the process changes, not just annually — a new dye chemistry is a new aspect.

APQP exists because quality cannot be inspected into a launch. Each phase retires a class of risk while retiring it is still cheap — a backing mistake costs a drawing change in Phase 2 and a recall conversation in Phase 5. The phases also sequence the other tools in this app: the FMEA is written twice (design, then process), the Gage R&R precedes the capability run, and the control plan is the document production actually runs to.

The warrant is the discipline: the PSW is signed only when the other 17 elements stand, so the signature means something. PPAP approval is also not permanent — a material, process or supplier change re-opens the affected elements, which is why this pack exists as a living view rather than a filed document. Interim approval, when a customer grants it, ships parts with named open items and dates — it is never a quiet yes.

Sequence is the method: separate internal from external first — most “changeover time” is fetching and waiting that never needed a stopped machine — then convert what remains, then streamline with standard work. Report stopped time, not total elapsed time: the external work still exists, it just happens while the loom earns. Changeover is what makes smaller batches, and so shorter lead time, affordable.

The honest reading of this view is sequencing. Sector certification follows a customer requirement, never precedes one in hope — the audit fees and surveillance visits are real costs against a hypothetical contract. What does precede it, cheaply, is the base: a closed ISO 9001 gap list and core tools in genuine use mean any of these three becomes an extension project measured in months, not a rebuild measured in years.

This view is the reading window onto the Knowledge Base: the articles live there as Knowledge Base articles — the knowledge base is not stored in this app. The list mirrors the Knowledge Base pattern: categories are parent articles in a nested tree, their articles sit under them, and selecting any article opens it in the reading view beside the list.