Find recurring causes across non-conformance reports

For: Quality manager or continuous improvement lead across several plants or lines

Pattern: Map, verify, reduceNeeds scaleDesigned for 40 to 500 agents

The pain today

Non-conformance reports are written by many people in free text. The same failure is described in different words at different plants, corrective actions are closed, and the defect returns without anyone linking the reports.

The ask

I attached our non-conformance reports and their corrective actions. For each report, pull out the part, process step, defect, stated cause and action. Then show me which causes keep recurring across plants, and which defects came back after a corrective action was closed.

Plain words, as you would say it to a colleague. Edit it to fit your case before you send it.

What you attach or connect

  • Non-conformance reports exported as text
  • Corrective and preventive action records
  • Part and process step lists

The unit of work

One worker task per one non-conformance report.

Why a swarm fits

Each report is normalised alone against a short taxonomy. Reports are independent and numerous; the recurrence only appears when all of them are coded the same way.

Not for

Finding the physical root cause of a defect. It organises what the reports say and cannot see the part, the process data or the machine.

The decision tree

6 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

  1. Planner, while planning

    Scope checkYes or no, with a probability

    Before work starts on a unit

    Does this report describe a nonconforming part or process output, rather than a safety observation, an audit note or a duplicate entry?

    Sees only: One report's header and description field

    Why: Keeps the coded set to real product non-conformances.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenEscalate to a strong model
    • No: below 0.30thenSkip this unit
  2. Before workers, before a task runs

    Small worker or strong modelA choice among options

    Before a task runs

    Does the report state a cause in its own cause field, or must the cause be read out of free-text discussion?

    Sees only: The cause field and description of one report

    Why: Reports with no stated cause are coded as such instead of being given one.

    • Cause field filledthenAccept
    • Cause only in free textthenEscalate to a strong model
    • No cause stated anywherethenMark unresolved
  3. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Does the quoted wording from the report describe the cause category the worker assigned from the frozen taxonomy?

    Sees only: One quoted sentence, the assigned cause code and its taxonomy definition

    Why: Keeps coding consistent so that recurrence counts mean the same thing at every plant.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry
  4. Reconciler, while merging

    Conflict checkA choice among options

    While reconciling

    Do these two coded reports share the same part family, process step and defect type?

    Sees only: Two coded reports with their quotes

    Why: A recurrence is claimed only on a full match; near matches stay visible but unclaimed.

    • All three matchthenAccept
    • Two of three matchthenMark unresolved
    • One or none matchthenContinue
  5. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Is the later report's occurrence date after the closure date recorded on the earlier report's corrective action?

    Sees only: The later report's date field and the earlier action's closure record

    Why: Confirms from dates on paper that a defect returned after an action was closed.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry
  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does the cluster include a defect that returned after a closed corrective action, or reports from more than one plant?

    Sees only: One recurrence cluster with member reports

    Why: The quality manager decides what is opened as a systemic issue and who owns it.

    Accountable: The quality manager owns which recurrences are opened as systemic issues and who is accountable for the corrective action.

    • Yes: 0.40 or higherthenAsk a person
    • Unsure: 0.15 up to 0.40thenAsk a person
    • No: below 0.15thenAccept

The fleet: who does what

Model tiers by role, not brands: you choose the models. Strong reasoning models plan and reconcile, small fast models do the wide work, and the judge is a decision model from a different family, so it does not share the workers' blind spots.

  1. Planner

    A strong reasoning model drafts the defect and cause taxonomy from a sample and fixes it before the wide pass.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from an open-weight family code one report into the taxonomy with the quoted wording.

    Designed for 40 to 500 agents, one worker task per one non-conformance report. Each worker receives only its own unit.

    Decisions here:2. Small worker or strong model

  3. Judge, from a different model family

    A decision model from a different family checks that the quoted text supports the assigned cause and defect codes.

    Decisions here:3. Evidence check5. Evidence check

  4. Reconciler

    A strong reasoning model clusters coded reports, links repeats to closed actions and keeps reports with disputed codes.

    Decisions here:4. Conflict check

  5. Accountable person

    The quality manager owns which recurrences are opened as systemic issues and who is accountable for the corrective action.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Every code is backed by a quote from the report
  • Reports with no stated cause are coded as such, not given one
  • A recurrence is claimed only when part, process step and defect all match

What comes back

  • Coded report table with quotes
  • Recurring cause clusters across plants
  • Defects that returned after a closed corrective action
  • Reports too thin to code

What to measure

  • Clusters the quality manager confirms as real recurrences
  • Coding agreement with a quality engineer on a sample
  • Review time per report
  • Cost per report

Names of measures only. No result is claimed for this template.

Templates open in the workspace chat with the ask filled in. Nothing runs until you send it.

Get early accessSign in to use

Trace failure modes through control plan to work instructions

For: Quality engineer or process engineer preparing for a customer or certification audit

The process failure analysis lists controls, the control plan should carry them, and the work instructions should tell the operator how.

Pattern: Hierarchical decompositionNeeds live models6 decisionsDesigned for 12 to 300 agents