CONTEXT(CLONE)
Sign in

Generate and narrow hypotheses for a chronic defect

For: Process engineer or plant quality lead facing a defect that keeps coming back

Pattern: TournamentNeeds a connectorDesigned for 24 to 300 agents
Get early accessSign in to use

The pain today

The clues to a chronic defect are scattered over shift logs, maintenance records, material lots, setup sheets and earlier reports. Each person sees one slice, and the team anchors on the first plausible cause.

The ask

“Look at the shift logs, maintenance records, lot records, setup sheets and defect reports for this line over the period the defect has been appearing. Propose as many candidate explanations as the records support, then knock out the ones other records contradict. Give me a short list with the evidence for and against each.”

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

  • Shift and production logs
  • Maintenance and changeover records
  • Material lot and supplier records
  • Defect and scrap reports
  • Setup and parameter sheets as text

The unit of work

One worker task per one record slice producing candidate hypotheses.

Why a swarm fits

Many workers each propose causes from one narrow slice, which avoids early anchoring. Independent judges then test every candidate against the other slices, round by round.

Not for

A defect with an obvious single cause, or one that only shows in sensor signals. It proposes candidates; it does not confirm any of them.

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

    Do the dates in this record slice overlap the period in which the defect reports were raised?

    Sees only: The date range of one slice and the defect report dates

    Why: Drops slices that cannot bear on the defect before hypotheses are generated from them.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenEscalate to a strong model
    • No: below 0.30thenSkip this unit

  2. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Do the quoted records mention the machine, lot, setting or event that the hypothesis names?

    Sees only: One hypothesis and the records it cites

    Why: Discards candidates that are not anchored in any record.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry

  3. Evidence checkA score

    After a worker answers

    Across the defect reports shown, in what share does a record in this slice place the hypothesised condition before the defect was found?

    Sees only: One hypothesis, one other slice and the defect report dates

    Why: Ranks candidates by how consistently other records line up with them, not by how plausible they sound.

    • High: 0.85 or higherthenAccept
    • Middle: 0.50 up to 0.85thenMark unresolved
    • Low: below 0.50thenSkip this unit

  4. Evidence checkA choice among options

    After a worker answers

    Does the quoted record show the defect occurring when the hypothesised condition was absent?

    Sees only: One hypothesis and one quoted record offered as a contradiction

    Why: Elimination needs a quoted contradicting record, never a judge's opinion.

    • Shows the defect without the conditionthenAccept
    • Shows something unrelatedthenReject and retry
    • Record is ambiguous on the conditionthenMark unresolved

  5. Run control, between rounds

    Another round?Yes or no, with a probability

    Between rounds

    Did the latest round leave the set of surviving hypotheses unchanged?

    Sees only: Surviving hypotheses before and after the round

    Why: Stops the tournament once further rounds no longer narrow the field.

    • Yes: 0.70 or higherthenStop
    • Unsure: 0.40 up to 0.70thenContinue
    • No: below 0.40thenContinue

  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does the short list include a hypothesis that would need a change to a machine setting, material or process step to test?

    Sees only: The surviving hypotheses with their suggested checks

    Why: The process engineer chooses what is tested on the line; the swarm confirms no cause.

    Accountable: The process engineer owns which hypotheses are tested on the line and the conclusion; the plant manager owns any process change.

    • 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 slices records by source, period and machine and sets the scoring rules for candidates.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from an open-weight family each propose hypotheses from one slice, citing the records.

    Designed for 24 to 300 agents, one worker task per one record slice producing candidate hypotheses. Each worker receives only its own unit.

  3. Judge, from a different model family

    Judges from a different family score each candidate against the other slices and eliminate contradicted ones per round.

    Decisions here:2. Evidence check3. Evidence check4. Evidence check

  4. Reconciler

    A strong reasoning model merges duplicates and presents the surviving candidates with evidence for and against.

    Decisions here:5. Another round?

  5. Accountable person

    The process engineer owns which hypotheses are tested on the line and the conclusion; the plant manager owns any process change.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Each hypothesis cites the records that prompted it
  • Elimination requires a quoted contradicting record, not a judge's opinion
  • Coincidence in time is labelled as such and never reported as cause

What comes back

  • Short list of surviving hypotheses with evidence for and against
  • Eliminated hypotheses with the contradicting record
  • Data gaps that would separate the survivors
  • Suggested checks for the engineer to run

What to measure

  • Whether the confirmed cause was on the short list
  • Hypotheses the engineer had not considered
  • Engineer hours to reach a test plan
  • Cost per investigation

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

More in Manufacturing and quality

Find recurring causes across non-conformance reports

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

Non-conformance reports are written by many people in free text.

Pattern: Map, verify, reduceNeeds scale6 decisionsDesigned for 40 to 500 agents

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