CONTEXT(CLONE)
Sign in

Check supplier material certificates against the order specification

For: Incoming quality engineer or supplier quality lead in a regulated or safety-critical supply chain

Pattern: Cross-examinationNeeds live modelsDesigned for 4 to 250 agents
Get early accessSign in to use

The pain today

Every lot arrives with a mill certificate or certificate of conformance. Someone must compare chemistry, mechanical values, heat treatment, standard revision and heat number with the order. Under time pressure this becomes a stamp.

The ask

“I attached the material certificates for received lots as text and the purchase orders and material specifications they were ordered to. For each certificate, tell me whether every required property is reported, within the specified range and to the right standard revision, and whether the heat and lot numbers match the order.”

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

  • Material and conformance certificates as text
  • Purchase orders with specification call-outs
  • Material specifications with required ranges
  • Receiving records with heat and lot numbers

The unit of work

One worker task per one certificate against its order line.

Why a swarm fits

One certificate needs one order line and one specification. Lots are independent. Two families reading the same table independently catch transposed values that one reader repeats.

Not for

Scanned or photographed certificates without a text layer, and detecting a forged certificate. It checks what is written, not the metal.

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 checkA choice among options

    Before work starts on a unit

    Do the heat or lot number and material grade on the certificate match the receiving record and the order line?

    Sees only: The certificate header, the receiving record and one order line

    Why: A certificate for the wrong heat is a finding in itself and is never checked further as if it matched.

    • Both matchthenAccept
    • Grade matches, heat or lot differsthenMark unresolved
    • No matching order linethenMark unresolved

  2. After workers, the judge checks

    Evidence checkA choice among options

    After a worker answers

    Does the certificate report a value for this required property, in the specification's unit, inside the quoted required range?

    Sees only: One specification row and the certificate's row for that property

    Why: A missing property is a finding and never assumed to conform.

    • Reported, inside the rangethenAccept
    • Reported, outside the rangethenMark unresolved
    • Reported in a different unitthenEscalate to a strong model
    • Not reportedthenMark unresolved

  3. Evidence checkYes or no, with a probability

    After a worker answers

    Does the standard and revision printed on the certificate match the standard and revision called out on the order line?

    Sees only: The certificate's standard line and the order's specification call-out

    Why: Catches material certified to a superseded revision.

    • 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

    Did the two readers quote the same value for this property from the certificate?

    Sees only: Both readers' quoted values for one property

    Why: Two families reading one table catch transposed digits that one reader repeats.

    • Same value quotedthenAccept
    • Different values quotedthenEscalate to a strong model
    • One reader found no valuethenEscalate to a strong model

  5. Run control, between rounds

    Retry or stopYes or no, with a probability

    After a rejection or low confidence

    Is the certificate's property table complete and readable as text, without merged or broken rows?

    Sees only: The raw text of the certificate's property table

    Why: Retries on a stronger model only when the text is usable; garbled tables go to a person.

    • Yes: 0.60 or higherthenEscalate to a strong model
    • Unsure: 0.40 up to 0.60thenMark unresolved
    • No: below 0.40thenMark unresolved

  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does the finding show an out-of-range, missing or mismatched entry, or a heat number that does not match?

    Sees only: One certificate's check result with quotes

    Why: Lot acceptance and any concession belong to the quality engineer, as does any suspicion of falsification.

    Accountable: The quality engineer owns lot acceptance and any concession; suspected falsification goes to the supplier quality lead.

    • 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 pairs each certificate with its order line and the specification revision called out.

    Decisions here:1. Scope check

  2. Workers

    Two sets of small workers from different families each check the same certificate against the required ranges.

    Designed for 4 to 250 agents, one worker task per one certificate against its order line. Each worker receives only its own unit.

  3. Judge, from a different model family

    A decision model from a third family settles disagreements by re-reading the quoted values, or keeps them open.

    Decisions here:2. Evidence check3. Evidence check

  4. Reconciler

    A strong reasoning model lists failed, missing and disputed properties per lot and patterns per supplier.

    Decisions here:4. Conflict check5. Retry or stop

  5. Accountable person

    The quality engineer owns lot acceptance and any concession; suspected falsification goes to the supplier quality lead.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Each reported value is quoted with its unit next to the required range
  • A missing property is a finding, never assumed to conform
  • Standard revision on the certificate is compared with the one ordered

What comes back

  • Check result per certificate with quoted values and ranges
  • Lots with out-of-range, missing or mismatched entries
  • Disagreements between the two readers
  • Recurring gaps per supplier

What to measure

  • Findings confirmed by the quality engineer
  • Non-conforming lots that passed the check
  • Review time per certificate
  • Cost per certificate

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