Main site

Manufacturing AI agent · Genealogy

Production-Genealogy & Recall-Scoping AI Agent

Maintain as-built genealogy from the records your plants already write, assemble the smallest population the evidence will defend, and hold it for the named person who decides scope.

Typical delivery
4–6 weeks
Deployment
Your stack
Person decides
Pre-filing
After launch
Agent Care

What this agent does

Assembles the population, not the decision

In

  1. 01

    A unit is built, and the lots, machines, tooling and test results the line captures are bound to its serial.

  2. 02

    A lot is consumed, and its split across work orders is carried down to the unit, not left at the order.

Reason

  1. 03

    A unit is reworked off the line, and the bench event is carried forward rather than left as a silent gap.

  2. 04

    A build is covered only to the level risk-graded in the IATF 16949:2016 8.5.2.1 plan, and that level is stated.

  3. 05

    A defect is confirmed, and the population is assembled with the evidence that puts the adjacent units outside it.

Decide

  1. 06

    A record is missing, and the population widens to the boundary the evidence holds, with the gap named.

  2. 07

    A lot is suspect, and forward tracing lists where it went, as far down the tiers as the distribution records reach.

Out

  1. 08

    A population is assembled, and it goes with its exclusion evidence to the named recall decision-maker.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent assembles the population; 49 CFR 573.6(c)(2), as in force in 2026, makes the basis for it a filed statement a named officer signs.

Example workflow

One confirmed defect, end to end

  1. 1Defect confirmedField failure, warranty signal, test escape or supplier notice
  2. 2Genealogy pulledLots, sub-assembly serials, machines, tooling, process data and rework events
  3. 3Population assembledSerials in scope, the units ruled out, open gaps and confidence
  4. 4Controls appliedExclusion-evidence checks, gap coverage, forward-trace checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and no unit is included or excluded at any of them — the agent is assembling, and the decision-maker's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the recall decision-maker to scope.

Low confidence

Adds a legal read first.

Scope decision

The population is held with its exclusion evidence, its open gaps and the confidence.

Sign off · Widen · Send to legal review

Scoped — decision recorded

  1. 6Quality systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedWidenings, units found late, and populations amended after the report was filed

Widenings

Every population the decision-maker widens is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Determining that a defect is safety-related.
    • Deciding the affected population and the remedy.
    • Judging whether a condition is reportable.
    • Filing a defect report with a regulator.

  • Automation boundary

    Agent acts unaided

    • ✓Maintain as-built genealogy from the records the line writes for the named owner.
    • ✓Assemble the smallest population the records will defend into the review queue.
    • ✓Show what puts each adjacent unit outside that boundary.
    • ✓Mark the gaps in the genealogy, and hold the population.

    Any write happens inside the boundaries agreed at implementation, never before the scope decision.

    • Narrowing a population the genealogy cannot support.
    • Telling a customer that suspect units shipped.
    • Setting the recall strategy or its depth.
  • Changes to genealogy, exclusion or retention rules.

Example output

One confirmed defect, annotated

Everything the agent assembles is attached to the record it was drawn from.

Scoping output · single defect

Illustrative example
Defect
Sensor sub-assembly
Population line
Traced to one supplier lot on one line
Units in scope
1,320 units
Basis of record
As-built genealogy
Confidence
86%
Units ruled out
Adjacent lots, evidenced

As received

Taken from the build records the line already writes — nothing on this side is inferred by the agent.

Evidence used

  • Lot-to-unit records
  • Machine and test data
  • Rework bench entries

Why these units

It is the boundary the build records will defend and a person still decides.

What the score decides

Below the configured threshold the population picks up a legal read first.

Action

Sign offWidenSend to legal review

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every unit

From the build record

03Scoping

Scope from the record

Draw on the lot, machine, test and rework records the plants already write.

01Approved path

Scope is an exclusion problem

NHTSA's November 2024 Ford consent order required a VIN-based traceability system.

02Human review

Send review to the thin records

Open gaps and low-confidence populations are marked, so the read starts where the genealogy is thin.

04

Build an evidence trail

The unit, the record it was built from and the engineer who scoped it stay on the campaign.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • MES and shop floor

    • Siemens Opcenter
    • Plex
    • Critical Manufacturing
    • AVEVA

  • ERP and serialisation

    • SAP
    • Oracle
    • Infor
    • Lot, batch and serial masters

  • Quality and PLM

    • Teamcenter
    • Windchill
    • ETQ
    • Sparta TrackWise

Agent

Genealogy & recall scoping

  • Reads the records
  • Assembles the population
  • Holds for the decision

  • Distribution and field

    • WMS
    • consignee records
    • Warranty and field-failure data

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the population

The controls sit six deep, innermost last. What escapes every one is set out in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeWiden the population back to the last defensible boundary when signals degrade.Roll back
L5Version monitoringTrack model, prompt, genealogy-rule and exclusion-rule changes.Track
L4TraceabilityRecord the build records read, the exclusions, the gaps and the sign-off time.Record
L3Scope decisionHold populations for the named decision-maker; it governs release, not whether the scope is right.Gate
L2Policy guardrailsTest populations against configured exclusion and gap rules; a failure returns the population.Restrict
L1Confidence thresholdsRoute low-confidence populations to a legal read before the decision-maker sees them.Require review
Model corePopulation assembled — serials in scope, units ruled out, open gaps and confidence
L1 – L2
Test whether an exclusion may stand
L3
Puts the scope in a named person's hands
L4 – L5
Keep the unit and the record behind it
L6
Widens the population when signals degrade

How Nestack evaluates it

Evaluate the genealogy underneath — not only the finished population.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the population that goes out

  1. E1Final-output evaluationDid every serial in scope, and every one ruled out, trace to a record?
  2. E2Step-level evaluationDid the agent use the right build records, lot links and rework events?
  3. E3Tool evaluationDid it read and write the correct unit and the correct build period?
  4. E4Confidence calibrationDo low-confidence populations actually attract more widenings?
  5. E5Slice evaluationHow does performance change across specific build periods?
  6. E6Business outcomeHow many populations were widened, or amended after the report was filed?

Floor — the population the plant defends

Failure modes

Where each failure originates in the agent

Seven failure modes, placed where each one begins.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • YG-03

      Rework bench unread

      Off-line rework is absent from the build trail.

    Stage gathers

    Lot links, machine data, rework events and tests

  2. 02 · Reasoning

    2 modes

    • YG-04

      Gap closed by inference

      A missing lot link is bridged without evidence.

    • YG-06

      Lot read at order level

      A split lot is scoped to the order, not the unit.

    Stage proposes

    Serials in scope, exclusions and confidence

  3. 03 · Tool / write

    2 modes

    • YG-02

      Population released early

      Agent acts despite insufficient certainty.

    • YG-05

      Lot code collision

      Two plants number lots the same way.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • YG-01

      Population too narrow

      Affected units fall outside the scope.

    Stage returns

    The population the decision-maker signs and files

  5. 05 · Change / Version

    1 mode

    • YG-07

      Silent scoping regression

      A rule change narrows what the population includes.

    Stage tracks

    Model, prompt, genealogy rules and exclusion rules

  • Sev-1 · scope closed outside the boundary
  • Sev-2 · affected units left out of scope
  • Sev-3 · record degrades, scope goes to review
See the complete Manufacturing failure-mode catalogue

Affected slices

A complete genealogy is not an even one

A plant-level completeness score can read strong while a few build periods carry most of the widening. Nestack reports the widening rate by slice, not only in total across the plant.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Lots split across work orders5.9%3.5×Review
Units reworked off the line3.8%2.2×Review
Build periods past retention2.6%1.5×Watch
Single-line serialised builds1.4%0.8×Normal

Bar: scope-widening-rate lift vs. single-line serialised baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Nothing closes until the gap is a test

A cycle is done when the hole in the genealogy is a case the next release must close. That suite is what the next population scoped is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Widening rate rises in a build period.

  2. 02

    Diagnose

    The recall that took every unit built that year is traced back until one cause is left.

  3. 03

    Improve

    Version the fix, and the units that exposed it stay attached to that version.

  4. 04

    Verify

    The release is held back until each touched scoping case passes again.

  5. 05

    Learn

    The case is retained, and the genealogy rules move with it in one change.

  6. Learn → Detect

    The return edge. The next population is scoped against a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, population assembly, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Scoping workflow and evidence boundary definition.Week 1
  2. 02Build-record and genealogy assessment.Week 1, Week 2
  3. 03Lot, serial, rework and process-record rule mapping.Week 1, Week 2
  4. 04Genealogy ingest and unit-link mapping.Week 2
  5. 05Population assembly and evidence binding.Week 2, Week 3
  6. 06Confidence scoring and gap routing.Week 3
  7. 07Scope-decision workflow.Week 3, Week 4
  8. 08MES and ERP-system integration.Week 5
  9. 09Population and evidence cases.Week 4, Week 5
  10. 10Guardrails and scoping controls.Week 4
  11. 11Unit-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne plant, one product lineProductionProduction shop-floor systemsAdvancedMultiple plants / systems
Introduced at Pilot
Genealogy from your build records✓✓✓
Scope decision✓✓✓
Genealogy-completeness baseline✓✓✓
Introduced at Production
Reporting by build period—✓✓
Scope-decision workflow in your systems—✓✓
Approved write-back—✓✓
Shop-floor record integration—✓✓
Introduced at Advanced
Multi-sector reporting rules——✓
Multi-stage quality approvals——✓
High build volume——✓
Multi-plant genealogy controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your build records and serialisation scheme

    Genealogy ingestion and unit-link mapping

    Week 1

  2. 02Representative builds, rework included

    Scoping baseline, evidence indexing and unit binding

    Week 2

  3. 03Your exclusion rules and retention limits

    Lot, serial and rework-rule mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    MES, ERP and quality assessment, then integration setup

    Week 2

  5. 05Populations you would not want defended

    Exclusion cases and failure-mode testing

    Week 4

  6. 06What no narrowing may rest on

    Confidence scoring, gap routing, guardrails and scope controls

    Week 3

  7. 07A named decision-maker to scope campaigns

    Scope-decision workflow, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Each band covers the weeks the build really takes, so evaluation and pilot share the fifth.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Scoping workflow discovery, rule mapping and the automation boundary

W2Source integration and the genealogy baseline

W3Population assembly, confidence logic and scope controls

W4Evaluation suite, exclusion and gap checks and failure-mode testing

W5Quality-system integration, pilot defects and targeted corrections

W6One build period reconstructed under the quality manager, then handover

Reading the band
A bar sits only on the weeks its own work is named in. Week 5 really does carry two phases.
At the end of W6
The final checks clear on live campaigns and monitoring moves to Agent Care.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Manufacturing AI agent

Build a recall-scoping agent before a regulator scopes it for you.

Show us your build records and who decides scope. The evidence that the units you left out are fine has to survive a regulator reading it, a customer spot-checking it and a field failure two years on.

Typical delivery
4–6 weeks
Deployment
Your shop-floor stack
Person decides
Pre-filing
After launch
Agent Care available

Nestack Agents · Genealogy & recall scoping

AGT-MFG-21 · Agent Care available after launch

Build a recall-scoping agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your build records

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.