Manufacturing AI agent · Calibration
Calibration, Metrology & Out-of-Tolerance Impact AI Agent
When a gauge comes back out of tolerance, assemble the product it measured since the last known-good calibration, mark where the usage record is thin, and hold the scope for a quality engineer.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Engineer signs
- Pre-disposition
- After launch
- Agent Care
What this agent does
Scopes the impact, not the disposition
In
- 01
A gauge falls due, and its record, interval and last certificate are pulled from the calibration system.
- 02
A certificate comes back, and the as-found and as-left readings are normalised point by point.
Reason
- 03
A calibration comes back out of tolerance, and the window opens at the last point the gauge was known good.
- 04
A drop or a repair sits in the history, and the window widens past the previous certificate to cover it.
- 05
A gauge moves between cells, and the population is built from usage records, not its registered home.
Decide
- 06
A reference standard fails, and the working gauges it qualified are cascaded before product is listed.
- 07
A device gauge fails, and the record is built to ISO 13485:2016 clause 7.6, not the reserved 21 CFR 820.72.
Out
- 08
A scope is assembled, and it is routed with its as-found data to the named quality engineer.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent scopes the population; ISO 9001:2015 clause 7.1.5.2 leaves the determination with the organisation, and an engineer signs it.
Example workflow
One out-of-tolerance event, end to end
- 1Gauge found out of toleranceCalibration certificate, in-use check, damage report or repair record
- 2Window anchoredAs-found readings, the previous certificate, interval history and any intervening event
- 3Population assembledSerials and lots, the characteristics measured, open holes and confidence
- 4Controls appliedAs-found-against-tolerance checks, usage-record coverage, cascade checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is dispositioned at any of them — the agent is scoping, and the engineer's lane opens at the confidence gate.
Goes to the quality engineer to disposition.
Adds a metrology-lead read first.
Engineer disposition
The scope is held with its as-found data, its open holes and the confidence.
Accept · Widen · Send to metrology review
Dispositioned — impact recorded
- 6Quality systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedScope changes, holes found late, and IATF 16949:2016 customer notifications
Widenings
Every scope the engineer widens is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Deciding earlier measurements are still valid.
- Dispositioning product in the suspect population.
- Telling the customer that suspect product shipped.
- Narrowing a window the usage record cannot support.
Automation boundary
Agent acts unaided
- ✓Assemble the product measured since the last known-good.
- ✓Carry each serial forward with the usage record behind it.
- ✓Show which acceptance decisions the as-found error touches.
- ✓Mark the holes the engineer should weigh, and hold the scope.
Any write happens inside the boundaries agreed at implementation, never ahead of disposition.
- Judging a measurement system capable from a certificate.
- Setting or extending a calibration interval.
- Starting a correction or a removal.
- Changes to interval, tolerance or disposition rules.
Example output
One out-of-tolerance event, annotated
Everything the agent scopes is attached to the record it was drawn from.
Impact-scoping output · single gauge
Illustrative example- Gauge
- Bore gauge, cell C
- As-found finding
- Reads high at the upper calibration point
- Error as found
- +18 µm
- Window anchor
- Previous certificate
- Confidence
- 88%
- Population
- Serials from the usage log
As received
Taken from the certificate and the usage log on file — nothing on this side is inferred by the agent.
Evidence used
- As-found data on file
- Previous certificate
- Usage log entries
Why this window
It runs from the last point the gauge was known good — the boundary the engineer weighs.
What the score decides
Below the configured threshold the scope picks up a metrology read before the engineer sees it.
Action
AcceptWidenSend to metrology reviewValue
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 gauge
From the calibration record
Scope from the record
Draw on the certificates, the usage logs and the confirmation rules ISO 10012:2026 sets out.
One bad gauge, months of parts
Recall status and due dates arrive already tracked against the asset register.
Send review to the thin evidence
Open holes and low-confidence scopes are marked, so the engineer's read starts where the record is thin.
Build an evidence trail
The gauge, the standard it was checked against and the engineer who dispositioned stay on the record.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Calibration management
- GAGEtrak
- CyberMetrics
- ProCalV5
- IndySoft
Metrology and gauging
- Zeiss Calypso
- PC-DMIS
- Mitutoyo MeasurLink
Quality and MES
- InfinityQS
- SAP QM
- Plex
- Siemens Opcenter
Agent
Calibration & impact scoping
- Reads the record
- Scopes the impact
- Holds for disposition
ERP and traceability
- SAP
- Oracle
- Infor
- Lot and serial records
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 scope
Six controls stacked inward. Whatever survives the lot of them is named in the map beneath.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a scope may stand
- L3
- Puts the disposition in an engineer's hands
- L4 – L5
- Keep the gauge and the standard behind it
- L6
- Holds the impact at unscoped when signals degrade
How Nestack evaluates it
Evaluate the whole scoping workflow — not only the finished population.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the scope the engineer signs
- E1Final-output evaluationDid every serial in the population trace back to a usage record?
- E2Step-level evaluationDid the agent use the right certificate, anchor and tolerance table?
- E3Tool evaluationDid it read and write the correct gauge and the correct event?
- E4Confidence calibrationDo low-confidence scopes actually attract more widenings?
- E5Slice evaluationHow does performance change across specific gauge families?
- E6Business outcomeHow many scopes were widened, or corrected after product had shipped?
Floor — the impact the plant answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, set at the stage each one starts in.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- MG-03
Home-location assumption
Usage read from the register, not from the floor.
Stage gathers
Certificates, usage logs, intervals and tolerances
02 · Reasoning
2 modes
- MG-04
Anchor set too late
The window opens at a certificate a drop preceded.
- MG-06
Exception averaged away
A safety-critical characteristic hides in a family.
Stage proposes
Serials, characteristics, holes and confidence
03 · Tool / write
2 modes
- MG-02
Hole closed silently
A gap in the usage log is filled without evidence.
- MG-05
In-transit stock missed
Finished goods in transit are left out of scope.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- MG-01
Population without evidence
Serials are listed with no chain behind them.
Stage returns
The scope the engineer dispositions and signs
05 · Change / Version
1 mode
- MG-07
Silent rule regression
A rule change narrows what the scope will include.
Stage tracks
Model, prompt, interval rules and tolerance tables
- Sev-1 · scope closed outside the boundary
- Sev-2 · suspect product left out of scope
- Sev-3 · record degrades, scope goes to review
Affected slices
No published out-of-tolerance rate to compare
A site-level calibration rate can look healthy while a few gauge families absorb most of the scope widening. Nestack reports the widening rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Gauges shared across cells | 6.4% | 3.6× | Review | |
| Certificates with no as-found | 4.1% | 2.3× | Review | |
| Reference standards and masters | 2.8% | 1.6× | Watch | |
| Fixed single-station gauges | 1.1% | 0.6× | Normal |
Bar: scope-widening-rate lift vs. fixed-gauge baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
No cycle closes without a new case
A cycle closes when the missed impact is a case the next release has to catch. That suite is what the next out-of-tolerance event is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Scope-widening rate rises in a gauge family.
- 02
Diagnose
The gauge that came back out of tolerance in March is read back until one cause is left.
- 03
Improve
The correction is numbered and the gauges that produced it are pinned to the note.
- 04
Verify
One impact case still failing is enough to keep the release on the bench.
- 05
Learn
The case is kept, and the interval rules are updated in the same change.
Learn → Detect
The return edge. The next event 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, impact scoping, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Calibration-workflow and scoping boundary definition.Week 1
- 02Calibration and metrology data assessment.Week 1, Week 2
- 03Tolerance, interval and cascade mapping and rule mapping.Week 1, Week 2
- 04Certificate ingest and usage-log mapping.Week 2
- 05Impact scoping and evidence binding.Week 2, Week 3
- 06Confidence scoring and hole routing.Week 3
- 07Engineer disposition workflow.Week 3, Week 4
- 08Quality and ERP-system integration.Week 5
- 09Recall and impact cases.Week 4, Week 5
- 10Guardrails and disposition controls.Week 4
- 11Gauge-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne site, one gauge family | ProductionProduction calibration systems | AdvancedMultiple sites / systems |
|---|---|---|---|
| Introduced at Pilot | |||
| Scoping to your records and rules | ✓ | ✓ | ✓ |
| Engineer disposition | ✓ | ✓ | ✓ |
| Impact-scoping baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by gauge family | — | ✓ | ✓ |
| Disposition workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Measurement-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-standard tolerance rules | — | — | ✓ |
| Multi-stage quality approvals | — | — | ✓ |
| High gauge volume | — | — | ✓ |
| Multi-site calibration controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom 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.
You bring
We build with it
01Your gauge register and interval rules
→Certificate ingestion and usage-log mapping
Week 102Representative certificates, as-found included
→Scoping baseline, evidence indexing and gauge binding
Week 203Your tolerance tables and cascade rules
→Tolerance, interval and cascade-rule mapping
Week 104Access to relevant APIs, feeds or exports
→Calibration and metrology assessment, then integration setup
Week 205Impacts you would not want scoped
→Out-of-tolerance cases and the evaluation suite
Week 406What no impact scope may omit
→Confidence scoring, hole routing, guardrails and disposition controls
Week 307A named quality engineer to disposition scopes
→Engineer disposition 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 lands where the work lands, which puts evaluation and the first live recalls in one week.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Calibration workflow discovery, rule mapping and the automation boundary
W2Source integration and the impact-scoping baseline
W3Impact scoping, confidence logic and disposition controls
W4Evaluation suite, anchor and cascade checks and failure-mode testing
W5Quality-system integration, pilot events and targeted corrections
W6One calibration cycle run under the metrology lead, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its work is named in. The week 5 overlap is work, not padding.
- At the end of W6
- The last checks clear on live events, and Agent Care assumes monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Manufacturing AI agent
Build an impact-scoping agent around your out-of-tolerance procedure.
Show us your gauge register and your usage logs. Who tells the customer that suspect product shipped? Your named quality manager — IATF 16949:2016 clause 7.1.5.2.1 makes it a record.
- Typical delivery
- 4–6 weeks
- Deployment
- Your calibration stack
- Engineer signs
- Pre-disposition
- After launch
- Agent Care available
Nestack Agents · Calibration & impact scoping
AGT-MFG-16 · Agent Care available after launch