Mining AI agent · Supply-chain evidence
Minerals Supply-Chain Evidence AI Agent
Keep current the evidence a producer is asked for again and again — chain of custody per shipment, certificate and audit clocks, answers still tied to a source. The certification sits downstream.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Manager answers
- Evidence only
- After launch
- Agent Care
What this agent does
Assembles the evidence, not the declaration
In
- 01
A parcel is despatched, and the despatch, weighbridge and seal records are pulled from supported sources.
- 02
A lot is blended from two pits, and each contributing lease is carried rather than the template default.
Reason
- 03
A questionnaire arrives, and each prior answer is re-checked against the source document that supported it.
- 04
An audit falls due, and the scheme cycle the operation is actually enrolled in is read, not a generic calendar.
- 05
A certificate nears expiry, and the shipments and answers leaning on it are listed while there is still time.
Decide
- 06
An origin claim rests on an assay, and the certificate is drawn from the assay copilot rather than restated.
- 07
A tracing request lands, and the extraction and transport records for that consignment are assembled to a date.
Out
- 08
A pack is retained as a dated version, beside the records under it and what the answering manager was shown.
- 09
A write action runs only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles and keeps the evidence current; a named manager answers the customer, and the operator carries every declaration it signs.
Example workflow
One shipment, pit to pack
- 1Records receivedDespatch, weighbridge, seal, assay and lease records
- 2Custody assembledTransfers, blend inputs, certificates and prior answers, each with its named source
- 3Evidence pack draftedCustody chain, supporting documents, open gaps and confidence
- 4Controls appliedCustody-continuity tests, certificate-currency checks, scheme-scope tests and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is answered at any of them — the agent is assembling evidence, and the manager lane opens at the confidence gate.
Goes to the commercial manager.
Adds a technical read first.
Manager review
The pack is held with its records, its open gaps and the confidence.
Approve · Amend · Send to technical review
Approved — released to answer
- 6Evidence register updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAmendment distance, open-gap outcomes, audit findings and post-answer corrections
Amendments
Every manager amendment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Certifying the origin of a parcel — a named person does that.
- Signing a conformity or due-diligence declaration.
- Answering a customer, a scheme or an auditor.
- Deciding whether a scheme criterion is met.
Automation boundary
Agent acts unaided
- ✓Assemble the chain-of-custody record for a despatched parcel.
- ✓Carry each claim with its source, and flag what has none.
- ✓Track certificate expiry and the audit cycles entered.
- ✓Draft the questionnaire response and hold it for the manager.
Writes run only inside the boundaries agreed at implementation, and never ahead of the manager.
- Committing the site to a scheme or its assurance clock.
- Warranting origin in a sales or offtake contract.
- Withdrawing or correcting a declaration already issued.
- Changes to scheme, evidence or approval rules.
Example output
One shipment record, annotated
Everything the agent assembles is attached to the record it was drawn from.
Evidence pack output · single shipment
Illustrative example- Counterparty
- Concentrate offtake
- Custody line
- Blended in a stockpile from two pits under separate leases
- Parcel reference
- MC-0412-B
- Source of record
- Stockpile transfer note
- Confidence
- 84%
- Basis of record
- Weighbridge and seal records
As received
Taken from the despatch record and the transfer note — nothing on this side is written by the agent.
Source facts used
- Weighbridge ticket
- Stockpile transfer note
- Container seal record
Why this is open
The parcel carries material from two leases, and the template names only one.
What the score decides
Below the configured threshold the pack picks up a technical read before the manager sees it.
Action
ApproveAmendSend to technical 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 shipment
From the despatch and assay records
Build the custody record
Records already on file are joined into a chain; what is claimed to a customer stays with the manager who answers.
The signer is downstream
Regulation (EU) 2017/821 binds the Union importer, not the mine. What is at stake upstream is market access.
One half is not a second product
If you run the shipped commercial and ESG agent, the questionnaire half is a feature of it. What is distinct here is the per-shipment record.
Build an evidence trail
The shipment, the origin evidence under it and the manager who answered stay on the pack.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Despatch and shipping
- Weighbridge
- seals
- Bills of lading
- draft survey
Assay and laboratory
- LIMS
- assay certificates
- Sampling records
Commercial systems
- SAP
- Oracle ERP
- Offtake and sales records
Agent
Supply-chain evidence
- Reads the records
- Assembles the pack
- Holds for the manager
Schemes and templates
- CMRT
- EMRT
- AMRT
- Scheme portals
- audit files
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 customer
Each of the six sits inside the last. What none of them stops appears in the map beneath it.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether the pack may stand
- L3
- Puts the release in a manager's hands
- L4 – L5
- Keep the shipment and the evidence behind it
- L6
- Holds the pack at unanswered when signals degrade
How Nestack evaluates it
Evaluate the assembly workflow — not only the finished pack.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the answer the customer reads
- E1Final-output evaluationDid every custody link trace to a document on file?
- E2Step-level evaluationDid the agent use the right scheme, cycle and evidence configuration?
- E3Tool evaluationDid it read and write the correct shipment and the correct counterparty?
- E4Confidence calibrationDo low-confidence answers actually attract more manager amendments?
- E5Slice evaluationHow does performance change across specific counterparty groups?
- E6Business outcomeHow many answers were amended, or corrected after they were sent?
Floor — the outcome the operation answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed where it first enters the pack.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- EF-03
Ticket never ingested
The custody link exists only on a phone.
Stage gathers
Despatch, weighbridge, assay and lease records
02 · Reasoning
2 modes
- EF-04
Template default origin
A blend is declared under the lease the template names.
- EF-06
Criterion mapped by analogy
A scheme answer is drawn from a near-equivalent criterion.
Stage proposes
Custody chain, supporting documents, confidence
03 · Tool / write
2 modes
- EF-02
Prior-year answer reused
The routing changed and the answer did not.
- EF-05
Certificate quietly expired
An answer still leans on a lapsed document.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- EF-01
Custody chain breaks
A transfer between two records has no document.
Stage returns
The evidence pack the manager answers from
05 · Change / Version
1 mode
- EF-07
Silent scheme drift
A rule change alters which evidence an answer needs.
Stage tracks
Model, prompt, scheme rules and evidence config
- Sev-1 · an answer sent outside the boundary
- Sev-2 · an unsupported claim reaches a buyer
- Sev-3 · source degrades, pack goes to review
Affected slices
A clean coverage figure can hide one counterparty
A shipment-level evidence-coverage figure can read as complete while a few counterparty and shipment types carry most of the amendments. Nestack reports by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Blended stockpile parcels | 6.4% | 3.7× | Review | |
| Refiner and smelter audits | 4.9% | 2.8× | Review | |
| Forced-labour tracing requests | 3.4% | 1.9× | Watch | |
| Routine offtake despatches | 1.2% | 0.7× | Normal |
Bar: amendment-rate lift vs. routine-despatch baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The loop closes on a new test
A cycle closes when the unanswerable question is a regression case. That suite is what the next pack assembled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Amendment rate rises for one counterparty slice.
- 02
Diagnose
The eleventh customer asking the same question a different way is read back through the packs until the cause narrows to one.
- 03
Improve
The fix carries a number, and the shipments that forced it travel with it.
- 04
Verify
Nothing goes out until every touched shipment case has passed again.
- 05
Learn
It is kept permanently, and the evidence rules are amended alongside it.
Learn → Detect
The return edge. Detection next time starts from 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, evidence workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Evidence workflow discovery and boundary mapping.Week 1
- 02Despatch and assay source assessment.Week 1, Week 2
- 03Scheme, certificate and audit-cycle rule mapping.Week 1, Week 2
- 04Record ingestion and custody joining.Week 2
- 05Pack assembly and source binding.Week 2, Week 3
- 06Confidence scoring and gap routing.Week 3
- 07Manager review workflow.Week 3, Week 4
- 08Despatch and assay-system integration.Week 5
- 09Custody and questionnaire cases.Week 4, Week 5
- 10Guardrails and disclosure controls.Week 4
- 11Shipment-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 commodity, one route | ProductionProduction source systems | AdvancedMultiple sites / schemes |
|---|---|---|---|
| Introduced at Pilot | |||
| Evidence assembly per shipment | ✓ | ✓ | ✓ |
| Manager answer approval | ✓ | ✓ | ✓ |
| Custody-continuity baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by counterparty | — | ✓ | ✓ |
| Exception workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Despatch-and-assay integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-scheme mapping rules | — | — | ✓ |
| Multi-stage commercial approvals | — | — | ✓ |
| High shipment volume | — | — | ✓ |
| Multi-scheme evidence 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 shipment types and counterparties
→Scheme, certificate and evidence rule mapping
Week 102Representative prior questionnaire responses
→Record ingestion, custody joining and the assembly baseline
Week 203Your origin-claim policy and evidence standards
→Evidence-pack mapping and automation-boundary definition
Week 104Access to relevant APIs, feeds or exports
→Despatch and assay-system assessment, then integration setup
Week 205Packs you would not want re-checked
→Certificate cases and failure-mode testing
Week 406What no origin claim may assume
→Confidence scoring, gap routing, guardrails and answer controls
Week 307Named managers to review the evidence pack
→Manager review 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
Phase widths follow the calendar the work needs, so the doubling seen at week five is not padding at all.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Evidence workflow discovery, scheme mapping and the automation boundary
W2Source integration and the custody-assembly baseline
W3Assembly workflow, confidence logic and answer controls
W4Evaluation suite, certificate checks and failure-mode testing
W5Assay-system integration, a pilot shipment and targeted corrections
W6One shipment cycle run under the commercial manager, then Agent Care handover
- Reading the band
- A bar covers only the weeks its own work is named in. The week 5 overlap is real, not padding.
- At the end of W6
- When the shipment pack validates, Agent Care takes the agent on.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Mining AI agent
Build the evidence pack before the next questionnaire arrives.
Show us your despatch records, the schemes you are enrolled in and who answers. The certification is carried downstream — by the smelter, the refiner, the listed manufacturer, the importer of record — never by the mine. When an answer that already went out proves unsupported, the manager withdraws it, and the pack shows what they were shown when they gave it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your despatch and assay data
- Manager answers
- Evidence only
- After launch
- Agent Care available
Nestack Agents · Minerals supply-chain evidence
AGT-MN-18 · Agent Care available after launch