Main site

Mining AI agent · Geological & assay data

Geological & Assay-Data AI Agent (Sample to Database)

Validate drill-hole and sample data on import, reconcile assay certificates against the database and score the QA/QC controls — handing a geologist the exceptions, while a named Competent Person carries every reported figure.

Typical delivery
4–6 weeks
Deployment
Your stack
Reporting sign-off
Named person
After launch
Agent Care

What this agent does

Checks the data behind the grade

In

  1. 01

    Take drill-hole, collar, survey, logging and sample records as they arrive from the field or the geological database.

  2. 02

    Read assay certificates as the laboratory issued them, carrying the method, the detection limits and the batch.

Reason

  1. 03

    Reconcile each certificate value against the database entry it became, and mark where the two disagree.

  2. 04

    Score the batch's standards, blanks, field and pulp duplicates and umpire results against the programme's own limits.

  3. 05

    Check collars, downhole surveys, sample intervals and logging codes for the contradictions that move a sample.

Decide

  1. 06

    Flag batches that fail their own control limits, samples with no certificate behind them, and method changes.

  2. 07

    Route what it cannot resolve to the project geologist, the database administrator or the laboratory.

Out

  1. 08

    Present the exception pack — each item beside the certificate, the batch and the control that failed.

  2. 09

    Retain every certificate read, the check applied, the rationale and every geologist correction.

Product statement

The agent checks, reconciles and flags. Accepting a batch, releasing a dataset for estimation and every reported figure stay with the named Competent Person.

Example workflow

One assay batch, end to end

  1. 1Batch receivedA laboratory certificate, a logging upload or a survey file arriving against a drill programme
  2. 2Records resolvedHole, collar, survey, sample interval, batch, method and detection limit as the certificate states them
  3. 3Checks runCertificate reconciled to the database, QA/QC controls scored, collar and survey geometry checked
  4. 4Exceptions markedControl failures, samples with no certificate, interval gaps and overlaps, and changes of method

No human action required

Stages 1 to 4 run without a person in the loop — reading, reconciling and scoring the controls finish before anyone is asked to look at anything. Nothing is accepted in that stretch.

5DecisionSplits on check confidence and whether a control limit was breached
Controls inside their limits

Reaches the geologist as a checked batch.

Control failure or an unmatched value

Goes to the project geologist before anything uses it.

Project geologist

Reads the exceptions, the certificates behind them and the controls that failed, then decides what happens to the batch.

Accept findings · Correct · Request re-assay

Findings accepted — handed back

  1. 6Pack routed to the geologistHanded over as exceptions; no batch is accepted or rejected and no dataset is released by the agent
  2. 7Outcome evaluatedWhat the geologist corrected, what the re-assay returned, and what the database audit found by cohort

Corrections

Values a geologist corrects at review are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Estimating or classifying a Mineral Resource or Ore Reserve.
    • Reporting any figure as an Exploration Result.
    • Signing off a dataset as fit for estimation.
    • Accepting or rejecting an assay batch.

  • Automation boundary

    Agent acts unaided

    • ✓Reconcile each certificate value against the database entry it became.
    • ✓Score the batch's standards, blanks and duplicates.
    • ✓Check collar, survey and interval data for errors.
    • ✓Assemble the exceptions with the failed control beside each one.

    Write actions run only inside the approval boundaries agreed during implementation. A reported figure is never one.

    • Overriding a QA/QC control limit or failure rule.
    • Changing a logging code, standard or sampling protocol.
    • Editing a certificate value held in the database.
  • Deciding a sample swap is only a labelling error.

Example output

One assay batch, annotated

Everything the agent flags is attached to the certificate and the sample it came from.

Exception pack · one certificate batch

Illustrative example
Batch as received
Certificate, 240 samples
Method as stated
Fire assay, 30 g charge
Detection limit
0.01 ppm
Flagged
Certificate mismatch
Confidence
91%
Batch
Not accepted by the agent

As received

The batch, the method and the detection limit as the certificate states them — nothing on this side is re-keyed.

Evidence used

  • Laboratory certificate
  • Batch QA/QC inserts
  • Sample dispatch record

Why it is held

The certificate and the database entry for that sample disagree.

What the score decides

Confidence decides how hard the geologist reads the pack, not whether it is accepted.

Action

Accept findingsCorrectRequest re-assay

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 batch entering the database

Certificate, logging upload or survey file

03Checking & reconciling

Check what the laboratory issued

Use the programme's sampling protocol, insertion rates and control limits, and the method the laboratory states for that batch.

01Approved path

Find the error nobody thought to query

A swapped sample or a re-keyed value looks like ordinary data, so the check runs across every batch rather than the ones someone already doubted.

02Human review

Send the failed controls up first

Batches that breached their own limits, and values with no certificate behind them, reach the geologist before the data reaches a model.

04

Build an evidence trail

Retain the certificates read, the controls scored, the exceptions raised, the confidence and every geologist correction — on both paths.

Integrations

Typical integrations

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

  • Geological database

    • acQuire
    • DataShed
    • Geobank
    • MX Deposit
    • core logging tools

  • Laboratory & certificates

    • ALS
    • SGS
    • Intertek
    • Bureau Veritas
    • LIMS certificate feeds

  • Modelling & estimation

    • Micromine
    • Datamine
    • Leapfrog
    • Vulcan
    • Surpac

Agent

Geological & assay data

  • Reads certificates
  • Reconciles the batch
  • Routes exceptions

  • Survey & field capture

    • Collar surveys
    • downhole gyro tools
    • Field tablets
    • sample dispatch

  • 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 database

Each control wraps the one inside it. A batch clears every layer before a geologist reads it, and the reported figure sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn checking to the geology team if evaluations or production signals degrade.Roll back
L5Change controlRecord the certificate, the protocol version, the check applied, the named approver and every correction.Record
L4Competent-Person gateAccepting a batch, releasing a dataset and reporting any figure stay with the named person.Gate
L3Open-exception holdA batch with an open exception stays marked unresolved until a geologist closes it.Hold
L2Control limitsStandards, blanks, duplicates and umpire results are scored against the programme's own limits.Flag
L1Certificate as sourceValues are scored against the certificate the laboratory issued, not the database entry; one with no certificate behind it is raised.Cite
Model coreCheck result produced — reconciliation, QA/QC scores, data-integrity flags and confidence
L1 – L2
Decide whether a value may stand
L3
Decides what a geologist must close
L4 – L5
Keep reporting with a person and the record intact
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the whole check — not only the pack a geologist opens.

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

Surface — the exception pack the geologist opens

  1. E1Final-output evaluationDid the pack carry the exceptions that actually mattered?
  2. E2Step-level evaluationDid it read the right certificate, batch, method and limit?
  3. E3Tool evaluationDid it resolve the correct hole, interval and sample number?
  4. E4Control recallHow many control failures were missed against the audited batch?
  5. E5Slice evaluationHow does check quality change across specific batch cohorts?
  6. E6Business outcomeHow much was corrected, and what did the re-assays return?

Floor — the database an estimate would be built from

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    2 modes

    • GA-01

      Certificate not reconciled

      A database entry differs from the certificate and passes.

    • GA-02

      Superseded certificate used

      The laboratory reissues a certificate and the first one stands.

    Stage gathers

    Certificates, sample records, survey files and the protocol

  2. 02 · Reconciling

    2 modes

    • GA-03

      Method change read as geology

      A change of assay method is scored as a change in the ground.

    • GA-04

      Over-limit result carried

      A value at the upper detection limit is used as if measured.

    Stage proposes

    Matched values, QA/QC scores and data-integrity flags

  3. 03 · Verification

    1 mode

    • GA-05

      Sample swap not caught

      Two exchanged samples each look ordinary on their own.

    Stage checks

    Control limits, method changes, interval gaps

  4. 04 · Handoff / write

    1 mode

    • GA-06

      Real failure buried in the pack

      Minor queries push the batch that actually failed down the list.

    Stage presents

    The exception pack the geologist reads and its queue

  5. 05 · Change / Version

    1 mode

    • GA-07

      Control limit quietly widened

      A limit change reaches scoring with no named approver.

    Stage tracks

    Model, prompt, protocol and control-limit changes

  • Sev-1 · bad data could reach an estimate
  • Sev-2 · a wrong value reaches the geologist
  • Sev-3 · checking degrades, more is corrected
See the complete Mining failure-mode catalogue

Affected slices

Failures follow the changes, not the volume

Which batches does this miss? Not grade control, which runs the same method every week — the ones where the method changed, the data is old, or the tool sat by steel. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Batches after a method change4.8%3.1×Review
Historic holes, legacy database3.5%2.3×Review
Magnetic downhole surveys2.9%1.9×Watch
Routine grade-control batches1.2%0.8×Normal

Bar: corrected-or-re-assayed item rate, lift vs. routine grade-control baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

One corrected value does not close the batch

Correcting the value leaves the reason it got there untouched. The check, the control limit or the import rule behind it changes under change control.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Geologist corrections or re-assay results move in a cohort.

  2. 02

    Diagnose

    Nothing is corrected until the check that let it through is named.

  3. 03

    Improve

    The check or the limit goes through change control, with an approver.

  4. 04

    Verify

    Run again over archived batches whose audit result is already known.

  5. 05

    Learn

    The missed exception is added to the protocol later batches are scored against.

  6. Learn → Detect

    The return edge. Nothing shipped here changes what a Competent Person must still decide.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, certificates and database access, reconciliation and control scoring, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Sampling protocol and QA/QC limit mapping.Week 1, Week 2
  3. 03Laboratory certificate and LIMS access.Week 2
  4. 04Drill-hole database and survey data access.Week 2
  5. 05Certificate-to-database reconciliation.Week 2, Week 3
  6. 06Standard, blank and duplicate scoring.Week 3
  7. 07Collar, survey and interval validation.Week 3, Week 4
  8. 08Method and detection-limit change detection.Week 4
  9. 09Evaluation suite, control recall and regression cases.Week 4, Week 5
  10. 10Geologist review and correction capture.Week 5
  11. 11Exception pack and database write-back.Week 5, Week 6
  12. 12Observability, deployment 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 programme, one laboratoryProductionProduction database integrationAdvancedMulti-site / multi-commodity
Introduced at Pilot
Certificate-to-database reconciliation✓✓✓
Standard, blank and duplicate scoring✓✓✓
Umpire and check-assay comparison✓✓✓
Collar, survey and interval validation✓✓✓
Method and detection-limit change flags✓✓✓
Reporting stays with the Competent Person✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Geologist review workflow and write-back—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Multi-site and multi-commodity programmes——✓
Multiple laboratories and database systems——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on certificate and database integrations, the number of programmes, laboratories and commodities, the QA/QC scheme in use, review 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 sampling and QA/QC protocol, with its insertion rates

    Sampling protocol and QA/QC limit mapping

    Week 1

  2. 02The reporting boundary and who your Competent Person is

    Workflow discovery and automation-boundary definition

    Week 1

  3. 03Access to the laboratory's certificates or its LIMS feed

    Laboratory certificate and LIMS access

    Week 2

  4. 04Access to the drill-hole database, collars and survey files

    Drill-hole database and survey data access

    Week 2

  5. 05The standards, blanks and duplicates you insert, and their limits

    Standard, blank and duplicate scoring

    Week 3

  6. 06Batches that failed a control, and what you did about them

    Evaluation suite, control recall and regression cases

    Week 4

  7. 07Named geologists to read the exception packs

    Geologist review workflow, then pilot batches 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

Phases are drawn over the weeks they actually occupy. Week 5 carries both the evaluation work and the first packs a geologist reads.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Sampling protocol, QA/QC limits and the automation boundary

W2Laboratory certificates, the drill-hole database and survey files wired in

W3Reconciliation and control scoring on the first drill programme

W4Evaluation suite, control recall and method-change detection

W5Database write-back, supervised batches and targeted corrections

W6Corrections made by the geologists who own the data, then handover

Reading the band
Week 4 scores the checks against batches you have already audited, so the recall figure comes out of your data, not ours.
At the end of W6
Geologists who own the data have worked live packs and read corrections back, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Mining AI agent

Build a geological data agent around your sampling protocol.

Show us your sampling and QA/QC protocol, one closed drill programme and the database it landed in. That programme is reconciled against the certificates it was built on, and the disagreements come back named, with the certificate behind each one.

Typical delivery
4–6 weeks
Deployment
Your geological data stack
Reporting sign-off
Named Competent Person
After launch
Agent Care available

Nestack Agents · Geological & assay data

AGT-MN-03 · Agent Care available after launch

Build a geological data agent

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

About you
Your setup

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