Main site

Mining AI agent · Reserve reporting

Mineral Resource & Reserve Reporting AI Agent

Assemble the drill-hole lineage, the QA/QC record and the modifying-factor evidence behind a resource and reserve statement, mark what the pack does not support, and hold it for the Competent Person who consents.

Typical delivery
4–6 weeks
Deployment
Your stack
CP consents
Pre-consent
After launch
Agent Care

What this agent does

Assembles the evidence, not the classification

In

  1. 01

    A hole is logged, and its assays, QA/QC inserts and the extract behind the estimate are pulled with it.

  2. 02

    A statement is prepared, and each figure is traced to the database extract current at the effective date.

Reason

  1. 03

    A Table 1 section is opened by activity under JORC 2012, mandatory since 1 December 2013, and silent rows listed.

  2. 04

    A reserve is converted, and each modifying factor is set beside the study, testwork or permit status behind it.

  3. 05

    An announcement repeats a figure, and it is matched to the statement, the consent and any drilling since.

Decide

  1. 06

    A prior statement is compared, and the movement is split by depletion, model, price and method, never netted.

  2. 07

    A consent goes on file under NI 43-101 s.8.3 as consolidated 9 June 2023, with the document version read.

Out

  1. 08

    A study level is set, and the technical report summary item set follows 17 CFR 229.601(b)(96), which scales.

  2. 09

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

Product statement

The agent assembles the support; the assay copilot proves the data on import, and a named Competent Person consents to the figure.

Example workflow

One statement, data to consent

  1. 1Estimate receivedDrill-hole database, assay certificates, model run and effective date
  2. 2Lineage assembledQA/QC records, Table 1 support, modifying-factor evidence and tenure status
  3. 3Pack draftedLineage map, silent Table 1 rows, open gaps and confidence
  4. 4Controls appliedSource-trace checks, code-section checks, consent-currency checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is reported at any of them — the agent is assembling, and the Competent Person's lane opens at the confidence gate.

5DecisionSplits at the confidence threshold
High confidence

Goes to the Competent Person to read.

Low confidence

Adds a peer-review geologist read first.

Competent Person review

The pack is held with its lineage, its silent rows and the confidence.

Accept · Annotate · Send to peer review

Accepted — pack complete

  1. 6Reporting record updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedLineage coverage, annotations, silent rows and post-statement corrections

Annotations

Every Competent Person annotation is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Classifying a resource or assigning a category.
    • Giving or withholding a Competent Person consent.
    • Deciding a Table 1 row is not applicable.
    • Setting the metallurgical recovery assumption.

  • Automation boundary

    Agent acts unaided

    • ✓Trace each reported figure to the database extract behind it.
    • ✓Assemble the QA/QC record and the modifying-factor evidence.
    • ✓Show which Table 1 rows the source pack leaves silent.
    • ✓Split the movement since the last statement by cause, and hold the pack.

    Any write happens inside the boundaries agreed at implementation, never ahead of the consent.

    • Determining the effective date of an estimate.
    • Signing, dating or sealing a technical report.
    • Judging whether a person meets the experience test.
  • Changes to code mapping or reporting thresholds.

Example output

One reported figure, annotated

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

Reserve pack output · single deposit

Illustrative example
Deposit
Open-pit gold deposit
Reported line
Indicated Resource carried into the annual statement
Effective date
30 June 2026
Database extract
Signed-off model run
Confidence
91%
Responsible person
Named Competent Person and organisation

As received

Taken from the drill-hole database and the signed model run — nothing on this side is written by the agent.

Evidence assembled

  • Assay certificates
  • QA/QC insert results
  • Survey and collar file

What supports this line

It traces to the extract at the effective date — the line the Competent Person weighs.

What the score decides

Below the configured threshold the pack picks up a peer read before it reaches the Competent Person.

Action

AcceptAnnotateSend to peer 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 reported figure

From the drill-hole database

03Assembly

Assemble from the record

Draw on the extract at the effective date, the QA/QC record and the studies on file.

01Approved path

The Competent Person carries it

Routine lineage, Table 1 support and cross-references arrive already assembled.

02Human review

Send review to the thin support

Silent rows and low-confidence packs are marked, so the Competent Person starts where the support is thinnest.

04

Build an evidence trail

The figure, the drill data beneath it and the person who consented stay on the statement.

Integrations

Typical integrations

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

  • Drill-hole databases

    • acQuire
    • Datamine
    • Micromine
    • Maxwell DataShed

  • Resource modelling

    • Leapfrog
    • Vulcan
    • Surpac
    • Deswik
    • Micromine

  • Laboratory systems

    • LIMS assay certificates
    • ALS
    • SGS
    • Bureau Veritas

Agent

Resource & reserve reporting

  • Reads the database
  • Assembles the pack
  • Holds for consent

  • Disclosure and records

    • Document management
    • Announcement and filing systems

  • 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 statement

Each of the six sits inside the last. What none of them catches appears in the map beneath this.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modePull assembly back to lineage-only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and code-mapping changes — NI 43-101 is proposed for repeal and replacement, not amendment.Track
L4TraceabilityRecord the evidence, the pack, the silent rows and the consent time.Record
L3Competent PersonHold packs for the named Competent Person; a wrong figure after consent still unwinds as a correction.Gate
L2Policy guardrailsTest packs against the section of JORC 2012 that attaches to the activity; a failure returns the pack.Restrict
L1Confidence thresholdsRoute low-confidence packs to a peer read before the Competent Person sees them.Require review
Model corePack produced — lineage map, modifying-factor evidence, silent rows and confidence
L1 – L2
Test whether a pack may stand
L3
Puts the consent in a named person's hands
L4 – L5
Keep the figure and the drill data behind it
L6
Reverts to lineage assembly when signals degrade

How Nestack evaluates it

Evaluate the whole reporting workflow — not only the statement that ships.

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

Surface — the figure the market reads

  1. E1Final-output evaluationDid every reported figure trace back to a source record?
  2. E2Step-level evaluationDid the agent read the extract current at the stated effective date?
  3. E3Tool evaluationDid it read and write the correct deposit and the correct model run?
  4. E4Confidence calibrationDo low-confidence packs actually attract more annotations?
  5. E5Slice evaluationHow does performance change across specific deposits and estimate types?
  6. E6Business outcomeHow many packs were corrected, retracted or refiled after the statement went out?

Floor — the statement the company answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed at the stage where it originates.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • DJ-03

      Extract taken after the date

      The model is re-run on a later database.

    Stage gathers

    Drill-hole data, QA/QC, studies and tenure status

  2. 02 · Reasoning

    2 modes

    • DJ-04

      Unsupported Table 1 row

      A silent row is passed on without its reason.

    • DJ-06

      Factor extrapolated silently

      A recovery is carried into untested domains.

    Stage proposes

    Lineage, silent rows, movement and confidence

  3. 03 · Tool / write

    2 modes

    • DJ-02

      Movement netted to one line

      Offsetting causes are shown as one change.

    • DJ-05

      Stale consent carried on

      The document changed after it was read.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • DJ-01

      Figure without lineage

      A reported figure has no traceable source.

    Stage returns

    The pack the Competent Person reads before consenting

  5. 05 · Change / Version

    1 mode

    • DJ-07

      Silent code regression

      A mapping change drops a row the section requires.

    Stage tracks

    Model, prompt, code mapping and thresholds

  • Sev-1 · acts outside the reporting boundary
  • Sev-2 · an unsupported figure is reported
  • Sev-3 · source degrades, pack goes to review
See the complete Mining failure-mode catalogue

Affected slices

A clean statement can hide one deposit

A statement-level lineage-coverage figure can read complete while a few deposits carry almost all of the unsupported rows. Nestack reports the unsupported-figure rate by deposit, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Deposits inherited by acquisition8.4%3.7×Review
First-time reserve conversions6.7%2.9×Review
Underground extensions at depth3.6%1.6×Watch
Established open-pit models1.6%0.7×Normal

Bar: unsupported-figure-rate lift vs. established open-pit baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Nothing closes without a new case

The loop shuts when the unsupported figure is a regression case. That suite is what the next statement prepared is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Unsupported-figure rate rises in a deposit.

  2. 02

    Diagnose

    The tonne that was reported before the assay behind it was reconciled is read back until one cause is left.

  3. 03

    Improve

    Any change goes out numbered, with the figures that caused it attached.

  4. 04

    Verify

    One statement case still failing is enough to hold the release back.

  5. 05

    Learn

    One case joins the suite, one line joins the estimation record.

  6. Learn → Detect

    The return edge. The next statement is prepared 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, assembly workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Reporting workflow discovery and boundary work.Week 1
  2. 02Database and model-source assessment.Week 1, Week 2
  3. 03Code-section, Table 1 and consent-record mapping.Week 1, Week 2
  4. 04Extract ingestion and lineage tracing.Week 2
  5. 05Evidence assembly and pack logic.Week 2, Week 3
  6. 06Confidence scoring and gap routing.Week 3
  7. 07Competent Person review workflow.Week 3, Week 4
  8. 08Resource-model and document integration.Week 5
  9. 09Lineage and modifying-factor cases.Week 4, Week 5
  10. 10Guardrails and consent controls.Week 4
  11. 11Estimate-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 deposit, one codeProductionProduction reporting systemsAdvancedMultiple deposits / codes
Introduced at Pilot
Assembly to your data and code✓✓✓
Competent Person review✓✓✓
Lineage-completeness baseline✓✓✓
Introduced at Production
Reporting by deposit—✓✓
Review workflow in your systems—✓✓
Approved write-back—✓✓
Resource-model integration—✓✓
Introduced at Advanced
Multi-code disclosure rules——✓
Multi-stage technical review——✓
High deposit count——✓
Multi-code disclosure 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 drill-hole database and model outputs

    Extract ingestion and lineage tracing

    Week 1

  2. 02Representative past statements

    Assembly baseline, evidence mapping and source binding

    Week 2

  3. 03Your code mapping and Table 1 practice

    Code-section, Table 1 and consent-record mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Database and model-source assessment, then integration setup

    Week 2

  5. 05Statements you would not want re-opened

    Reconciliation cases and the evaluation suite

    Week 4

  6. 06What no consent may cover

    Confidence scoring, gap routing, guardrails and review controls

    Week 3

  7. 07Named Competent Persons to read packs

    Competent Person 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

Each band covers the weeks its own phase really needs, and that is why week five holds two of them.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Reporting workflow discovery, code mapping and the automation boundary

W2Source integration and the assembly baseline

W3Assembly workflow, confidence logic and review controls

W4Evaluation suite, lineage checks and failure-mode testing

W5Document integration, pilot deposits and targeted corrections

W6One reporting cycle run under the Competent Person, then Agent Care handover

Reading the band
A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
At the end of W6
Once the statement validates, Agent Care assumes the running agent.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Mining AI agent

Build a reserve-reporting agent around the Competent Person's consent.

Show us your drill-hole database, your last statement and the people who consent. A named Competent Person carries every reported figure, and this is the page our geological assay copilot defers to when it says so.

Typical delivery
4–6 weeks
Deployment
Your database and models
CP consents
Pre-consent
After launch
Agent Care available

Nestack Agents · Resource & reserve reporting

AGT-MN-10 · Agent Care available after launch

Build a reserve-reporting agent

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

About you
Your statements

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