Main site

Mining AI agent · Ground control records

Ground Control Decision-Record AI Agent

Reconstruct what the monitoring showed when a ground decision was taken, which trigger rule was in force that day and who was notified — the statutory official on shift decides the ground.

Typical delivery
4–6 weeks
Deployment
Your stack
Named officials
Post-decision
After launch
Agent Care

What this agent does

Reconstructs the record, not the decision

In

  1. 01

    A trigger is crossed, and the reading is pulled from supported seismic, radar, extensometer and survey sources.

  2. 02

    A threshold is read against the trigger table in force on the day, not the table that is in force now.

Reason

  1. 03

    An alarm fires, and the escalation the plan required at that level is set beside the notifications that went out.

  2. 04

    A re-entry is called, and the record holds the named official, the stated basis and the clock they ran.

  3. 05

    An exclusion zone is declared, and the basis for declaring it and for releasing it are both carried forward.

Decide

  1. 06

    A plan review falls due, and the cycle applied is the one that jurisdiction sets, not one cadence group-wide.

  2. 07

    A sector is mined, and the approval covering it is read against the rock mass actually being extracted.

Out

  1. 08

    Support is installed, and the as-installed record is reconciled against the design revision it was cut from.

  2. 09

    A write action runs only inside the approval boundaries agreed during implementation.

Product statement

The agent reconstructs the record; a named statutory official decides the ground, and a record found wrong is corrected after people were already underground.

Example workflow

One decision, alarm to record

  1. 1Alarm or trigger receivedSeismic system, radar, extensometer, inspection and shift records
  2. 2Record assembledTriggers in force, plan requirements, notifications and roster, each with its named source
  3. 3Pack draftedDecision window, escalation record, open items and confidence
  4. 4Controls appliedTrigger-version checks, escalation-rule tests, notification completeness and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is decided at any of them — the agent is reconstructing a record, and the official lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the statutory official.

Low confidence

Adds a geotechnical read first.

Official review

The pack is held with its readings, its open items and the confidence.

Approve · Amend · Send to geotechnical review

Approved — released to the record

  1. 6Ground control record updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedAmendment distance, open-item outcomes, review findings and post-decision corrections

Amendments

Every official amendment is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Predicting a ground failure or a rockburst.
    • Clearing a re-entry, or lifting an exclusion.
    • Setting, tuning or re-baselining a trigger level.
    • Certifying a ground support design.

  • Automation boundary

    Agent acts unaided

    • ✓Reconstruct the decision window from the record the shift saw.
    • ✓Set each reading against the trigger table in force that day.
    • ✓Name who the plan required to be told, and whether they were.
    • ✓Reconcile designed support against installed, and hold the pack.

    Writes run only inside the boundaries agreed at implementation, and never ahead of a decision.

    • Raising, suppressing or standing down an alarm.
    • Authorising the plan — the named statutory role.
    • Notifying the inspectorate of a ground event.
  • Changes to trigger, escalation or approval rules.

Example output

One ground decision, annotated

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

Decision record output · single decision

Illustrative example
Ground domain
Seismically active panel
Decision reconstructed
Re-entry authorised after a seismic exclusion
Time to escalation
41 minutes
Source of record
Seismic catalogue on file
Confidence
88%
Decision of record
Underground manager, named on shift

As received

Taken from the shift log and the event catalogue — nothing on this side is written by the agent.

Source facts used

  • Trigger table, versioned
  • Shift notification log
  • Seismic event record

Why this is open

The plan required a geotechnical escalation at that level, and the log does not show one.

What the score decides

Below the configured threshold the item picks up a geotechnical read before the official sees it.

Action

ApproveAmendSend to geotechnical 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 ground decision

From the monitoring record

03Reconstruction

Rebuild the decision window

Readings are placed beside the trigger table in force that day, and beside the plan requirement that applied at that level.

01Approved path

Reconstruct what was known

Press reporting of the Sernageomin El Teniente report puts the failure in a missing mandatory escalation, not in the array. The agent shows which rule was in force.

02Human review

Not the alarm, and not the permit

Our process control copilot works the live plant loop; our safety compliance assistant owns the SWMS, isolation and permit packs. This page owns the basis of a ground decision.

04

Build an evidence trail

The alarm, the plan it was read against and the official who decided 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.

  • Seismic monitoring

    • IMS
    • ESG Solutions
    • Microseismic arrays

  • Deformation monitoring

    • Slope stability radar
    • Extensometers
    • prisms

  • Mine planning and survey

    • Deswik
    • Vulcan
    • As-mined survey data

Agent

Ground control decision record

  • Reads the record
  • Rebuilds the window
  • Holds for review

  • Plans and shift records

    • GCMP
    • TARP
    • SRMP files
    • Shift logs and rosters

  • 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 record

Each of the six sits within the last. What none of them holds shows in the map beneath this.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow to record reconstruction when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, trigger-table and plan-version changes.Track
L4TraceabilityRecord the readings, the rules applied, the open items and the approval time.Record
L3Official approvalHold the pack for the named statutory official; it governs release of the pack, not whether the ground held.Gate
L2Policy guardrailsTest the pack against the escalation and review-cycle rules on file; a failure returns the pack.Restrict
L1Confidence thresholdsRoute low-confidence items to a geotechnical read before the official sees them.Require review
Model corePack assembled — decision window, escalation record, open items and confidence
L1 – L2
Test whether the pack may stand
L3
Puts the release in the hands of an official
L4 – L5
Keep the alarm and the plan behind it
L6
Narrows to record reconstruction when signals degrade

How Nestack evaluates it

Evaluate the reconstruction workflow — not only the finished pack.

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

Surface — the record the official reads

  1. E1Final-output evaluationDid every item trace to the reading or record behind it?
  2. E2Step-level evaluationDid the agent use the trigger table and plan version in force that day?
  3. E3Tool evaluationDid it read and write the correct panel and the correct shift?
  4. E4Confidence calibrationDo low-confidence items actually attract more official amendments?
  5. E5Slice evaluationHow does performance change across specific ground domains?
  6. E6Business outcomeHow many items were amended, or corrected after the decision had been acted on?

Floor — the outcome the operator answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed where it first enters the record.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • DW-03

      Post-processed catalogue

      The raw feed the control room saw is not the one read.

    Stage gathers

    Readings, trigger tables, plan versions and logs

  2. 02 · Reasoning

    2 modes

    • DW-04

      Superseded trigger table

      A revised level is applied back across the year.

    • DW-06

      Design reconciled to itself

      The plan changed and the model was never told.

    Stage proposes

    Decision window, escalation record, confidence

  3. 03 · Tool / write

    2 modes

    • DW-02

      Zone declared verbally

      A zone has a start with no recorded basis.

    • DW-05

      Notification to a vacant role

      The plan named a role the roster left empty.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • DW-01

      Escalation not evidenced

      The plan required a call the log cannot show.

    Stage returns

    The decision record the statutory official reads

  5. 05 · Change / Version

    1 mode

    • DW-07

      Silent threshold drift

      A model or rule change alters what reads as out of range.

    Stage tracks

    Model, prompt, trigger tables and plan versions

  • Sev-1 · a wrong pack reaches a decision
  • Sev-2 · a wrong basis reaches the record
  • Sev-3 · source degrades, item goes to review
See the complete Mining failure-mode catalogue

Affected slices

One ground domain can carry most of the returns

A panel-level reconstruction rate can read as complete while a few ground domains carry most of the returned items. Nestack reports the return rate by ground domain, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Seismically active panels6.6%3.7×Review
Block-cave drawpoint ground4.5%2.5×Review
Development and access drives2.7%1.5×Watch
Stable, supported ground1.8%1.0×Normal

Bar: return-rate lift vs. stable-ground baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Each cycle leaves one more standing test

A cycle is done when the lost decision trail has become a case the next release must pass. That suite is what the next re-entry logged is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Return rate rises in one ground domain.

  2. 02

    Diagnose

    The alarm that fired at two in the morning and was explained six months later is read back until the cause narrows to one.

  3. 03

    Improve

    Stamp the change; the decisions behind it are filed against that number.

  4. 04

    Verify

    Each touched decision case is run again, and one red stops the release.

  5. 05

    Learn

    It stays a standing test, and the trigger rules move with it.

  6. Learn → Detect

    The return edge. Detection next time runs against one more case than last.

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, reconstruction workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Decision workflow discovery and boundary mapping.Week 1
  2. 02Monitoring and plan-source assessment.Week 1, Week 2
  3. 03Trigger, escalation and review-cycle rule mapping.Week 1, Week 2
  4. 04Reading ingestion and reconciliation.Week 2
  5. 05Decision-window assembly and binding.Week 2, Week 3
  6. 06Confidence scoring and exception routing.Week 3
  7. 07Official review workflow.Week 3, Week 4
  8. 08Monitoring and plan-system integration.Week 5
  9. 09Alarm and decision cases.Week 4, Week 5
  10. 10Guardrails and escalation controls.Week 4
  11. 11Decision-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 panel, one cycleProductionProduction monitoring systemsAdvancedMultiple mines / jurisdictions
Introduced at Pilot
Decision-record reconstruction✓✓✓
Statutory-official approval✓✓✓
Reconstruction-completeness baseline✓✓✓
Introduced at Production
Reporting by ground domain—✓✓
Exception workflow in your systems—✓✓
Approved write-back—✓✓
Monitoring-system integration—✓✓
Introduced at Advanced
Multi-jurisdiction plan rules——✓
Multi-stage ground approvals——✓
High panel count——✓
Multi-panel decision controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, panel count, 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 panels, ground domains and plans

    Trigger, escalation and review-cycle rule mapping

    Week 1

  2. 02Representative prior decision records

    Reading ingestion, reconciliation and the reconstruction baseline

    Week 2

  3. 03Your GCMP, TARPs and re-entry protocols

    Decision-record mapping and automation-boundary definition

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Monitoring and plan-source assessment, then integration setup

    Week 2

  5. 05Decisions you would not want reconstructed

    Re-entry cases and the evaluation suite

    Week 4

  6. 06What no ground decision may omit

    Confidence scoring, exception routing, guardrails and escalation controls

    Week 3

  7. 07Named statutory officials to review the pack

    Official 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

The widths come from the weeks the work actually occupies, which is the reason that week five overlaps.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Decision workflow discovery, rule mapping and the automation boundary

W2Source integration and the reconstruction baseline

W3Reconstruction workflow, confidence logic and review controls

W4Evaluation suite, escalation checks and failure-mode testing

W5Monitoring integration, a pilot panel and targeted corrections

W6One review cycle run under the statutory mine 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
Validation closes on live panels, and Agent Care picks up monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Mining AI agent

Build the ground decision record before someone hostile asks for it.

Show us your monitoring feeds, your trigger tables and the plan that governs escalation. The next step is one panel and one month: we reconstruct the ground decisions inside it, name what the record cannot answer, and set the automation boundary with you — the officials keep the decisions.

Typical delivery
4–6 weeks
Deployment
Your monitoring data
Named officials
Post-decision
After launch
Agent Care available

Nestack Agents · Ground control decision record

AGT-MN-14 · Agent Care available after launch

Build a ground control record agent

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

About you
Your ground control

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