Mining AI agent · Tailings governance
Tailings Storage Facility Governance AI Agent
Assemble the governance record a facility already owes: the engineer of record file, readings set against their triggers, the review cycle and the deviations — then hold it for the people who sign.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named signers
- Assembly only
- After launch
- Agent Care
What this agent does
Assembles the record, not the review
In
- 01
An instrument reports, and the reading is pulled from supported piezometer, survey, inspection and SCADA sources.
- 02
A reading is set against the trigger level current on the day it was taken, not the level current today.
Reason
- 03
A review falls due, and the interval is the stricter of the cycles that apply to that facility, not one global number.
- 04
A raise is completed, and the record is rebuilt against the as-built survey rather than the design geometry.
- 05
A change is approved, and it joins the deviation record the deviance accountability report is drawn from.
Decide
- 06
An OMS review date passes, and the assumptions inside the EPRP are checked as well as the date on its cover.
- 07
A disclosure round opens, and each per-facility field is traced back to the engineer of record file behind it.
Out
- 08
A pack is retained as a dated version, beside the readings behind it and what each named signer was shown.
- 09
A write action runs only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles and keeps the record current; the engineer of record reviews and signs, and the accountable executive approves the design criteria.
Example workflow
One facility, reading to review
- 1Readings receivedPiezometers, survey, inspection reports and SCADA history
- 2Record assembledTriggers, review dates, deviations and document currency, each with its named source
- 3Pack draftedSurveillance record, deviation log, open items and confidence
- 4Controls appliedTrigger tests, review-cycle checks, deviation completeness and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is certified at any of them — the agent is assembling a record, and the engineer lane opens at the confidence gate.
Goes to the engineer of record.
Adds a site read first.
Engineer review
The pack is held with its readings, its open items and the confidence.
Approve · Amend · Send to site review
Approved — released to the signers
- 6Governance record updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAmendment distance, open-item outcomes, review findings and post-disclosure corrections
Amendments
Every engineer 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 dam safety — the engineer of record.
- Setting or revising a trigger level.
- Clearing a TARP alarm, or standing one down.
- Approving design criteria — the accountable executive.
Automation boundary
Agent acts unaided
- ✓Assemble the surveillance record against the triggers on file.
- ✓Track the review cycle the consequence classification sets.
- ✓Name deviations from the design basis report, and date them.
- ✓Check the OMS manual and the EPRP for currency, and hold the pack.
Writes run only inside the boundaries agreed at implementation, and never ahead of a signature.
- Determining the consequence classification.
- Predicting whether a facility will fail.
- Notifying a regulator or a chief inspector.
- Changes to trigger, review or approval rules.
Example output
One governance item, annotated
Everything the agent assembles is attached to the document it was drawn from.
Governance record output · single item
Illustrative example- Facility
- Very high consequence
- Governance item
- OMS manual annual review, read against its own date
- Time since review
- 11 months
- Source of record
- OMS manual on file
- Confidence
- 86%
- Signer of record
- Engineer of record, named on the file
As received
Taken from the OMS manual and the review log — nothing on this side is written by the agent.
Source facts used
- OMS manual, dated
- Annual review log
- Current DSR record
Why this is open
The annual review under GISTM 2020 Requirement 6.4 has a date, and this one has passed it.
What the score decides
Below the configured threshold the item picks up a site read before the engineer sees it.
Action
ApproveAmendSend to site 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 facility
From the surveillance record
Read the data against the triggers
Readings are set beside the trigger levels already on file; the response stays with the TARP owner named in it.
The engineer of record signs
Design responsibility sits with the engineer of record — GISTM 2020, Requirement 10.4. Assembling evidence is not the review.
Independence is the instrument
A dam safety review is independent by design — GISTM 2020, Requirement 10.5. The agent prepares what a reviewer reads and replaces neither.
Build an evidence trail
The reading, the trigger it crossed and the engineer who reviewed stay on the surveillance file.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Geotechnical monitoring
- Piezometers
- inclinometers
- Vibrating-wire loggers
Survey and imagery
- Drone and LiDAR survey
- InSAR deformation feeds
Operational data
- SCADA
- historian
- Deposition and water balance
Agent
Tailings governance record
- Reads the record
- Assembles the pack
- Holds for review
Documents and disclosure
- OMS
- EPRP
- DBR files
- Disclosure and ESG 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 file
Six layers, outermost first. What the whole set still lets through is drawn in the map below.
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 the hands of an engineer
- L4 – L5
- Keep the reading and the trigger behind it
- L6
- Holds the dam file at unreviewed 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 record the engineer reads
- E1Final-output evaluationDid every item trace to the reading or document behind it?
- E2Step-level evaluationDid the agent use the right triggers, intervals and facility configuration?
- E3Tool evaluationDid it read and write the correct facility and the correct period?
- E4Confidence calibrationDo low-confidence items actually attract more engineer amendments?
- E5Slice evaluationHow does performance change across specific facility classes?
- E6Business outcomeHow many items were amended, or corrected after the disclosure went out?
Floor — the outcome the operator answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each set at the stage where it first arises.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- DG-03
As-built never ingested
The pack is assembled against design geometry.
Stage gathers
Readings, documents, triggers and review dates
02 · Reasoning
2 modes
- DG-04
Superseded trigger level
A revised level is not applied back across the year.
- DG-06
Deviations added, not assessed
Each change passed alone, and none was read together.
Stage proposes
Surveillance pack, deviation log and confidence
03 · Tool / write
2 modes
- DG-02
Old review interval carried
A reclassified facility keeps its former cycle.
- DG-05
Date received, not conducted
The review date published is the date it arrived.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- DG-01
EPRP checked by date only
Its inundation assumption is the superseded one.
Stage returns
The governance record the engineer of record reads
05 · Change / Version
1 mode
- DG-07
Silent trigger drift
A model or rule change alters what reads as an exceedance.
Stage tracks
Model, prompt, trigger rules and facility config
- Sev-1 · a wrong pack reaches a signature
- Sev-2 · a wrong date reaches the disclosure
- Sev-3 · source degrades, item goes to review
Affected slices
A clean portfolio total can hide one facility class
A portfolio-wide trigger-response rate can read as healthy while a few facility classes carry most of the returned items. Nestack reports the return rate by facility class, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Extreme-consequence facilities | 9.6% | 3.7× | Review | |
| Facilities raised in the year | 7.4% | 2.8× | Review | |
| Facilities moving to closure | 4.3% | 1.7× | Watch | |
| Stable, unraised facilities | 1.9% | 0.7× | Normal |
Bar: return-rate lift vs. stable-facility 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 test
A cycle closes when the missed trigger is a regression case. That suite is what the next surveillance pack assembled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Return rate rises for one facility class.
- 02
Diagnose
The reading that crossed a trigger on a Friday and was reviewed on a Tuesday is read back until the cause narrows to one.
- 03
Improve
The change ships with a number, and the triggers that forced it travel beside it.
- 04
Verify
Nothing releases while one touched surveillance case is still red.
- 05
Learn
It is kept for good, and the trigger rules are amended in the same commit.
Learn → Detect
The return edge. The next cycle runs 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, governance workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Governance workflow discovery and boundary mapping.Week 1
- 02Instrumentation and document assessment.Week 1, Week 2
- 03Trigger, review-cycle and deviation rule mapping.Week 1, Week 2
- 04Reading ingestion and reconciliation.Week 2
- 05Pack assembly and source binding.Week 2, Week 3
- 06Confidence scoring and exception routing.Week 3
- 07Engineer review workflow.Week 3, Week 4
- 08Instrumentation and document integration.Week 5
- 09Trigger and surveillance cases.Week 4, Week 5
- 10Guardrails and review controls.Week 4
- 11Surveillance-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 facility, one cycle | ProductionProduction monitoring systems | AdvancedMultiple sites / jurisdictions |
|---|---|---|---|
| Introduced at Pilot | |||
| Governance record and reconciliation | ✓ | ✓ | ✓ |
| Engineer-of-record approval | ✓ | ✓ | ✓ |
| Trigger-coverage baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by facility | — | ✓ | ✓ |
| Exception workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Instrumentation integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-jurisdiction cycle rules | — | — | ✓ |
| Multi-stage governance approvals | — | — | ✓ |
| High facility count | — | — | ✓ |
| Multi-facility review controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, facility 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.
You bring
We build with it
01Your facility list and consequence classes
→Trigger, review-cycle and deviation rule mapping
Week 102Representative prior annual packs
→Reading ingestion, reconciliation and the assembly baseline
Week 203Your design basis reports and OMS manuals
→Governance-record mapping and automation-boundary definition
Week 104Access to relevant APIs, feeds or exports
→Instrumentation and document assessment, then integration setup
Week 205Surveillance packs you would not want read back
→Trigger cases and failure-mode testing
Week 406What no dam-safety review may omit
→Confidence scoring, exception routing, guardrails and review controls
Week 307Named engineers of record to review the pack
→Engineer 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 bands sit on the weeks each phase genuinely fills, which is why the fifth of them carries two.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Governance workflow discovery, rule mapping and the automation boundary
W2Source integration and the reconciliation baseline
W3Assembly workflow, confidence logic and review controls
W4Evaluation suite, trigger checks and failure-mode testing
W5Instrumentation integration, a pilot facility and targeted corrections
W6One surveillance cycle run under the engineer of record, then Agent Care handover
- Reading the band
- A bar spans 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 facilities, and Agent Care picks up the monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Mining AI agent
Build the governance record before the next review falls due.
Show us your facilities, your trigger levels and who signs. If your evidence was built to carry a self-assessment, then it has not yet been asked the question a third-party assessor asks at every criterion — which record shows this.
- Typical delivery
- 4–6 weeks
- Deployment
- Your monitoring data
- Named signers
- Assembly only
- After launch
- Agent Care available
Nestack Agents · Tailings storage facility governance
AGT-MN-09 · Agent Care available after launch