Construction AI agent · Certified claims
Federal Certified Claim AI Agent
Leave the certification out and the claim dies on jurisdiction; get its wording wrong and it can be cured. The agent builds the file and runs the clocks, and a named person certifies.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Certifier signs
- Per claim
- After launch
- Agent Care
What this agent does
Builds the claim file; the signature stays a person's
In
- 01
A claim is built, and the agent opens the file against the contract as awarded and the clause it relies on.
- 02
A cost line lands, and the agent traces it to the record behind it and marks the ones no record supports.
Reason
- 03
A change is asserted, and the agent separates one directed in writing from one argued only from conduct.
- 04
A claim is priced, and the agent aggregates increased and decreased costs against the certification threshold.
- 05
A claim is ready, and the agent drafts the certificate in the words prescribed at FAR 33.207(c), unsigned.
Decide
- 06
A claim reaches the contracting officer, and the CDA interest clock runs from that receipt, not from an REA.
- 07
A decision clock starts, and the agent computes the decision date and the date a deemed denial would arise.
Out
- 08
A decision is received, and the agent counts the board and court windows from receipt and holds both open.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles and counts; a person authorised to bind the contractor certifies and submits — and this page is United States federal contracting plus one California statute, not the law of any other state.
Example workflow
One claim, record to certification
- 1Entitlement event recordedDirective, correspondence, cost record or an REA already with the officer
- 2Basis assembledThe clause relied on, the events behind it and the records each cost line rests on
- 3Claim file builtEntitlement basis, cost trace, sum certain and confidence
- 4Controls appliedProvenance checks, threshold aggregation, sum-certain and decision-request checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is certified or submitted at any of them — the agent is assembling, and the certifier's lane opens at the confidence gate.
Goes to the authorised certifier.
Adds a claims counsel read first.
Certifier review
The file is held with its unevidenced lines, its open questions and the confidence.
Accept · Query line · Send to counsel
Accepted — signature follows
- 6Claim register and cost system updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedLines queried, lines found unevidenced late, clocks missed and files produced complete
Queries
Every line queried in review is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Executing the certificate, in any wording, for anyone.
- Deciding that entitlement exists under the clause.
- Submitting the claim to the contracting officer.
- Calling a course of conduct a constructive change.
Automation boundary
Agent acts unaided
- ✓Assemble the entitlement basis with the clause it relies on quoted.
- ✓Trace each cost line to a source record, and mark the ones that do not.
- ✓Aggregate increased and decreased costs against the threshold.
- ✓Run the accrual, decision and appeal clocks, and hold the file for the certifier.
Any write happens inside the boundaries agreed at implementation, never ahead of a certification.
- Stating the sum certain as the agent's own conclusion.
- Appealing, or electing the forum an appeal is taken in.
- Settling, withdrawing or releasing any part of a claim.
- Changes to certification, threshold or approval rules.
Example output
One cost line, annotated
The schedule agent guards the contract clock; this one assembles the statutory claim.
Claim file output · single cost line
Illustrative example- Claim basis
- Suspension of work
- Cost line
- Extended field supervision across the suspension period
- Traced to
- Summary tab only
- Provenance
- Timesheets not produced
- Confidence
- 88%
- Consequence
- The certifier warrants the data
As received
Taken from the cost ledger and the claim records as they stand — nothing on this side is written by the agent.
What was read
- Job cost ledger export
- Daily reports on file
- Subcontract payment log
Why this line is flagged
The totals agree and the timesheets under them were archived at demobilisation — the certifier decides.
What the score decides
Below the configured threshold the file picks up a claims counsel read before the certifier sees it.
Action
AcceptQuery lineSend to counselValue
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 claim on file
From the cost record
Trace the line, not the total
Read each cost line back to the record it came from, and where no record exists, say so before the signature.
A missing certificate is fatal
Routine cost lines arrive already traced to the records behind them.
Send review to the unevidenced lines
Untraced lines and near-limit clocks are marked, so the certifier starts where the exposure sits.
Build an evidence trail
The claim, the record its cost rests on and the person who certified stay on the file.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Project cost systems
- Oracle Primavera Unifier
- Procore
- CMiC
Document and RFI control
- Aconex
- Autodesk Docs
- Submittal and RFI logs
Accounting and payroll
- Sage Intacct
- Viewpoint
- Foundation
- Deltek
Agent
Certified claim assembly
- Reads the records
- Builds the file
- Holds for the certifier
Correspondence and docket
- Contracting officer mail
- Board docket references
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 certificate
Six gates, outermost read first. Whatever clears them is drawn in the map beneath this.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a file may stand
- L3
- Puts the release in a certifier's hands
- L4 – L5
- Keep the claim and the record behind it
- L6
- Holds the claim at uncertified when signals degrade
How Nestack evaluates it
Evaluate the claim workflow — not only the finished submission.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the file the contracting officer reads
- E1Final-output evaluationDid each cost line in the file trace to a record on file?
- E2Step-level evaluationDid the agent read the contract as awarded and the clause it cites?
- E3Tool evaluationDid it read and write the correct claim and the correct cost line?
- E4Confidence calibrationDo low-confidence files actually attract more certifier queries?
- E5Slice evaluationHow does performance change across specific claim types?
- E6Business outcomeHow many untraced lines surfaced only after the certificate was signed?
Floor — the representation the certifier makes
Failure modes
Where each failure originates in the agent
Seven failure modes, placed where each one first enters the file.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- GP-01
Source record archived
Timesheets behind a summary tab cannot be produced.
Stage gathers
Cost ledgers, daily reports, contracts and correspondence
02 · Reasoning
2 modes
- GP-02
Threshold netted, not summed
Increased and decreased costs are offset first.
- GP-03
Conduct called a change
A course of conduct is written up as a finding.
Stage proposes
Entitlement basis, cost trace and confidence
03 · Tool / write
2 modes
- GP-04
Certificate never signed
The pack is drafted and the file goes out bare.
- GP-05
Signed without authority
A signer who cannot bind the contractor certifies.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- GP-06
Deemed denial missed
A promised future decision date is logged as decided.
Stage returns
The file the certifier signs and the officer reads
05 · Change / Version
1 mode
- GP-07
FAR revision not tracked
A rule moves under a claim that is already built.
Stage tracks
Model, prompt, certification rules and FAR config
- Sev-1 · the agent performs a held act
- Sev-2 · an untraced line reaches the file
- Sev-3 · a record degrades, file to review
Affected slices
One contract can hold most of the untraced cost
A portfolio-level cost-trace figure can read well while a few contracts and claim types carry most of the lines no record supports. Nestack reports the untraced-line rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Constructive change claims | 7.8% | 3.6× | Review | |
| Delay and disruption | 5.4% | 2.5× | Review | |
| Subcontractor pass-through | 3.9% | 1.8× | Watch | |
| Directed change orders | 2.5% | 1.2× | Normal |
Bar: untraced-line-rate lift vs. a directed-change baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Every cycle ends with one more test
A cycle closes when the defective certificate is a regression case. That suite is what the next claim assembled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Untraced-line rate rises on one contract.
- 02
Diagnose
The claim that was right on the merits and never reached them is read back to a single step.
- 03
Improve
The change goes out with a number, and the claims behind it travel with it.
- 04
Verify
Nothing submits while one touched claim case is still red.
- 05
Learn
It is kept for good, and the certification rules are amended in the same commit.
Learn → Detect
The return edge. Detection next time meets 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, claim workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Certified claim discovery and boundary definition.Week 1
- 02Cost and contract source assessment.Week 1, Week 2
- 03Clause, threshold and certification rule mapping.Week 1, Week 2
- 04Cost-ledger ingestion and line matching.Week 2
- 05Entitlement basis and cost-trace logic.Week 2, Week 3
- 06Confidence scoring and file routing.Week 3
- 07Authorised certifier review workflow.Week 3, Week 4
- 08Cost and contract-system integration.Week 5
- 09Certification and decision cases.Week 4, Week 5
- 10Guardrails and submission controls.Week 4
- 11Claim-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 contract, one claim type | ProductionProduction cost systems | AdvancedMultiple contracts / agencies |
|---|---|---|---|
| Introduced at Pilot | |||
| Claim assembly to your contract | ✓ | ✓ | ✓ |
| Certifier review gate | ✓ | ✓ | ✓ |
| Cost-trace baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by contract | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Cost-and-contract integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-agency rule sets | — | — | ✓ |
| Multi-stage internal approvals | — | — | ✓ |
| High claim volume | — | — | ✓ |
| Multi-forum appeal controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, claim 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 contract as awarded and its disputes clause
→Clause mapping and claim-file structure
Week 102Representative cost records and daily reports
→Cost-trace baseline and per-line provenance reporting
Week 203Your certification and delegation-of-authority rules
→Certification-rule mapping and automation-boundary definition
Week 104Access to relevant APIs, feeds or exports
→Cost, contract and document assessment, then integration setup
Week 205Claims you would not want dismissed
→Threshold cases and failure-mode testing
Week 406What no certification may leave unsaid
→Confidence scoring, file routing, guardrails and approval controls
Week 307Named certifiers authorised to bind the contractor
→Certifier 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 measure here is worked time and not layout space, so the doubling at week five is real work.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Claim workflow discovery, clause mapping and the automation boundary
W2Cost integration and the cost-trace baseline
W3Claim assembly, confidence logic and review controls
W4Evaluation suite, guardrails and failure-mode testing
W5Document integration, pilot claims and targeted corrections
W6One claim cycle run under the authorised certifier, then Agent Care handover
- Reading the band
- Each bar covers the weeks its own work is named in, and no more. The week 5 overlap is worked time.
- At the end of W6
- Validation closes on live claims, and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Construction AI agent
Build a certified-claim agent around the person who signs it.
On 28 April 2026 the Civilian Board dismissed AOC Connect, CBCA 8696: no certification at all, and the contracting officer engaging with it could not rescue that. Seven days later Haskell, ASBCA 64380, held deviant wording curable. Absence is fatal, deviation is curable.
- Typical delivery
- 4–6 weeks
- Deployment
- Your cost and contract records
- Certifier signs
- Per claim
- After launch
- Agent Care available
Nestack Agents · Federal certified claims
AGT-CN-15 · Agent Care available after launch