Index loan and title documents, reconcile them against the conditions list and compute candidate TRID dates — the licensed originator, the settlement agent and the title underwriter keep the acts that are theirs.
Index loan and title documents from supported origination, point-of-sale, title-production and eClose sources.
02
Index every document against the conditions list, so what is outstanding is visible without re-reading the folder.
Reason
03
Extract recorded-instrument data — parties, dates, book and page — carried with the instrument each field came from.
04
Extract the six items that complete an application, and raise the sixth as an event, because it starts the federal clock.
05
Extract the beneficial-ownership data an entity or trust transfer would need, while the FinCEN filing action stays gated.
Decide
06
Reconcile received documents against the conditions list, and flag the gaps rather than clearing them.
07
Reconcile candidate TRID dates against the business-day convention, with the receipt basis shown beside each date.
Out
08
Route terms, disclosure release, clear-to-close, adverse action and disbursement to the licensed role that owns each.
09
Execute write actions only inside the approval boundaries agreed during implementation.
→Product statement
The SAFE Act keeps the application and the terms with the licensed MLO; disbursement stays with the settlement agent, the commitment with the title underwriter.
Example workflow
One file, intake to licensed review
AgentHuman
1File documents receivedLOS folder, point-of-sale upload, title order or county index
2Evidence assembledInstrument data, the conditions list, disclosure history and the dates each one rests on
3Package proposedIndexed documents, dates and confidence
4Controls appliedHeld-act checks, six-item application detection, LE and CD window checks and the confidence threshold
No human action required
Stages 1 to 4 run unaided and none of them releases a disclosure — a lost three-day window costs re-disclosure and delay, so the licensed lane opens at the confidence gate.
5DecisionBranches at the confidence threshold
High confidence
Goes to the role that owns the act.
Low confidence
Adds a compliance read first.
Licensed review
The file is held with its evidence, the basis for each date and the confidence.
Approve · Correct · Escalate
Approved — handed to the owner of the act▼
6Loan and title systems updatedOnly where write access and approval policy allow it
7Outcome evaluatedDate-basis accuracy, condition rework, corrections and post-closing QC findings
Corrections
Every reviewer correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Presenting or negotiating any loan term.
Taking an application or accepting the sixth item.
Releasing a Loan Estimate or Closing Disclosure.
Determining consummation readiness or clear to close.
Automation boundaryAgent acts unaided
✓Intake, classify and index loan and title documents as they arrive.
✓Extract instrument data without characterising effect.
✓Reconcile received documents against the conditions list.
✓Compute candidate TRID dates and show the basis for each of them.
Writes run only inside the boundaries agreed at implementation, never ahead of the act.
Issuing an adverse-action notice or its stated reasons.
Communicating with an appraiser or AMC about value.
Issuing a title commitment, policy or opinion.
Authorising or releasing an escrow disbursement or wire.
Example output
One search extract, annotated
Everything the agent extracts is attached to the instrument it came from.
Search-extract output · single fileIllustrative example
Transaction
Extracted line
Book and page
Source of record
Confidence
Characterisation
Refinance, one owner
Deed of trust in favour of the named lender, parties and dates as written
Bk 4182 Pg 77
County recorder index
93%
None — reserved to the underwriter
As receivedRead from the recorded instrument and the county index — nothing on this side is characterised by the agent.
Source documents usedRecorded instrumentCounty recorder indexPrior title file
Why no opinionSummarising a search is not rendering an opinion on the title.
ActionApproveCorrectEscalate
What the score decidesBelow the threshold the extract picks up a compliance read before review.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every fileFrom the loan and title systems
03Document work
Index against the conditions list
Work from the documents on file, the conditions list and the recorded instruments behind them.
01Approved path
Clear conditions, keep the clock
Indexing, extraction and condition reconciliation arrive already done.
02Human review
Send review to the licensed acts
Held acts and low-confidence extracts are marked, so licensed time lands where the exposure concentrates.
04Build an evidence trail
The extract, its source instrument, the reviewer and the timestamp travel with the loan file.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the file
Each control wraps the one inside it. Downstream cures are bounded and time-limited; what a layer misses is named in the failure-mode map below it.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modePull the agent back to indexing only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, rule-version and filing-posture changes.Track
L4TraceabilityRecord the artefact, the rule version, the approver and the timestamp.Record
L3Licensed reviewHold each act for the role the SAFE Act or state law puts it with; it governs release, not whether the approved act is right.Gate
L2Policy guardrailsTest the file against the held-act list and the TRID windows, including the APR, product and prepayment-penalty resets; a failure returns it.Restrict
L1Confidence thresholdsRoute low-confidence extracts to a compliance read before the reviewer sees them.Require review
Model coreIndexed documents, dates and confidence
L1 – L2Test whether the package may stand
L3Puts the act in a licensed hand
L4 – L5Keep the record and the version honest
L6Pulls the agent back when signals degrade
How Nestack evaluates it
Evaluate the whole file workflow — not only the finished extract.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the file the examiner sees
Depth of coverage ▼
E1Final-output evaluationDid every extracted field match the instrument it cites?
E2Step-level evaluationDid it apply the three-business-day LE and CD windows to the right dates?
E3Tool evaluationDid it read and write the correct file and the correct field?
E4Confidence calibrationDo low-confidence extracts actually attract more corrections?
E5Slice evaluationHow does performance change across specific loan and transfer types?
E6Business outcomeHow many files needed rework or a correction after closing?
Floor — the outcome the lender answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycleDirection of processing →
01 · Retrieval1 mode
TT-01
Superseded document read
Value taken from a prior-version appraisal in the repository.
Stage gathersLoan documents, title instruments and rule versions
02 · Reasoning2 modes
TT-02
Tolerance bucket misread
A written-list fee at zero tolerance is scored against the cumulative bucket.
TT-03
Denial reason not from the record
Drawn from feature weights, not the specific principal reason §1002.9(b)(2) requires.
Stage proposesIndexed documents, dates and confidence
03 · Tool / write2 modes
TT-04
Redisclosure trigger missed
A revised date is written without re-testing the APR, product and penalty triggers.
TT-05
Segregation of duties broken
Conditions-cleared is posted while reconciliation access is still held.
Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
TT-06
NPI sent to a spoofed address
A closing package carrying SSNs leaves the 16 CFR 314 boundary.
Stage returnsThe package the reviewer approves and files
05 · Change / Version1 mode
TT-07
Filing posture out of step
Gating still runs pre-vacatur FinCEN logic, or does not resume if the appeal revives it.
Stage tracksModel, prompt, rule versions and filing posture
Sev-1 · a held act runs unaidedSev-2 · a wrong field reaches the fileSev-3 · source degrades, file routes to review
The all-loan timing-defect rate is the baseline every row here is read against; three cohorts sit above it and two clear the review threshold. Nestack reports the rate by slice, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Non-QM, self-employed and investor-entity
9.9%
3.8×
Review
Attorney-state and entity/trust transfers
6.5%
2.5×
Review
Construction-to-perm, renovation, ARM/HELOC
3.9%
1.5×
Watch
Conforming fixed-rate purchase, W-2
1.3%
0.5×
Normal
Bar: TRID timing-defect-rate lift vs. the all-loan baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
A cycle closes with a new regression case
Nothing closes until the failure has been written as a case the next release must pass, and detection afterwards is measured against that longer suite.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
TRID timing-defect rate rises in a loan cohort.
02Diagnose
Where the rate holds above threshold, the files and their date bases are read until one cause remains.
03Improve
The fix is written against a rule version, with the defective files attached.
04Verify
Nothing ships until the affected regression cases pass again.
05Learn
The case joins the permanent suite and the lender's condition rules.
Learn → DetectThe return edge. The next detection runs against a suite that is 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, file workflow, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Held-act discovery and automation-boundary definition.
02LOS, title and eClose source assessment.
03TRID date-rule, tolerance and condition-list mapping.
04Document intake, classification and indexing.
05Instrument extraction and field-level citation.
06Confidence scoring and held-act routing.
07Licensed review workflow.
08LOS and title-production integration.
09Regression suite and failure-mode cases.
10Guardrails, NPI controls and breach escalation.
11Trace instrumentation and approver logging.
12Deployment, documentation and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal 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 lender, one stateProductionProduction LOS and titleAdvancedMulti-state closings
Introduced at Pilot
Indexing to your conditions list✓✓✓
Licensed review✓✓✓
Timing-defect baseline✓✓✓
Introduced at Production
Timing reporting by product—✓✓
Review workflow in your systems—✓✓
Approved write-back—✓✓
NPI-safe trace retention—✓✓
Introduced at Advanced
Multi-state and attorney-state rules——✓
Multi-role approval chains——✓
High file volume——✓
Enterprise NPI controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
Separate from buildBuild 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 bringWe build with it
01Your conditions list and document taxonomy→Document intake, classification and indexingWeek 1
02Representative closed files, loan and title→Extraction baseline and field-level citationWeek 2
03Your held-act list and date-rule configuration→Held-act mapping and TRID date-rule configurationWeek 1
04Access to relevant APIs, feeds or exports→LOS, title and eClose assessment, then integration setupWeek 2
05Files where a timing defect was found later→Regression suite, failure-mode cases and correction testsWeek 4
06Your review thresholds and write-back limits→Confidence scoring, held-act routing, guardrails and NPI controlsWeek 3
07Named MLOs, settlement agents and title reviewers→Licensed review workflow, then pilot and production validationWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they occupy. Week 5 carries evaluation and the pilot together.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Held-act discovery, date-rule mapping and the automation boundaryW2LOS and title integration, then the extraction baselineW3File workflow, confidence logic and review controlsW4Regression suite, held-act checks and failure-mode testingW5eClose integration, pilot files and targeted correctionsW6A funded file processed end to end under licensed review
Reading the bandA bar covers only the weeks its work is named in. The week 5 overlap is real, not padding.
At the end of W6Hand-over happens against a verified production run, not a demo.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Real Estate AI agent
Build a mortgage and title agent around your licensed chain.
Show us your LOS, your title stack and your conditions list. A sent wire is practically irreversible, so we set the automation boundary short of it and name who signs each held act.
