Grade each requested document by sensitivity before it moves, name the audience cleared to read it, hold anything unclassified, and leave the release itself to a named person.
A document is requested, and clearance depends on who may read it, not on who asked.
02
A room opens at signing, and both parties stay independent competitors until closing.
Reason
03
A current price list is asked for, and current pricing is what practice holds back.
04
A customer-specific term is sought, and the aggregated form is the usual accommodation.
05
A forward plan reaches the room, and forward plans sit with the too-sensitive material.
Decide
06
A cost structure crosses early, and the Clayton Act waiting period is what it can breach.
07
A pricing decision is coordinated pre-closing, and Sherman Act § 1 governs that conduct.
Out
08
A deal is notifiable in Europe, and Article 7 of Regulation 139/2004 bars implementation.
09
Execute write actions only inside the approval boundaries agreed during implementation.
→Product statement
Classification and the log belong to the agent. Counsel sets the clearance rules and a named person releases each document; nothing here is signed or filed at a regulator's requirement.
Example workflow
One document, request to release
AgentHuman
1Document request receivedWorkstream questions, index requests, buyer follow-ups or seller uploads
2Clearance context assembledThe document, the workstream that wants it, the sensitivity class it falls in and the audience cleared for it
3Clearance decision draftedThe document, its class, the recipient tier and the reason
4Controls appliedSensitivity checks, recipient-role checks, redaction checks and clearance confidence
No human action required
Stages 1 to 4 run unaided, and nothing crosses at any of them — the agent only proposes, and clean-team counsel enters at the clearance gate.
5DecisionSplits at the clearance gate
Clearance clear-cut
Goes to the named releaser to send.
Anything borderline
Adds a clean-team counsel read first.
Clean-team review
The document waits with its class, its recipient tier and the reason drafted for it.
Release · Redact · Send to clean-team review
Released — by a named person▼
6Data-room and log records updatedOnly where write access and records policy allow it
7Outcome evaluatedClearance accuracy, sensitivity coverage, counsel corrections and what review found
Corrections
Each reviewer correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Clearing a document for release to the other side.
Signing the clean-team protocol for a party.
Deciding a document is not competitively sensitive.
Setting who may sit on the clean team.
Automation boundaryAgent acts unaided
✓Map each requested document to a sensitivity class.
✓Hold anything still unclassified until counsel has set its rule.
✓Record who released which document, to which team, and on what day.
✓Flag a request that reads as integration execution, not planning.
No document crosses to the other side except by a named person, and the log keeps it.
Judging whether a document may lawfully cross.
Telling a regulator the room was run cleanly.
Confirming that a clean team was properly staffed.
Changes to clearance rules, tiers or the release log.
Example output
One clearance decision, annotated
The deal-underwriting due-diligence copilot in real estate reconciles asset-level financials for a discretionary investment committee; this decides what may cross between two companies that still compete.
Clearance decision · single documentIllustrative example
Document
Recorded as
Sensitivity class
Evidence of record
Confidence
Held for
Customer pricing schedule, current
Cleared to the clean team, not the deal team
Current customer terms
Seller upload, 3 August 2026
Held unreleased
The clean-team releaser, by name
As receivedSourced from the seller upload and the request behind it, and from no wider place than those two.
What the record holdsSeller uploadWorkstream requestClean-team roster
Why nothing crosses at this pointWhether a document may cross is a judgement counsel makes, not a model.
ActionReleaseRedactSend to clean-team review
What the score decidesBelow the configured threshold a document picks up a clean-team read before release.
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 documentFrom the team that requested it
03Evidence
Where the evidence is used
Early implementation shows up as pre-closing consent rights, day-to-day involvement in the target and sensitive-information exchange, none of which a log alone can cure.
01Approved path
Two companies, one room
The parties still compete until closing, and Altice, C-746/21 P, decided by the CJEU on 9 November 2023, made Articles 4(1) and 7(1) autonomous and separately fineable.
02Human review
What was checked, and not found
No US gun-jumping enforcement was found after the January 2025 settlement; the LivCor and Willow Bridge decree terms were not verified, so no terms are stated; and the Vivendi statement of objections over Lagardère is preliminary and pending.
04Build an evidence trail
The document, the team cleared to read it and the person who released it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Data room and indexVirtual data room · request log Uploads and index items
Commercial and financePricing systems · contract store Price lists and terms
Deal and clean teamsTeam roster · role register Membership and role records
Agent
Data-room clearance decisions
Reads each request Classifies the document Holds for the releaser
Counsel and advisersOutside counsel · advisers Clearance rules and escalations
Integration availability depends on the client's existing systems and API access.
Agent controls
Six baffles between the model and the release
Six baffles set across one duct, the last the narrowest. What comes through is drawn in the map below.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeNarrow the agent to classification only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and clearance rules; XCL, Verdun and EP Energy (D.D.C., 7 January 2025) covered coordinating on prices for the other party's customers.Track
L4TraceabilityRecord each document, the class it was given, the audience it reached and the reason behind it.Record
L3Named releaseHold each document for the person counsel names; that hold decides release, never whether the class itself was right.Gate
L2Clean-team guardrailsCheck membership against the roster; the only agency writing is an FTC staff blog of 20 March 2018, framing clean teams as safeguards parties should employ.Restrict
L1Confidence thresholdsRoute a borderline document to clean-team counsel first; Legends (S.D.N.Y., 5 August 2024) turned on operational control taken before closing.Require review
Model coreEvidence assembled — the document, its class, the audience and the reason
L1 – L2Test whether a document may cross
L3Leaves the release to a named person
L4 – L5Keep the document and the clearance behind it
L6Holds the document unreleased when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the clearance decision that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the log an enforcer would read
Depth of coverage ▼
E1Final-output evaluationDid the class match what the document actually contains?
E2Step-level evaluationDid the agent read the right roster, the right request and the live clearance rules?
E3Tool evaluationDid it read and write the correct document and the correct release entry?
E4Confidence calibrationDo low-confidence clearances actually attract more counsel corrections?
E5Slice evaluationHow does performance change across specific workstreams?
E6Business outcomeHow many documents needed a correction before a person released them?
Floor — the record the parties answer for
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed at the stage where it first surfaces.
Agent lifecycleDirection of processing →
01 · Retrieval1 mode
LZ-03
Stale roster read
The clean-team roster is not the one now in force.
Stage gathersThe requests, the documents, the tiers and the dates
02 · Reasoning2 modes
LZ-04
Class asserted, not reasoned
A document is classed with no reason recorded.
LZ-06
Clean team read as law
A private protocol is worked as a legal rule.
Stage proposesThe classes, their reasons and the cleared audience
03 · Tool / write2 modes
LZ-02
Thin clearance passed forward
A document moves on without the counsel read.
LZ-05
Wrong audience on the release
The document goes to the team it was withheld from.
Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
LZ-01
Released, reason unrecorded
The log shows a release but not the reason for it.
Stage returnsThe clearance a person makes and the log keeps
05 · Change / Version1 mode
LZ-07
Silent tier drift
A rule change moves the tier while the log keeps the old one.
Stage tracksModel, prompt, clearance rules and log fields
Sev-1 · a document cleared on no reasonSev-2 · wrong audience reads the documentSev-3 · source degrades, document held back
A workstream-level clearance-accuracy figure can read clean while pricing and customer terms carry most of the corrections. Nestack reports the correction rate by workstream, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Pricing and customer terms
9.2%
3.6×
Review
Cost and margin detail
6.6%
2.6×
Review
Forward plans and roadmaps
4.1%
1.6×
Watch
Historic aggregated financials
1.8%
0.7×
Normal
Bar: correction-rate lift vs. historic aggregated baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
What an early crossing costs
A loop ends when the document crossing early becomes a case the next release must pass. That suite is what the next room cleared is measured against.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Correction rate rises on pricing and customer terms.
02Diagnose
The price file that entered the buyer room before clearance landed is read back until one cause remains.
03Improve
Number the change; the documents that drove it are filed underneath it.
04Verify
Each touched document case is run once more, and one red holds it back.
05Learn
It stays on as a standing test, and the clearance rules move alongside it.
Learn → DetectThe return edge. The next document is measured 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, clearance logic, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Clearance-boundary and sensitivity-class definition.
02Data-room, pricing and roster sources.
03Document-to-class and audience-coverage mapping.
04Requested document ingestion.
05Document, class and audience binding.
06Clearance scoring and review routing.
07Named-releaser workflow.
08Data-room platform integration.
09Clearance and sensitivity cases.
10Guardrails and release controls.
11Document-trail instrumentation.
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 deal, one roomProductionProduction clearance workflowAdvancedMultiple rooms / workstreams
Introduced at Pilot
Clearance decisions to your rules✓✓✓
Release by a named person✓✓✓
Requested-document baseline✓✓✓
Introduced at Production
Reporting by workstream—✓✓
Clean-team review workflow in your systems—✓✓
Approved write-back—✓✓
Data-room-platform integration—✓✓
Introduced at Advanced
Multi-party rooms——✓
Cross-workstream clearance packs——✓
Large document estates——✓
Multi-team clearance controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, request 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 live rooms and the workstreams behind them→Workstream mapping and request captureWeek 1
02Representative pricing, cost and plan documents→Class binding, audience logic and the clearance baselineWeek 2
03Your clean-team protocol and the roles it names→Sensitivity mapping, audience binding and the automation boundaryWeek 1
04Access to relevant APIs, feeds or exports→Data-room, pricing and roster source assessment, then integration setupWeek 2
05Documents you would not want disclosed→Sensitivity cases and the evaluation runWeek 4
06What no data room may settle→Clearance scoring, review routing, guardrails and release controlsWeek 3
07A named person who releases each document→Handover to the named releaser, 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
The chart is honest about overlap: week five runs two phases together because both of them need it.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Clearance workflow discovery, sensitivity mapping and the automation boundaryW2Source integration and the requested-document baselineW3Clearance logic, audience binding and release controlsW4Evaluation suite, sensitivity cases and failure-mode testingW5Data-room integration, pilot releases and targeted correctionsW6One deal period run under the clean-team counsel, then Agent Care handover
Reading the bandNo bar runs wider than the work it names, and week five is where two of them overlap.
At the end of W6When the clearance record validates, Agent Care takes the agent on.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Strategy AI agent
Build a data-room agent around the document that should never have crossed.
Show us one document and the reason it crossed. Counsel sets the clearance rules and a named person releases each document, and nothing is signed or filed at a regulator's requirement. A clean-team protocol is one private contract, not a recurring statutory certification.
