Screen each competitive input at the door — where it came from, whether it was ever published, whether a rival is behind it — and leave the admission to a named analyst.
An input is offered, and the agent records the source before the landscape reads it.
02
A source is proposed, and the first question is whether the material was ever published.
Reason
03
An input that is current or forward-looking, rather than historical, is held for counsel.
04
A figure that is competitively sensitive — price, cost, capacity, bid or wage — raises the bar.
05
An item attributable to one named rival, rather than aggregated, is quarantined on that ground.
Decide
06
A source that runs both ways, where each side knows the others contribute, is refused.
07
An input reaching you directly or indirectly from a competitor is refused and escalated.
Out
08
A customer relaying a rival quote is logged against that customer, and the rival is never asked.
09
Execute write actions only inside the approval boundaries agreed during implementation.
→Product statement
The agent records provenance and applies the rules. A named analyst admits an input; competition counsel owns the admissibility rules themselves.
Example workflow
One input, offer to admission
AgentHuman
1Input offeredSales notes, vendor feeds, public filings or an inbound message
2Provenance assembledThe item, the person or system that offered it, the tier it falls in and the day it arrived
3Hallmark answers draftedThe item, the five answers, the doubts and confidence
4Controls appliedTier checks, reciprocity checks, sensitivity checks and admissibility confidence
No human action required
Stages 1 to 4 run unaided, and nothing is admitted at any of them — the agent is only recording, and counsel enters at the admissibility gate.
5DecisionSplits at the admissibility gate
Provenance clean
Goes to the named analyst to admit.
Anything doubtful
Adds a competition counsel read first.
Counsel review
The item is held with its tier, its doubts and the source it arrived from.
Admit · Append provenance · Send to counsel
Admitted — by a named analyst▼
6Landscape and input records updatedOnly where write access and records policy allow it
7Outcome evaluatedProvenance completeness, tier coverage, counsel corrections and what review found
Corrections
A counsel correction is counted in the evaluation run.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Admitting an input into the landscape.
Setting the admissibility rules themselves.
Deciding that an exchange was lawful.
Clearing a competitor-sourced item for use.
Automation boundaryAgent acts unaided
✓Record where every input came from before it is used.
✓Test each offered input against the five hallmarks.
✓Refuse an input traced to a competitor and escalate it to counsel.
✓Hold the tier, the timestamp and the reason on one input record.
Nothing is admitted and no rule changes except by a named person, inside the boundary.
Ruling on a contested source classification.
Telling a regulator the record is complete.
Approving what the company may contribute.
Changes to the rules, the log or the tier map.
Example output
One input, annotated
Our audience-targeting copilot documents the permission behind an audience input; this documents the provenance behind a competitive input, and the restraint here is antitrust rather than privacy.
Admissibility record · single inputIllustrative example
Input
Recorded as
Tier
Provenance of record
Confidence
Held for
Rival price list, by email
Offered by a new hire who came from the rival
Competitor-sourced
Received 3 August 2026
Refused, escalated
Competition counsel, by name
As receivedDrawn from the message it arrived in and the person who passed it on, and nowhere else.
What the record holdsSender and mailboxPrior employer namedTier and timestamp
Why nothing is cleared hereWhether an exchange broke Sherman Act § 1 is a judgement for counsel.
ActionAdmitAppend provenanceSend to counsel
What the score decidesBelow the configured threshold an input picks up a counsel read before it is admitted.
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 input offeredFrom the source that offered it
03Provenance
Where provenance decides
Gibson v. Cendyn (9th Cir., 15 August 2025) affirmed dismissal where no rival data was pooled; Cornish-Adebiyi v. Caesars (3d Cir., 29 July 2026) reversed dismissal where it was. Both are pleadings.
01Approved path
Where the input came from
Each item carries the tier it was given, the day it arrived and the reason it was let in or turned away.
02Human review
What was checked, and not found
DOJ withdrew the 1993, 1996 and 2011 healthcare statements on 3 February 2023 and the FTC withdrew on 14 July 2023. A joint request for information ran from 20 February 2026 to 24 April 2026; no replacement zone was found, and later guidance could not be verified.
04Build an evidence trail
The input, the provenance behind it and the analyst who admitted it 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.
Public and published sourcesWebsites · filings · press Published price lists and tariffs
Purchased researchSyndicated reports · panels Subscription market data
Field and customer signalsCRM · win-loss notes Customer-reported rival quotes
Agent
Competitive landscape provenance
Records the source Applies the rules Refuses and escalates
Inbound and refusalMailboxes · ticketing Escalations to counsel
Integration availability depends on the client's existing systems and API access.
Agent controls
Six sluices between the model and the landscape
Six sluices opened in turn, the last a bare crack. Whatever gets past is named in the map below.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeNarrow the agent to public sources when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and admissibility rules; the April 2000 collaboration guidelines were withdrawn on 11 December 2024 with no replacement.Track
L4TraceabilityRecord each input, the tier it was given, the source behind it and every read of the log.Record
L3Analyst admissionHold an input for the named analyst who admits it; the hold governs admission, not whether an exchange was lawful.Gate
L2Hallmark guardrailsTest each input against the five hallmarks and record the answers; no withdrawn safety zone may be cited here as authority.Restrict
L1Confidence thresholdsRoute a doubtful input to a counsel read; the workers guidelines of 16 January 2025 reach an exchange made “through an algorithm or through a third party’s tool” even where recommendations bind nobody.Require review
Model coreInput received — the source, the tier, the hallmarks and confidence
L1 – L2Test whether an input may enter
L3Leaves the admission to a named person
L4 – L5Keep the input and the provenance behind it
L6Narrows to public sources when signals degrade
How Nestack evaluates it
Evaluate the whole intake — not only the landscape that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the landscape a planner reads
Depth of coverage ▼
E1Final-output evaluationDid the record show where each input actually came from?
E2Step-level evaluationDid the agent read the right source, the right tier and the live rule set?
E3Tool evaluationDid it read and write the correct input record and the correct log?
E4Confidence calibrationDo low-confidence inputs actually attract more counsel corrections?
E5Slice evaluationHow does performance change across specific input classes?
E6Business outcomeHow many inputs needed a correction before the analyst admitted them?
Floor — the conduct the company answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed where the intake first goes wrong.
Agent lifecycleDirection of processing →
01 · Retrieval1 mode
LY-03
Stale tier map read
The tier map in use has fallen behind what counsel approved.
Stage gathersThe sources, the tiers, the offers and the dates
02 · Reasoning2 modes
LY-04
Tier asserted, not traced
An input is tiered with no source behind it.
LY-06
Withdrawn zone read as live
A withdrawn safety zone is worked as current.
Stage proposesThe inputs, their tiers and the hallmark answers
03 · Tool / write2 modes
LY-02
Doubtful input passed forward
An input moves on without the counsel read.
LY-05
Input bound to wrong rival
The item is filed against the wrong competitor.
Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
LY-01
Admitted, provenance unrecorded
The log shows an admission but not the source behind it.
Stage returnsThe landscape a planner and counsel both read
05 · Change / Version1 mode
LY-07
Silent rule drift
A release moves the rules while the log keeps the old ones.
Stage tracksModel, prompt, tier rules and input fields
Sev-1 · an input admitted with no sourceSev-2 · rival material reaches the planSev-3 · source degrades, input held back
A competitor-level provenance figure can read clean while competitor-sourced offers carry most of the rework. Nestack reports the correction rate by input class, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Competitor-sourced offers
9.7%
3.6×
Review
Customer-reported pricing
6.9%
2.6×
Review
Purchased vendor research
4.3%
1.6×
Watch
Published public sources
2.4%
0.9×
Normal
Bar: correction-rate lift vs. public-source baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
What a refused input costs
A cycle closes when the competitor-sourced figure is a regression case. That suite is what the next landscape issued is measured against.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Correction rate rises on competitor-sourced offers.
02Diagnose
The competitor price list that arrived by email from someone who used to work there is traced back until a single cause is left.
03Improve
The change ships numbered, and the inputs that forced it ride with it.
04Verify
Nothing releases while one touched input case is still red.
05Learn
It is retained for good, and the admissibility rules are amended in that same commit.
Learn → DetectThe return edge. The next landscape 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, admissibility rules, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Source-tier discovery and automation-boundary work.
02Public, purchased and field sources.
03Input-to-tier and hallmark-coverage rule mapping.
04Competitive input ingestion.
05Source, tier and timestamp binding.
06Admissibility scoring and review routing.
07Analyst admission workflow.
08Research-vendor and mailbox integration.
09Provenance and admissibility cases.
10Guardrails and admission controls.
11Input-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 market, one yearProductionProduction landscape workflowAdvancedMultiple markets / entities
Introduced at Pilot
Admissibility to your rules✓✓✓
Named-analyst admission✓✓✓
Admissible-source baseline✓✓✓
Introduced at Production
Reporting by competitor—✓✓
Escalation workflow in your systems—✓✓
Approved write-back—✓✓
Research-vendor integration—✓✓
Introduced at Advanced
Multi-market landscapes——✓
Cross-source provenance packs——✓
Large input estates——✓
Multi-source provenance controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, input 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 markets and the rivals you track in each→Source-tier mapping and provenance captureWeek 1
02Representative public, purchased and field inputs→Source binding, tier logic and the admissibility baselineWeek 2
03Your standing antitrust policy and the counsel it names→Rule mapping, tier binding and the automation boundaryWeek 1
04Access to relevant APIs, feeds or exports→Public, purchased and field source assessment, then integration setupWeek 2
05Inputs you would not want produced→Admissibility cases and failure testingWeek 4
06What no landscape view may prove→Admissibility scoring, review routing, guardrails and release controlsWeek 3
07A named analyst and the counsel who owns the rules→Analyst admission 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
Duration here reflects the work rather than the drawing, so the fifth column shows a pair running on.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Landscape workflow discovery, tier mapping and the automation boundaryW2Source integration and the admissible-input baselineW3Admissibility logic, hallmark tests and release controlsW4Evaluation suite, provenance cases and failure-mode testingW5Vendor integration, pilot inputs and targeted correctionsW6One planning year run under the competition counsel, then Agent Care handover
Reading the bandA bar runs across only the weeks its own work is named for, and the fifth takes two because the work does.
At the end of W6Validation closes on live inputs, and Agent Care picks up the watch.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Strategy AI agent
Build a landscape agent around the input nobody should have opened.
Show us one input and where it came from. Not a certificate. A dated record of why each input was let in or turned away. The RealPage consent judgment of 24 November 2025 is proposed, not entered; Cornish-Adebiyi v. Caesars, No. 24-3006, went no further than the pleadings.
