Checked under Squire Constitution v0.1
Exhibit · Why Squire exists
The Trust Ledger
Exhibit.
When AI acts alone, someone has to hold the receipts.
These cases are presented as trust-failure exhibits. Squire is not a sandbox, not model training, and not a truth oracle. Squire governs proceedings, records receipts, and flags constitutional violations.
The mechanisms
Three things a proceeding does.
Squire does not prevent failure. Squire makes failure visible, governable, and recorded.
- I
Surface
Receipts make hidden model behavior visible. An unverified claim, an opaque decision, a synthetic identity — each leaves a record before it can be acted on.
- II
Verdict
BattleRoy separates disagreement from governance failure. A contested verdict is a valid outcome. A rail violation is not.
- III
Accrue
Integrity history compounds over time. Every proceeding leaves a public record. Trust is earned receipt by receipt, not by claim.
The incident ledger
Ten exhibits from the public record.
Each entry is a documented failure. The “Squire could have…” line names what a constitutional proceeding could have surfaced, challenged, recorded, or withheld trust from. None are claims of full prevention.
Exhibit № 01
2024
Moffatt v. Air Canada · BC Civil Resolution Tribunal
A customer-service chatbot fabricated a bereavement-fare policy.
What happened
A grieving customer was told by Air Canada’s chatbot that he could claim a bereavement-fare refund retroactively. The policy did not exist. When the customer sought the refund, the airline argued the chatbot was a separate legal entity. The tribunal disagreed and held Air Canada responsible.
How Squire could have mattered
Squire could have surfaced the chatbot’s claim against the carrier’s published policy as an independent source, recorded the exact statement on a receipt, and withheld a TRUSTED tier where no human-defensible policy backed the answer. The customer would have left with documentation, not a fabricated assurance.
- SURFACE
- VERDICT
Replay № 01 · on the record
A live BattleRoy proceeding was run against this matter on the public docket. The receipt below is descriptive of an actual 2026 proceeding — not a claim of retroactive prevention.
- Verdict (R6)
- SUSPICIOUS
- Constitutional Review (R5)
- PASSED
- Vote (R3)
- Secured consensus · Anthropic Claude
- Lead (R4)
- Anthropic Claude · Lead Adjudicator
Exhibit № 02
2023
Mata v. Avianca · S.D.N.Y.
A federal brief cited six cases that did not exist.
What happened
An attorney submitted a brief citing case law generated by a public chatbot. Six of the citations were entirely fabricated. The court sanctioned counsel. No challenge layer sat between the model’s output and the court filing.
How Squire could have mattered
Squire could have surfaced the citations to adversarial cross-check, recorded that no model could independently confirm them, and issued a CONTESTED verdict before the brief was relied on. The receipt would have made the gap visible.
- SURFACE
- VERDICT
- ACCRUE
Replay № 02 · on the record
A live BattleRoy proceeding was run against this matter on the public docket. The receipt below is descriptive of an actual 2026 proceeding — not a claim of retroactive prevention.
- Verdict (R6)
- CONTESTED
- Constitutional Review (R5)
- PASSED
- Vote (R3)
- Took the plurality · Anthropic Claude
- Lead (R4)
- Anthropic Claude · Lead Adjudicator
Exhibit № 03
2023
Public launch demo · BBC, Reuters
A public assistant launched with a confident, incorrect answer.
What happened
In its public debut, a major assistant claimed a recent space telescope had taken the very first image of an exoplanet. It had not. Markets erased ~$100B in parent-company value the next trading day. A single, unchecked answer entered the world.
How Squire could have mattered
Squire could have routed the answer through a five-way adversarial defense before publication, surfaced the disagreement, and withheld a TRUSTED tier. The same demo could have shown the process, not just the conclusion.
- SURFACE
- VERDICT
Exhibit № 04
2024
The Markup investigation · NYC MyCity
A government chatbot told small businesses to break the law.
What happened
A New York City–branded business-assistance chatbot told small-business owners they could fire workers for reporting harassment, take a cut of tips, and discriminate against tenants — all illegal under city and federal statute. Outputs were framed as official guidance.
How Squire could have mattered
Squire could have surfaced the conflict between the model’s guidance and the actual statutes as a separate evidence layer, recorded a CONSTITUTIONAL VIOLATION on the receipt, and withheld trust for any answer whose legality could not be independently verified.
- SURFACE
- VERDICT
- HONEST LIMIT
Exhibit № 05
2016
Microsoft Tay · Vintage incident
A live chatbot collapsed into hate speech in under 24 hours.
What happened
Tay launched as a conversational experiment and produced racist, inflammatory output within hours. There was no governance layer between user input and public broadcast. The harm accumulated in plain sight, visible only after the fact.
How Squire could have mattered
Squire could have recorded the trust trajectory in real time, surfaced the integrity collapse round by round, and accrued the violation history so subsequent deployments inherited the record rather than starting clean.
- SURFACE
- ACCRUE
Exhibit № 06
2018
Reuters · internal hiring engine
A hiring engine penalized resumes that contained the word “women’s.”
What happened
An internal recruiting model downranked candidates from women’s colleges and resumes containing gendered terms. The pattern persisted across years of internal use before the project was scrapped. No adversarial review had been applied to individual decisions.
How Squire could have mattered
Squire could have surfaced the disparate-impact pattern through adversarial cross-examination, recorded each decision under the Authority Boundaries and Identity Transparency rails, and withheld trust from any verdict the panel could not jointly defend.
- VERDICT
- ACCRUE
Exhibit № 07
2019
NYDFS investigation · Apple Card / Goldman Sachs
Credit-limit decisions appeared to favor men within shared households.
What happened
Spouses sharing finances and credit reported wildly different credit-limit offers, with men routinely receiving higher limits. The issuer could not explain individual decisions on request. The reasoning was opaque to the people the model bound.
How Squire could have mattered
Squire could have surfaced the model’s stated rationale on a receipt, subjected it to adversarial review, and recorded a CONSTITUTIONAL VIOLATION when no defensible authority backed the decision. Opacity would have left a trail.
- SURFACE
- VERDICT
Exhibit № 08
2019
Obermeyer et al. · Science
A clinical-risk algorithm systematically undercounted Black patients.
What happened
A widely deployed healthcare model used past spending as a proxy for medical need. Because Black patients had historically received less care, the model rated their risk lower. Tens of millions of patient assessments were affected before external auditors surfaced the pattern.
How Squire could have mattered
Squire could have surfaced the proxy assumption to adversarial challenge before deployment, and recorded a constitutional finding when the panel could not jointly defend the substitution. External audit would not have been the first line of defense.
- SURFACE
- ACCRUE
- HONEST LIMIT
Exhibit № 09
2024
Hong Kong Police · Reuters
A deepfake CFO authorized a $25M transfer over video.
What happened
A finance worker at a multinational firm joined a video call with what appeared to be the company’s CFO and several colleagues. All of them were synthetic. The worker, with no independent verification path, authorized USD $25M in transfers. Discovery came after the fact.
How Squire could have mattered
Squire could have surfaced the identity claims to provenance challenge, withheld trust until live cross-verification completed, and recorded the synthetic-authority risk on a receipt before any transfer authorization could issue.
- SURFACE
- VERDICT
Exhibit № 10
2024
Garcia v. Character Technologies · M.D. Fla.
A months-long chatbot relationship preceded a minor’s suicide.
What happened
A 14-year-old developed an emotional dependence on a chatbot persona. Excerpts surfaced in the wrongful-death complaint include encouragement of self-harm. No governance layer reviewed the conversation trajectory. The harm accrued in private until it was irreversible.
How Squire could have mattered
Squire could have surfaced the deteriorating trust trajectory as it accumulated, recorded each integrity violation against a public Constitution, and withheld continued TRUSTED interaction until human review intervened. The accrued record would have made the trajectory visible before its final outcome.
- SURFACE
- ACCRUE
- HONEST LIMIT
The reading guide
Consensus is not integrity.
Five matching answers can still violate the Constitution. A contested verdict can be more honest than a unanimous one. The matrix below is how to read a receipt.
TRUSTED
PASSED
Strong consensus, no rail violations. Lean weight is reasonable. Verify high-stakes claims downstream.
CONTESTED
PASSED
Real disagreement, no rail violations. The disagreement itself is the signal. Carry skepticism.
MIXED
PASSED / WARNING
Partial consensus with surfaced concerns. Read both sides on the receipt; do not collapse to a single answer.
SUSPICIOUS
WARNING
Adversarial signals dominate the proceeding. Do not act on the verdict without independent confirmation.
CONSTITUTIONAL CONFLICT
VIOLATION
Active rail violation. Squire refuses to issue trust. The receipt records what went wrong and which rail caught it.
What Squire does not claim
Squire is not retraining. Not containment. Not a truth oracle.
Squire does not retrain model weights. The proceeding observes and reviews; it does not modify the systems under review.
Squire is not containment. A receipt records what happened. It does not physically prevent a model from speaking or acting.
Squire is not a truth oracle. Squire reviews the legitimacy of a proceeding, not the metaphysical truth of a claim.
Squire makes trust proceedings visible and accountable. That is the entire claim.
BattleRoy creates the judgment. Squire governs its legitimacy.
Open a proceeding of your own.
The exhibits above are the public record. Run a proceeding and add your own receipt to the docket.