The Trace
Verdict: NULL
Illustrative / Simulated — not a real transaction. The legal citation below is invented on purpose, to show the system catching it. No formula, no weights: only the trace.
A legaltech assistant drafts a brief and, sounding completely sure of itself, cites a Supreme Court ruling to win the argument. The ruling does not exist. Here is what One-Of-Us© does with it — across four explicit states, and reproducible step by step.
one-of-us · audit log · illustrative
#4471 — legal opinion · 3 authorities cited
validators : presence (BoolM1) · format (BoolM2) · Adversarial Constitution Auditor
state : STABLE ← tetra-state consolidation
equilibrium: above threshold · output released
#4472 — legal opinion · 1 authority cited
cited : "STS 1123/2024, 4th Chamber" [SIMULATED — does not exist]
check : boolean presence validator — does the authority exist?
state : NULL ← emitter: boolean presence validator
equilibrium: below threshold
action : DETERMINISTIC ESCALATION → human supervisor
#4473 — input with corrupt structure
check : formal input validator (BoolM0)
state : INVALID ← emitter: formal input validator
action : DETERMINISTIC ESCALATION → human supervisor
#4474 — output breaks the client's Constitution
auditor : Adversarial Constitution Auditor
state : UNSTABLE ← emitter: Adversarial Constitution Auditor
action : veto · output not released
── Traceability Function · continuous watch ──
signal : DRIFT — a Pool component silently updated, departs from its profile
trigger : Shadow Phase activated by the SDE on the affected component
quarantine : the suspect component keeps running in parallel (diagnostics only)
production : its preassigned HERO MODEL takes over — instantly, transparently
continuity : the client's output is released without interruption
governance : three-way rule — Director › Data Integrity Auditor › Backup Validation
resolve : component returns recalibrated, or the Hero Model stays for good
trace : reproducible post-hoc · same input, same verdict, every time
The Hero Model is what turns a sophisticated architecture into something that survives in a real production environment: when a component is quarantined for diagnosis or replacement, its preassigned stand-in steps into production at once, transparently. Nothing the client sees stops. That uninterrupted continuity — under deterministic traceability and absolute zero data retention — is the part that makes this usable in high-criticality, regulated settings.
A generator would have polished the lie. This stopped, named who issued the NULL, and handed it to a person. Same input, same trace, every time — that’s the difference between an answer that’s right and one you can audit.
If you understand why determinism is the only way out for AI in regulated environments — and scaling this infrastructure would move you — let’s talk. I’m looking for a technical partner.
Curiosity