A federation of six verification instruments that turns what an AI system says into executed, sealed evidence — so the claim that reaches your customer, your regulator, or your board is one a machine already tried to break.
Anchored claims verify with zero model calls. Unverifiable never becomes true.
The decision ledger. Every non-obvious guard, magic number, or special case gets its reason recorded as a git-native trailer and note — quoted from history, or the lookup reports "not recorded." It never invents a plausible-sounding why.
Legal and compliance rules as cited, versioned, dated data — never model memory. Every rule carries its source, its effective date, and its verification trail behind it. The model is never the source of a rule.
A deterministic compliance engine: same inputs, same rule version, same decision — reproducible to the hash. AI output is typed a Suggestion until a named human converts it, and real-world effects stay dry-run until a logged token authorizes them.
The pipeline is CLAIM → PLAN → EXECUTE → JUDGE → GATE → DOSSIER. Five check kinds — exec, read, grep, fetch, json — cover code, text, and now structured data; json is the newest, anchored deep-equality against executed output. Verbatim anchors run at least eight characters; a pure, non-LLM gate renders the verdict.
Sealed dossiers travel between machines as portable, signed artifacts. A counterparty re-checks the seal, not the story — verification that doesn't require trusting the sender.
The outcome ledger. Append-only and hash-chained, it records what a system claimed against what actually happened, so calibration is measured, not asserted.
The assertion is stated in the open claim-batch format: what's claimed, and what would prove it.
The model proposes a check — exec, read, grep, fetch, or json — and nothing else. It plans; it does not run anything.
The check runs for real, in your environment, under your keys. The model never executes.
Output is compared against the claim: verbatim anchors must match character-for-character; json checks run anchored deep-equality.
A pure, deterministic, non-LLM function renders the verdict: SUPPORTED, REFUTED, or UNVERIFIABLE. No temperature, no prompt, no drift.
The verdict, the evidence, and the run are sealed with a sha256 hash. Change one byte and the seal shows it.
json semantics: anchored deep-equality against executed output. A mismatch fails closed to UNVERIFIABLE — the gate never guesses. REFUTED is reserved for executed evidence that contradicts the claim.
claim-batch@0 is a format, not a platform. Anything that writes JSON can emit it; anything that reads JSON can consume it. Specification available on request →
Regulators stopped asking whether you have a policy. They ask for evidence it ran.
The model plans and judges. It never executes. Execution happens in your environment, under your keys.
Bring your own key. Your credentials never transit our infrastructure; a viral day costs you nothing and leaks nothing.
The verdict gate is pure, deterministic, non-LLM code. No temperature, no prompt, no drift.
Anything unproven is UNVERIFIABLE. The system has no way to say "probably."
Dossiers are sha256-sealed. Change one byte and the seal shows it.
Instruments are vendored, never imported across trust boundaries — no transitive supply-chain surface between layers.
We say verification and evidence — never audit, attestation, or opinion. Those are reserved professional terms (AICPA AT-C 205), and using them casually would be its own false claim.
A six-stage market diligence with an adversarial QA gate — findings published to the ledger, kill criteria included.
Every application carries an honest maturity label. On a diligence site, "in validation" means in validation.
679 codified rules served in production, each carrying its citation and verification status.
Seven verticals on one deterministic spine, running full pipelines on draft rule packs.
A public, falsifiable ledger of our own claims — the stack pointed at itself.
Every instrument is inspectable — deterministic gates, published verifiers, dossiers you can re-check offline.
The dossier covers every property: what's live, what's dry-run, what's still in validation — with the receipts to check us.
sealed · sha256 · fail-closed · model:null