Benchmarks & surveillance

Insights

Every value is computed twice by independently written engines, Ed25519-signed, and published onto an append-only hash chain. Settlement uses the value as first published.

CIOI · Core Industrial Output Index

183.5 +0.0%

Chain verified ✓Two engines agreed

120 published fixings · 30–200 corridor · rounding applied exactly once, at publication.

✦Model · proposes only

Logistic typology scorer

Encodes known typologies — round-tripping, related-party dealing, concentration, structuring. Weights are declared, not fitted: a venue that has not launched has no labelled abuse cases.

✦Model · proposes only

Isolation forest

Genuinely unsupervised (Liu, Ting & Zhou 2008). Scores how few random splits isolate a member — the half with any chance of surfacing a typology nobody wrote a weight for.

AI proposes, deterministic code disposes. A block always needs a rule at severity ≥ 0.5 · a conclusive rule blocks alone · a degraded solver run can flag but never block.

Detected subgraph34 members

✦Quantum-formulated surveillance

Anomalous Flow Subgraph Detection

Members form a weighted flow graph. AFSD finds the size-k subset whose internal flow most exceeds what the members' own volumes predict — a QUBO consumed identically by four solver backends, so annealing hardware is a substitution, not a rewrite.

Covert ring blocked — vantage · kestrel · solstice. Three separate legal identities entity resolution cannot link. All three gates green; blocked on flow structure alone.

Backend	n = 15	n = 20	n = 100
Statevector QAOA	262 ms	15.6 s · missed	—
Simulated quantum annealing	9 ms	7 ms · optimum	116 ms
Tabu (classical control)	1 ms	1 ms	144 ms

Statevector QAOA is capped at 15 qubits by the 1500 ms scrutiny budget; simulated quantum annealing runs above it.