Skip to content
Proof

Every claim, evidenced.

(001) · [BENCHMARK]

The ON/OFF experiment.

Governed debate across three models (ON) against single models and a simple no-debate combination of outputs (OFF), judged blind on real federal bid decisions. The result: under governance, model debate made better bid/no-bid calls than any single model or a simple average of them.

Pilot caveated
Pairwise win rate · governed vs blind
93% GOVERNED 18% BLIND · loses 82%

Exploratory directional pilot · 7 decision points · no statistical significance claimed.

Harness Lift · Decision Quality
+0.62+0.87

Governed debate lifted decision quality across the pilot’s 7 decision points.

Exploratory directional pilot · 7 decision points · no statistical significance claimed.

(002) · [ENGAGEMENTS]

What the work looked like.

DEFENSE-AEROSPACE PRIME

ISSM and cyber-IA support under an active federal program.

FEDERAL SERVICES FIRM

Bid/no-bid design partner on the rEfracta engine.

ARGUS™ · BUILT ON rDENZ TECHNOLOGY

An open source intelligence solution, developed using rDenz technology including the AELUM harness.

Visit ARGUS (opens in a new tab)
(003) · [METHOD]

How we verify.

Every number on this site carries a verdict: verified, pilot caveated, or roadmap. The claim ledger is the method. It is also the proof.

[TRUTH GATE]
421 claims · swept 2026-07-07
352 matched 15 caveated 54 awaiting row 0 contradictions
VERIFIED · grounded in source documents
PILOT CAVEATED · measured, caveats attached
ROADMAP · planned, labeled as such
DO NOT PUBLISH · blocked by the truth gate

Ask for the evidence.