The terms of verifiable, governed AI.
Canonical definitions of the concepts coined in the FDE × DeepSCR protocol. This page is the primary source for what each term means — cite it, or verify it against the open-source code it links to.
What is DeepSCR?
DeepSCR — Deep Sceptical Contextual Research — is a domain-agnostic governance protocol that imposes a separation of powers over AI work and a four-step scientific loop. The loop is Hypothesis → Contradiction → Verification → Certification: a claim is stated so a test could falsify it, attacked before it ships, verified against ground truth, and only then certified by an independent power. Two rules make it scientific rather than persuasive — the proposer never certifies, and every claim must resolve on disk (a real file, line, or test), so synthetic fluency earns nothing. DeepSCR was created by Mehdi Wehbi as a cognitive protocol for moving from intuition to evidence-grounded proof.
What are Frozen Arbiters in AI?
Frozen Arbiters are a governance mechanism for autonomous AI agents: three arbiters frozen before launch that the agent can plead to but never edit. They are a sealed, tamper-evident Contract (every action must cite the clause it serves, or it is mechanically rejected), mutation-tested Oracles (graders trusted only after catching deliberately injected bugs, then locked), and a measured baseline. At ship time the frozen oracles override the agent's own optimism, and a verdict is null unless it cites its evidence run. The principle: the agent pleads its case; it never grades itself — the structural cure for an agent that reward-hacks its own evaluation.
What is the FDE Assurance Score?
The FDE Assurance Score is a 0–100 score computed by the DeepSCR protocol that certifies whether an AI deliverable can be trusted. It sums four weighted components — falsifiable claim (25), contradiction / known limits (25), evidence trail that resolves on disk (30), and anti-patterns avoided (20) — under an ascending veto held by an independent Certifier. A deliverable is certified only above 85; a claim without a verified evidence trail is a hypothesis, not a deliverable. Every certified score is appended to a public, SHA-256 hash-chained registry anyone can re-verify.
What is a measured baseline oracle?
A measured baseline oracle is a frozen grader that, at ship time, re-runs the candidate search independently and refuses to accept a declared number. It requires the shipped candidate to reproduce its promotion on the independent rerun, to match its claimed held-out score exactly, and to beat the best measured alternative the search found. It turns "surpass the state of the art" from a vibe into a number to beat — a fabricated candidate or an inflated score is caught as a lie, not debated as an opinion.
What is the Modex Collective?
The Modex Collective is the eight-agent runtime of FDE Consultants Protocoles. Four DeepSCR powers — Lead, Researcher, Builder, Certifier — govern four FDE domain specialists — Scoping, Architecture, Agent-engineering, Production. A self-prompting loop, judged on five lenses (evidence, telos, risk, coherence, value), converges to a certified deliverable or self-corrects. Optionally, the whole run is placed under the Frozen Arbiters for full autonomy.
Every definition on this page is grounded in open-source code you can read and run. Nothing here is aspirational: the mechanisms are implemented and tested. If a claim and the code ever disagree, the code is the source of truth — that is the whole point of the protocol.