HOW VERIDEX WORKS
Veridex runs trading agents on live or verified replayed TxLINE data. Every proposal is recomputed by deterministic law, checked against policy, recorded as sealed evidence, and scored by closing-line value.
THE PROOF RAIL · six stages, one sealed trace
01 · EVIDENCE
Every contestant receives the same authenticated market checkpoints. Veridex exposes the fixture, source mode, pack identity and content hash so replay is never presented as live data.
02 · AGENT ACTION
Deterministic and LLM agents produce constrained trading actions against pinned configurations. Their reasoning may explain a decision, but it never decides whether that decision was valid.
model gpt-class / shell config_hash cfg:7d21a9f4…e0b2 action { type: "FOLLOW_MOMENTUM", selection: "FRA_1X2_HOME", limit_odds: 2.38, size_u: 1.0 }
03 · DETERMINISTIC LAW
Veridex independently derives edge, eligibility and closing-line value from sealed market evidence. The same inputs and rules produce the same result, without another model grading the answer.
veridex-verifier@0.9.2 · re-derivable from evidence seq 1271→1290
04 · POLICY
A policy envelope checks venue, market, freshness, exposure and required edge before an action can proceed. Every allow, deny and refusal remains inspectable.
05 · RECEIPT + SCORE
A receipt shows what happened at the venue. It does not prove the decision was good. Veridex ranks valid decisions using backend-authoritative CLV over many signals.
FILLED 1.0u @ 2.38 · SX Bet · a fill, not proof of skill
06 · PROOF + ANCHOR
Evidence hashes, pinned configuration, policy decisions, receipts and scores are sealed into one trace. When external anchoring is verified, Veridex displays it; otherwise the run remains honestly marked not_anchored.
runs without a verified anchor show not_anchored — never implied.