Skip to main content
These limitations are stated precisely because honesty is a core design principle of Veridex. The platform is built to prove both wins and non-wins, and a clear record of what has not yet been demonstrated is as important as what has.
A trustworthy “no executable edge yet” is part of what Veridex ships.

​
Real-Money Trading

No real-money order has been placed yet. The live_guarded execution surface is built, reviewed, and fail-closed. The path to real money requires every gate in the live-money conjunction to be open simultaneously. The first 1-share FAK smoke is deliberately a human operator’s decision — not the platform’s. No agent, no HTTP call, and no API parameter can arm real-money execution. Specifically:
  • The API deliberately passes no live dependencies. POST /competitions/{id}/start calls start_competition(...) with no live_deps argument. Real execution is operator-direct-only, by construction.
  • live_ready requires two operator-verified confirmations (the on-chain neg-risk approval and a 1-share FAK smoke) to be explicitly set. A preflight that merely “didn’t object” is not consent.
  • Two independent locks must open: the route arm gate and the adapter’s own lock. The money gate does not trust the routing layer.

​
Trading Edge

No executable venue edge has been demonstrated. Veridex ran two end-to-end experiments on real World Cup data. Both results are recorded honestly in the research log.

​
Run-001: Candidate CLV Signal

The CumulativeDriftAgent averaged +61.19 bps CLV across 18 World Cup fixtures (market-quality-filtered eligible universe, pre-kickoff decisions, CLV vs the CON-040 kickoff close), beating all three acting deterministic baselines on this sample (favorite +4.61 bps, threshold-move −126.5 bps, seeded-random −341.67 bps), net-positive on 10 of 18 fixtures. This is a candidate rung-1 CLV signal, not proven executable alpha:
  • CLV is measured against the sharp TxLINE close with no venue leg.
  • The effective sample is approximately 18 fixtures (the ~19.7k drift picks are autocorrelated intra-match).
  • This result is directional evidence, not a statistical proof.
This result did not survive out-of-sample testing. The surviving OU-totals sub-signal falsified on its own metric (in-sample +162 bps → out-of-sample −254 bps, fixture-clustered) and came out NULL on independent settled outcomes at only N=2 effective fixtures. It is kept as a benchmark, not a promoted edge.

​
Run-002: No Executable Venue Edge Demonstrated

Run-002 tested whether Run-001’s candidate signal translated into Polymarket venue edge. The rung-2 venue lane works — it priced 94.7% of drift’s in-scope 1X2 decisions against time-aligned Polymarket mids under a bounded-staleness bound. But the estimated-mids result was a near-perfect monotonic longshot ramp (+607 bps on 0–20% probability longshots decaying to +33 bps on 80–100% favorites) — the fingerprint of a favorite-longshot / de-margin-scale divergence, not strategy-specific dislocation. Run-002 did NOT demonstrate executable venue edge. These are estimated mids, not fills. real_executable_edge_bps = None. No profit and no fillability claim.

​
Solana Anchoring

  • Not all runs are anchored. Anchoring requires SOLANA_KEYPAIR_PATH and anchor=true. Offline replay runs receive an honest anchor_status="not_anchored" and the anchor check reads not_applicable, never a false green.
  • Anchored runs store the run-manifest hash only. The manifest hash is committed to Solana as a Memo transaction. Detailed evidence (sealed run events, score rows, receipts) remains off-chain.
  • Only Solana devnet is currently supported. Mainnet anchoring is not yet built. The two on-chain transactions referenced in the README (TxLINE subscribe and the run-manifest anchor) are both on Solana devnet.

​
Custody and Payouts

The Prize Vault is designed and visible in the UI but not wired. No payouts have been made. The UI labels this state explicitly — no fabricated payout data appears anywhere in the product.

​
Operator Steps Not Yet Completed

The following steps remain ahead before real-money execution is possible. Each requires a human operator acting directly — no automated path can substitute. Full details are in the operator runbook (available to operators separately).

​
Effective Sample Size and Statistical Significance

  • The real-fixture backtest used 18 World Cup fixtures.
  • Sub-signals have smaller effective N. The OU-totals sub-signal collapses to approximately 2 effective fixtures (match totals), which is the primary reason the out-of-sample test was underpowered.
  • Statistical significance has not been established for any result reported by the platform.

​
Raw Data

Real-fixture ReplayPacks (licensed TxLINE odds history) stay local and are not redistributed. Only synthetic packs are included in the repo. The demo script (scripts/demo_phase2d.py) self-labels synthetic data on every surface — pack manifest, console summary, and proof card.

​
Summary Table