EVERY CALL. ON THE RECORD.

No curated backtests. No survivor bias. Every directional signal is measured against real post-signal prices and posted here live — wins, losses, and the types we demote. This page renders the exact gate the pipeline uses to flag conviction: strong, so it can never claim edge the product doesn't actually act on.

proven · live
68%
deleverage · +0.4pts vs BTC
win rate · 24h forward return
2 proven signal types
810 signals measured · 30d
68% best proven win-rate
1h–24h per-type proven horizon · real HL prices

The gate

Every signal type is forced through the same three checks, per direction, regime-aware. Pass all three → the API flags it. Fail any → it stays published here as intel only. No type is grandfathered in.

candidates
loading…
forward-return gate
n ≥ 30 measured signals
beats BTC + profitable, per (type, direction)
regime-aware — gated on the regime that holds
expected move > round-trip cost
proven → conviction: strong
intel only — stays public

The full audit — failures included

Types that don't clear the bar stay on the board as intel only. Public demotion is the feature: when edge decays, this board says so before our marketing does.

methodology

Every directional signal is measured against real post-signal price action — Hyperliquid 1-hour candles — at the horizon where each type is proven (1h, 4h or 24h) over a rolling 30-day window. For each signal type we publish the win rate, the mean return of the trade itself, the mean excess vs BTC over the same window, and 95% confidence intervals. A type is marked proven only when it clears 30+ measured signals AND both beats BTC AND is profitable on its own — judged per (type, direction) and regime, not on the pooled blend. (So a type can read proven here while its pooled excess-return CI, shown in the table, still dips below zero — the pool mixes a winning direction with a losing one; the gate scores them separately.) Beating BTC alone isn't enough — a type that only looks good by being short in a downtrend, while still losing money, does not qualify. The live trade_signal gate applies this same per-(type, direction), regime-aware test, and additionally downgrades any call whose expected move is too small to clear round-trip trading cost (fees + slippage) — a move you can't profit from after costs isn't a trade.

Edge accumulation

Does the per-signal edge above actually compound, or is it a few lucky outliers? Below is an illustrative account starting at index 100 that takes every proven-type signal, one at a time, risking 1% of equity per signal and compounding the realized 24h return — in fire-time order. Fixed fractional size, no leverage, no stacking. Illustrative of edge accumulation on measured returns, not a live-traded or guaranteed account.

loading…

We grade our own grades

Every pre-trade check verdict we hand out is logged the moment we give it and scored at its horizon against real prices. If our grades mean anything, A-graded setups outperform F-graded ones. Small sample for now — we show it anyway, from day one.

loading…

These numbers update whether they flatter us or not.

Pipe the proven signals straight into your agent — REST, WebSocket, Webhook, MCP, or x402.