Strategy adjudication · agent-native · net of fees
EdgeVerdict is a paid endpoint that puts any trading strategy on trial — full history, net of fees, out-of-sample, against buy-and-hold — and returns a ruling your agent can act on. We built it to kill our own ideas. It killed 40 of them.
The docket — three services, one island
Everything the court offers is information, not custody — your keys never leave your account. Priced per-call over x402, agent-native.
Put any textbook strategy on trial against the battery. Net of fees, both disjoint halves. Instant on cache.
The daily target weights of the long-term strategy we run ourselves — a low-turnover rebalancer that trims into strength and buys back weakness, for a smoother ride through cycles (materially lower drawdowns) than white-knuckling every one. You execute; we never touch funds.
Arbitrary specs through the full engine: admissibility oracle, execution realism, multi-asset. Opens on Canon demand.
Exhibit A — the retail canon, on trial
Seven textbook systems. Canonical parameters. Long/flat on BTC daily, 10 bps a side, judged on both disjoint halves of 2018–2026. The bar: beat buy-and-hold in each half, net. A few famous convictions are shown free below in full — don't trust us, reproduce them and verify the court. Which of the rest survive — and which don't — is sealed: build the whole honest-backtest apparatus yourself, or make one call.
| Defendant | Full | 1st half | 2nd half | Trades | Ruling vs baseline |
|---|---|---|---|---|---|
| Buy & hold — required baseline | +784.8% | +412.9% | +68.8% | 1 | — |
| MACD 12/26/9 free sample | +687.3% | +384.4% | +62.5% | 107 | Guilty |
| Golden Cross 50/200 free sample | +623.9% | +288.9% | +86.1% | 8 | Guilty |
| RSI(14) 30/70 free sample | +30.4% | +36.2% | −4.3% | 13 | Guilty |
| Bollinger(20,2) rev | ••• | ••• | ••• | ••• | Sealed |
| Turtle 20/10 | ••• | ••• | ••• | ••• | Sealed |
| TSMOM(90) | ••• | ••• | ••• | ••• | Sealed |
| Supertrend(10,3) | ••• | ••• | ••• | ••• | Sealed |
The samples above are shown free, in full — reproduce them and verify the court. Which of the remaining defendants survive, and which don't, is sealed. Put your own strategy on the stand and find out where it lands.
Famous systems get convicted here. Which of them survive the honest test — and whether any do — is a sealed verdict. Knowing it the hard way means clean multi-source data, a fee-accurate engine, and out-of-sample discipline. Knowing it the easy way is one call. Put your strategy on the stand and find out where it lands.
Procedure
Your agent submits a spec, pays $0.05 over x402, and gets a signed verdict with the full methodology attached. Identical specs return instantly — the court remembers its rulings.
# put a strategy on trial — one paid HTTP call curl -X POST https://verdict.example/v1/verdict/canon \ -H "Content-Type: application/json" \ -d '{"family":"macd","params":{"fast":12,"slow":26,"signal":9},"symbol":"BTCUSDT"}' # → 402 Payment Required → pay $0.05 USDC (x402) → retry → { "ruling": "DOES NOT WORK", "beats_buy_and_hold": { "full": false, "h1": false, "h2": false }, "net_return_pct": { "strategy": 687.3, "buy_and_hold": 784.8 }, "methodology": "net 10bps/side · disjoint halves · pre-registered gate", "caveats": ["family-wise comparison disclosed", "not investment advice"] }
Why trust the court
Any vendor can claim their backtest is great. We're the only one whose pitch is a table of failures — including our own.
The two places every backtest lies. A verdict that skips either isn't a verdict.
The pass rule is declared before the test runs and never renegotiated after it fails.
Missing history returns an error, not a guess. The court would rather say "unknown."
On the record
Our own flagship strategy claimed a large edge. We ran it through these gates, the edge failed its mechanism tests, and we demoted it in writing. The judge ruled against the house — and the ruling stood.
For agents
Agents don't read landing pages. Point yours here.