StillOS · Trust layer for the agent economy

Is your edge real, or did you just get lucky?

Feed us a settled trade history. We run the out-of-sample split, fee-adjust the EV, and hand back one fail-closed verdict — REAL_EDGE, REGIME_LUCK, NEGATIVE_EV, or INSUFFICIENT_DATA — with a cryptographically signed receipt you can hand to anyone.

We did this to our own data first

Earlier today we thought we'd found a real edge in a live Kalshi market: 782 real settled trades, positive expected value, looked strong. Before risking a dollar on it, we ran it through this exact tool.

Sample
782
real settled trades, 2 months
Raw win rate
48.8%
t-statistic
-3.57
need ≥ 1.5 to call it real
Verdict
REGIME_LUCK

The most recent third of the data looked great (+9.1¢/trade). The other two-thirds — the part nobody wants to look at — was net negative (-11¢/trade). That's the whole story of most "edges": real until you check the part of the data you were hoping to ignore.

That's the product. Not a backtesting platform, not a course, not a Discord. One HTTP call, one honest number, a signature you don't have to trust us on.

Try it right now

curl -s -X POST https://nolawealthfinancial.com/notary/grade-strategy \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "you",
    "trades": [
      {"t":"2026-06-01T00:00:00Z","price":0.35,"side":"yes","outcome":1},
      {"t":"2026-06-02T00:00:00Z","price":0.40,"side":"yes","outcome":0},
      {"t":"2026-06-03T00:00:00Z","price":0.30,"side":"yes","outcome":1}
    ]
  }'

Real format: one row per settled trade — t (timestamp), price (your entry price, 0-1), side (yes/no), outcome (1 win / 0 loss). Minimum 30 trades to get a real verdict; fewer than that and you get INSUFFICIENT_DATA — on purpose, because a coin flip isn't a strategy.

The four verdicts, no fifth option

REAL_EDGE+EV, survives the out-of-sample split, t ≥ 1.5
REGIME_LUCKlooks positive, not statistically distinguishable from noise yet
NEGATIVE_EVfee-adjusted expectancy is negative — stop trading this
INSUFFICIENT_DATA< 30 settled trades — come back with more evidence

What's actually under the hood

Chronological train/test split

Most recent 30% held out and never touched during training — the same discipline that killed our own WTI finding today. No peeking.

Fee + slippage adjusted

Every trade is repriced at real transaction cost before any verdict is computed. A "win rate" without cost isn't a number, it's marketing.

t-stat significance bar

Positive expectancy alone isn't enough — it has to clear a real statistical bar (t ≥ 1.5) on the held-out slice before we'll call it REAL_EDGE.

Ed25519-signed receipt

The paid tier returns a signature you — or anyone you show it to — can independently verify without trusting our server at all.

How this compares

Pulled directly from each product's own public pricing page — not our characterization, their numbers.

ApproachSetupCostVerifiable by a third party
Self-hosted DSR/PBO Python librariesInstall, wire your own data, know the mathFree (your time)No — you ran it yourself
VARRD (closest AI-native comparable)Auto-created account, credential stored locally~$0.30/session, $1/edgeNo — tied to your account/session
LuxAlgo AI Backtesting AssistantMust subscribe first — gated to top tier$27.99–$32.99/moNo — proprietary dashboard output
StillOS Edge GateZero signup, everFree (30/day), then $0.05/callYes — Ed25519 signature, verify independently, no account needed on either end

Not a knock on either product — VARRD's statistical guardrails (Bonferroni correction, OOS lock) are genuinely rigorous, and LuxAlgo's platform does far more than this one endpoint. The gap is narrower and more specific than "better": neither publishes an output a third party can verify without an account or without trusting the platform that produced it. That's the one thing we do that we could find nobody else doing.

We're not trying to replace the research layer — we're built to sit downstream of it. Run your discovery in whatever tool finds the idea (VARRD's chat, LuxAlgo's optimizer, your own notebook), then settle the "is this real" question here: same trades, one call, a receipt anyone can check without an account on either end.

Pricing

Free

Unsigned verdict, unlimited use up to the daily limit. Good enough to know the truth. Not enough to show a counterparty.

$0.05 / call x402 · USDC on Base

Same verdict, Ed25519-signed and independently verifiable. No signup, no account, no subscription — pay per call from your own agent or wallet.