Most backtests are lies.
Run the gauntlet.
Every other tool is built to tell you your edge is real. EdgeProven runs the deterministic de-bias suite quants run before they trust a number — variant-deflation, regime split, tail & single-name dependence, real-cost re-pricing. Same trades in, same verdict out.
Every other tool in this category is engineered to confirm your edge. We engineered the opposite — and then proved we meant it by killing six of our own.
- No signals to sell
- No return promises
- No “this will make money”
- No live-order path — paper-only, in the code
- You can’t buy a better verdict
Proof we can actually stand behind
The falsification wall
We couldn’t fool our own validator. Now try to fool it with yours.
Six edges we built, hoped would work, and killed when they failed the gauntlet. No competitor publishes their own failures — which is exactly why this is the one proof a five-user startup can’t fake.
- ✗ killedPairs reversion
→ hooked us Clean mean-reversion, in-sample.
! cause of death No robust out-of-sample edge across two independent rounds. Stopped a losing auto-trader before it cost us.
- ✗ killedThe scanner zoo
→ hooked us Survivor-inflated win rates across ~30 scanners.
! cause of death 15-year anti-survivor gauntlet: only one thin scanner is all-weather; two were outright negative.
- ✗ killed13F overlay
→ hooked us Looked like +4%/qtr buy-minus-sell.
! cause of death An event-bucket artifact — double-counted buckets plus one GME tail. No tradeable portfolio survived de-tailing — kept only as a genuine decorrelated diversifier, never as standalone alpha.
- ✗ killedPEAD (earnings drift)
→ hooked us Top-magnitude names, t = 2.71.
! cause of death 76% of the edge was three names (TSLA / AMD / FCX). Ex-top-3, it’s a coin flip.
- ✗ killedInsider clusters (Form 4)
→ hooked us +5.4%/yr, t = 0.96.
! cause of death Dead out-of-sample; negative in the bear; 95% of it was three tickers.
- ✗ killedFundamental quality/value
→ hooked us In-sample t = 2.29.
! cause of death Premium decay — dead out-of-sample (t = 0.51); the long/short leg was negative.
This is the method working. Most edges should die. The rare one that survives — like our put-write research — earns the green light precisely because the other six didn’t.
Try to fool it →How the gauntlet works
Four honest checks, run the same way every time.
Deterministic and reproducible — every check shows its work. Hit “Show the math” on any of them in the validator.
- Deflated Sharpe
We discount your Sharpe for every variant you tried.
Naive backtests report the best run of many and call it skill — that’s luck wearing a suit.
- Regime stratification
We split your record into bull and bear.
A single backtest blends a 2020 melt-up with a 2022 drawdown and hides that you only work in one.
- Tail & single-name dependence
We check whether a few names or days carry the whole result.
An average return looks systematic right up until you learn 71% of it was three tickers.
- Cost sensitivity
We re-price with realistic round-trip costs.
Plenty of edges are real on paper and gone after the spread.
The honesty ladder
“Survives these checks” is as high as it goes.
Fails a check — and we show you exactly which one, with the math.
Survives, but thin — leans on a tail, a thin sample, or one fragile spot.
Clears all four — and that is not a promise it will make money.
If you ever see us print “validated” or “profitable,” we broke our own rule. Tell us.
From verdict to proof
Free gives you the verdict. Paid gives you the receipt.
A signed, tamper-evident, shareable report — a permanent link you can hand to an allocator, a partner, or future-you, with an embeddable badge they can verify hasn’t been altered. You’re paying for the permanent record and the signature, never a better grade. The verdict is identical whether you pay or not.
The rest of the bench
A full research bench — and everything runs the gauntlet first.
Design a strategy, get an AI second opinion, paper-trade it — but nothing earns a green light until it survives the same honest checks.
AI second opinion
A senior trader to sanity-check every order — never to tell you to buy.
Submit a proposed entry, stop, and target. Claude returns a thesis-check, a risk-factor breakdown, and a concrete list of what would invalidate the setup. Analytical, never directive. Streams in real time and saves every analysis with its order context.
- Bull / bear / risk decomposition with severity tags
- Streams output token-by-token — no spinner-fest
- Analytical language by design — never “buy this”
Pullback to rising 21-EMA on declining volume. R:R 2.0, in line with regime defaults.
Earnings in 4 trading days. AI suggests halving size or moving stop to $458.50.
If SPX breaks 4400, semis tend to decouple to the downside on the day. Watch /ES at 09:30.
Strategies + Backtests
Build in plain English. Then it has to survive the gauntlet.
Describe a setup in natural language and the AI Playground returns a validated DSL strategy. Or paste Python. Either way it's schema-validated and smoke-tested before it touches your saved strategies — and you can run the de-bias gauntlet on the result.
- Plain-English → typed strategy with schema validation
- Backtests run against historical bars
- Run the validator on the result before you trust it
{
"name": "52w breakout, low-vol",
"entry": "high.52w × + atr.20<2 + rvol>1.5",
"stop": "atr × 2",
"target": "3R",
"size": "0.5% equity risk"
}Auto-Trader
Hands-off paper trading with the safety rails you set.
The bot polls signals every minute, runs them through a rule engine you control — daily-loss cap, consensus filter, time-of-day window — and submits bracket orders into Alpaca's paper sandbox. Kill switch is a single click. There is no live-order path.
- Bracket orders: entry + stop + target, linked at the broker
- Drift detection on your win-rate — surfaces regime shifts
- Paper-only — the live-order path doesn’t exist in the code
| Time | Sym | Side | Qty | Status | P&L |
|---|---|---|---|---|---|
| 09:31:04 | AAPL | BUY | 14 | FILLED | +$182 |
| 10:05:22 | NVDA | BUY | 6 | FILLED | +$340 |
| 10:42:11 | MSFT | BUY | 12 | FILLED | — |
| 11:18:50 | TSLA | BUY | 9 | SKIPPED | — |
And the rest of the toolbox
Everything else you’d expect — and a few things you wouldn’t.
Scanners
ORB / VWAP / RVOL / failed-breakout / gap-fill across a 400-name day-trade universe.
Movers
Today's biggest gainers, losers, and volatility leaders with catalyst-aware ranking.
Watchlist
Your tracked symbols, unioned into every scanner and the daily brief.
Fast-Tick
WebSocket stream with sub-minute setups — ORB, momentum, mean-reversion, stat-arb pairs.
Swing Signals
Source-performance panel tracks realized P&L per feed so you know which to trust.
Email Digest
Pre-open + post-close email digest with gappers, top setups, and open-position commentary.
Built with the tools traders trust
- Powered by ClaudeAnthropic API
- Brokered by AlpacaPaper trading sandbox
- Real market dataLive + historical bars
- Encrypted at restAES-256-GCM
How we treat your account
Boring by design where it counts.
No live order path
The codebase literally cannot place a live order. Every executor talks only to Alpaca's paper sandbox; we accept no live-trading flag at any layer.
Your keys, encrypted at rest
Alpaca API credentials are stored under AES-256-GCM with a master key kept in our deployment secret store. You can rotate or clear them from Settings any time.
Per-user data scoping
Every protected route resolves to your user and every database read is auto-filtered to your rows. You see only your strategies, your decisions, your trades.
Informational only
EdgeProven is not a broker, not a financial advisor, and nothing on this site is investment advice. AI critiques use analytical language by design.
FAQ
The questions an honest tool should expect.
Is this just talking your book?
No. We don't sell signals and we don't trade against you — there's no live-order path in the product at all. The validator is deterministic: same trades in, same verdict out. You can't pay us for a better grade, and we publish the six of our own signals we killed.
Why would a tool tell me my edge is fake?
Because most edges are. Overfitting, one lucky name, and costs kill more strategies than bad ideas do. We'd rather you find that out for free, on paper, than with money. When something genuinely survives, that verdict means more because we kill the rest.
Does passing mean it works?
No. The highest verdict we print is “survives these checks.” It means your strategy cleared an honest de-bias gauntlet — not that it will make money. Markets change; past performance isn't predictive. We will never print “validated” or “profitable.”
Is this real trading? Is it advice?
Neither. Every order routes through Alpaca's paper sandbox — the live-trading code path doesn't exist. And EdgeProven is an impersonal research tool and publisher, never an adviser. The AI is analytical — thesis, risk, counter — never “buy this.”
What do the paid plans add?
Pro ($29/mo) unlocks unlimited runs of the Edge Lab validation gauntlet plus the AI second-opinion. Premium ($79/mo) adds the Premia Book and the all-weather research suite — alpha decomposition, decorrelation, and 13F awareness — research and honest verdicts, never signals. Neither buys a better verdict; the gauntlet is deterministic and identical whether you pay or not.
How much does it cost?
The validator is free, no signup. A free account adds saved verdicts, the AI analyzer, and paper trading. Pro ($29/mo) adds unlimited Edge Lab validation; Premium ($79/mo) adds the Premia Book and the research suite. The signed, shareable report is a separate per-run purchase (Validation-as-a-Service), not a subscription perk — you pay for the artifact, never a better grade. The free tier keeps working regardless.
Bring a backtest you believe in. Let’s try to break it.
Free, no signup, no card. Paste your trades and get an honest verdict in seconds — then create a free account to save it and validate the rest of your strategies.
By signing up you agree that this tool is informational and not investment advice.
Honest edge validation and paper trading with AI-assisted analysis. Most backtests are lies — we test yours like it's our money. Not a broker. Not personalized investment advice.
Important. EdgeProven is an informational research and paper-trading tool. Nothing on this site is investment advice, an offer to buy or sell securities, or a recommendation tailored to your situation. Trading involves risk; past performance is not predictive of future results. Backtests use historical data and may not reflect live execution conditions. AI-generated analyses are probabilistic and may be wrong.
© 2026 EdgeProven · Paper trading only.