Indicators we build — and test in public
Indicators our AI just made
Every night the forge invents new indicator formulas and runs each through the same gauntlet as everything else here: out-of-sample testing, realistic costs, and a multiple-testing hurdle that RISES with every attempt. Survivors publish with the formula shown in full and a public forward track record from birth. Most nights nothing survives — that is the gate working.
Nothing has survived the gauntlet yet. The forge keeps inventing nightly — most candidates die in testing, which is the point.
22 indicators, every one verified
150/1527 beat buy-and-hold, 831 OOS-qualified - low-lag Ehlers filter.
177/1521 beat buy-and-hold, 1,005 OOS-qualified.
DecisionPoint PMO. 171/1492 beat buy-and-hold, 1,134 OOS-qualified - the strongest of this batch.
Dual-CCI trend system. 176/1531 beat buy-and-hold, 864 OOS-qualified.
165/1533 beat buy-and-hold, 804 OOS-qualified.
Pivot-anchored SuperTrend. 135/1324 beat buy-and-hold, 1,052 OOS-qualified.
Wilder ATR trailing stop. 145/1484 beat buy-and-hold, 1,087 OOS-qualified.
MACD on zero-lag EMAs. 152/1529 beat buy-and-hold, 1,022 OOS-qualified.
Pullback-entry oscillator. 141/1523 beat buy-and-hold, 830 OOS-qualified.
Regime filter that sits out chop. 115/1455 beat buy-and-hold, 741 OOS-qualified.
Gauntleted across 741 markets. Beats buy-and-hold on only 170/1489 - a trend FILTER, not a standalone edge - but its real edge is intraday (e.g. AAPL 4h: Sharpe 0.93, +3.2% alpha). 1,111 setups held up out-of-sample.
Auto-trendline breakout. Beats buy-and-hold on 147/1318 markets, 1,042 OOS-qualified - works on trending names, fades on chop.
The volatility-regime edge is real but narrow - 94/1467 beat buy-and-hold. Best as a risk-ON filter (calm vol + uptrend), not a return driver.
An OHLCV proxy: with no tick/order-flow data, we sign each bar's volume by where it closes in its range and accumulate it (true CVD needs order-flow we don't have). The honest result: across 1,400+ asset-timeframe tests it beat buy-and-hold AND survived out-of-sample on only ~6-7% of cases, and among 360+ indicators it's almost never the top pick. The faint edge that survives sits almost entirely in high-volume crypto (KuCoin, Algorand, Coinbase, Hedera, Canton) where volume-pressure means more — and is mostly noise everywhere else. A narrow, asset-specific signal, not the broad alpha the hype implies.
The standout. Best indicator on 14 assets and a genuine volatility-regime edge on VIX (out-of-sample Sharpe 1.18 over 547 trades). But it loses to buy-and-hold on BTC, SPY and AAPL — the viral "60x" is the asset rising, not alpha.
Adds a 'hidden Markov' confirmation gate so it trades less. Wins on 1 asset (Fortinet). Trading less didn't make it better — a clean negative result worth knowing.
Momentum continuation (buys a z-score breakout). Wins on 0 assets as the #1 fit; avg Sharpe 0.26. The card claimed Sharpe 5.36 — costed and out-of-sample, it doesn't beat buy-and-hold.
Net signed-volume flow oscillator. Avg Sharpe 0.31, but wins 0 assets — no broad edge once costs and an out-of-sample split are applied.
Rides a stacked-EMA trend. Big raw returns on trending names (it's just long the trend) but negative alpha almost everywhere; best fit on 1 asset (Kimco Realty).
Longs only in an uptrend, exits fast. Best on 1 asset (Warner Bros. Discovery); avg Sharpe 0.16. Mostly underperforms buy-and-hold.
Buys confirmed 2x-volume spikes. The weakest of the batch — avg Sharpe 0.09 — though it is the best fit on 3 niche assets (Soybean Oil, Ondo, Incyte).
Fades a 2-sigma stretch when signed-volume flow shows sellers are exhausted. High win-rate reversion that gives it back on big moves; best on 2 assets (Datadog, Uber), avg Sharpe 0.25. The card's +458% was 15m-specific overfit.
Most custom indicators — including most of these — do not beat buy-and-hold once you add costs and an out-of-sample test. That is the single most useful thing we can tell you. Anyone can post a winning backtest; almost no one shows you the ones that quietly lose. Educational information only — not investment advice. Hypothetical backtested results; past performance does not guarantee future results. Trading involves risk of loss.
Get pinged when an edge flips
Pick the tickers you care about and we'll email you the moment their best indicator flips long, flat, or short — or leave it blank to get every flip. No app, no noise.