OPEN-SOURCE SCRIPT
Verdict Calibrator Edge vs Base Rate
Overview
A meta-tool that answers one honest question about any signal: does it actually beat chance?
Point it at another indicator's output (via the source input) or use a built-in reference signal, and it forward-tests every long/short call with a triple-barrier outcome, compares the hit rate to the unconditional base rate of the same move, and only calls an edge "proven" when a confidence-interval lower bound clears that base rate. It splits the result by side (long vs short) and by regime (trend vs range).
It grades a signal; it does not make one.
Why this exists
Most "win rate" readouts are misleading. 60% right means nothing until you know how often the same move happened anyway — if price rose 60% of the time regardless, your signal has an edge of exactly zero. And a 60% on 12 samples is noise, not evidence.
This tool is built so it can only ever say "no proven edge" when there isn't one. The base-rate comparison and the significance gate make false confidence structurally hard to produce.
Why these parts are ONE tool
Signal as a parameter. Connect any plot on your chart through the source input, or pick a built-in reference. The signal is an input, not baked in — so you can audit your indicators, not just this one. The ten built-ins span the families traders actually use:
FamilyBuilt-insTrend-followMA cross · MACD cross · Supertrend flipMomentumRSI 50-cross · Stochastic crossBreakoutDonchian breakout · VWAP crossMean-reversionRSI 30/70 reversal · Bollinger fadeControlRandom (coin flip)
The random control — the tool's own self-test. A deterministic coin flip that by construction has no edge. Grade it and the verdict should read "Not proven" with an Edge near zero. If a random signal ever comes back "PROVEN", the harness is broken — distrust the tool, not the market. No other calibrator on TradingView ships with a falsification test built in. It is also the single fastest way for a sceptical user to satisfy themselves that this thing is honest.
Triple-barrier outcome. From each signal: did price reach +target, −target, or neither within the horizon? A well-defined outcome, not a vague "did it go up eventually".
Base-rate comparison. The honest yardstick — the unconditional rate of the same outcome, matched to the signal's own side and regime mix. Edge = Hit% − Base%, never raw Hit%.
Significance gate. A score-interval lower bound must clear the base rate before an edge is called proven — which matters most at small samples, exactly where point estimates lie.
Regime and side split. Edge is reported for long vs short and trend vs range separately, because a real edge usually lives in one and not the other.
Remove any one and the tool can be fooled into reporting confidence it hasn't earned.
How to use it
Pick the signal (external source or a built-in), set the outcome (horizon + target in ATR), and read the verdict: PROVEN +X% ★ / Not proven / Gathering data.
The two plotted lines are the running Hit % (of the signal) and Base % (unconditional) — the gap between them IS the edge, and you can watch it stabilise as samples accumulate. Switch the dashboard to Pro to see where the edge lives (long/short, trend/range).
If it says "no proven edge", believe it. That's the tool working, not failing.
Data & scope
Works on any symbol and timeframe — it needs only OHLC, no volume. Give it enough history to reach the minimum sample count, or the verdict will honestly read "Gathering data". Because the base rate is measured on the same chart, the yardstick always matches the instrument you're on.
Non-repainting & honest limits
Confirmed-bar reads; samples log at the signal bar and resolve on closed bars.
This is an in-sample, forward-from-signal study aid — NOT a walk-forward backtest. No costs or slippage. Overlapping forward windows correlate samples (the minimum-gap setting mitigates this; it does not eliminate it). Small-n edges are provisional even when starred. A proven in-sample edge is not a guarantee out-of-sample. Nothing here predicts price.
Concept credits
Built on standard, published techniques — triple-barrier forward labelling (M. López de Prado), base-rate / skill-vs-chance evaluation (a long tradition in forecast verification), the efficiency-ratio regime read (Perry Kaufman), and the Wilson score interval for a proportion (Edwin B. Wilson). The signal-agnostic intake, the coupling and the plain-language verdict are this script's own. No third-party Pine code is reused.
Disclaimer
Research and educational tool only. Not financial advice, no recommendation, no guarantee of results. Indicators describe past behaviour; they do not predict the future. Trading carries risk of loss. Test out-of-sample and make your own decisions. The author accepts no liability.
A meta-tool that answers one honest question about any signal: does it actually beat chance?
Point it at another indicator's output (via the source input) or use a built-in reference signal, and it forward-tests every long/short call with a triple-barrier outcome, compares the hit rate to the unconditional base rate of the same move, and only calls an edge "proven" when a confidence-interval lower bound clears that base rate. It splits the result by side (long vs short) and by regime (trend vs range).
It grades a signal; it does not make one.
Why this exists
Most "win rate" readouts are misleading. 60% right means nothing until you know how often the same move happened anyway — if price rose 60% of the time regardless, your signal has an edge of exactly zero. And a 60% on 12 samples is noise, not evidence.
This tool is built so it can only ever say "no proven edge" when there isn't one. The base-rate comparison and the significance gate make false confidence structurally hard to produce.
Why these parts are ONE tool
Signal as a parameter. Connect any plot on your chart through the source input, or pick a built-in reference. The signal is an input, not baked in — so you can audit your indicators, not just this one. The ten built-ins span the families traders actually use:
FamilyBuilt-insTrend-followMA cross · MACD cross · Supertrend flipMomentumRSI 50-cross · Stochastic crossBreakoutDonchian breakout · VWAP crossMean-reversionRSI 30/70 reversal · Bollinger fadeControlRandom (coin flip)
The random control — the tool's own self-test. A deterministic coin flip that by construction has no edge. Grade it and the verdict should read "Not proven" with an Edge near zero. If a random signal ever comes back "PROVEN", the harness is broken — distrust the tool, not the market. No other calibrator on TradingView ships with a falsification test built in. It is also the single fastest way for a sceptical user to satisfy themselves that this thing is honest.
Triple-barrier outcome. From each signal: did price reach +target, −target, or neither within the horizon? A well-defined outcome, not a vague "did it go up eventually".
Base-rate comparison. The honest yardstick — the unconditional rate of the same outcome, matched to the signal's own side and regime mix. Edge = Hit% − Base%, never raw Hit%.
Significance gate. A score-interval lower bound must clear the base rate before an edge is called proven — which matters most at small samples, exactly where point estimates lie.
Regime and side split. Edge is reported for long vs short and trend vs range separately, because a real edge usually lives in one and not the other.
Remove any one and the tool can be fooled into reporting confidence it hasn't earned.
How to use it
Pick the signal (external source or a built-in), set the outcome (horizon + target in ATR), and read the verdict: PROVEN +X% ★ / Not proven / Gathering data.
The two plotted lines are the running Hit % (of the signal) and Base % (unconditional) — the gap between them IS the edge, and you can watch it stabilise as samples accumulate. Switch the dashboard to Pro to see where the edge lives (long/short, trend/range).
If it says "no proven edge", believe it. That's the tool working, not failing.
Data & scope
Works on any symbol and timeframe — it needs only OHLC, no volume. Give it enough history to reach the minimum sample count, or the verdict will honestly read "Gathering data". Because the base rate is measured on the same chart, the yardstick always matches the instrument you're on.
Non-repainting & honest limits
Confirmed-bar reads; samples log at the signal bar and resolve on closed bars.
This is an in-sample, forward-from-signal study aid — NOT a walk-forward backtest. No costs or slippage. Overlapping forward windows correlate samples (the minimum-gap setting mitigates this; it does not eliminate it). Small-n edges are provisional even when starred. A proven in-sample edge is not a guarantee out-of-sample. Nothing here predicts price.
Concept credits
Built on standard, published techniques — triple-barrier forward labelling (M. López de Prado), base-rate / skill-vs-chance evaluation (a long tradition in forecast verification), the efficiency-ratio regime read (Perry Kaufman), and the Wilson score interval for a proportion (Edwin B. Wilson). The signal-agnostic intake, the coupling and the plain-language verdict are this script's own. No third-party Pine code is reused.
Disclaimer
Research and educational tool only. Not financial advice, no recommendation, no guarantee of results. Indicators describe past behaviour; they do not predict the future. Trading carries risk of loss. Test out-of-sample and make your own decisions. The author accepts no liability.
オープンソーススクリプト
TradingViewの精神に則り、このスクリプトの作者はコードをオープンソースとして公開してくれました。トレーダーが内容を確認・検証できるようにという配慮です。作者に拍手を送りましょう！無料で利用できますが、コードの再公開はハウスルールに従う必要があります。
免責事項
これらの情報および投稿は、TradingViewが提供または承認する金融、投資、取引、またはその他の種類の助言もしくは推奨であることを意図したものではなく、またこれらに該当するものでもありません。詳細は利用規約をご覧ください。
オープンソーススクリプト
TradingViewの精神に則り、このスクリプトの作者はコードをオープンソースとして公開してくれました。トレーダーが内容を確認・検証できるようにという配慮です。作者に拍手を送りましょう！無料で利用できますが、コードの再公開はハウスルールに従う必要があります。
免責事項
これらの情報および投稿は、TradingViewが提供または承認する金融、投資、取引、またはその他の種類の助言もしくは推奨であることを意図したものではなく、またこれらに該当するものでもありません。詳細は利用規約をご覧ください。