Twenty classics.
Zero passed.
Every one ran on the same 18 market × timeframe cells, out of sample, with a 0.05% fee on every fill. 16 came back Conditional, 4 had no alpha at all, and 17 of 20 finished behind simply holding the asset.
//@version=5strategy("Golden Cross (SMA 50/200)", initial_capital=100000, default_qty_type=strategy.percent_of_equity, default_qty_value=100)fast = ta.sma(close, 50)slow = ta.sma(close, 200)longSig = ta.crossover(fast, slow)exitSig = ta.crossunder(fast, slow)if longSigstrategy.entry("L", strategy.long)if exitSigstrategy.close("L")plot(strategy.equity, "equity")
The whole ledger, including the ugly rows.
open the desk →Paste a Pine script. It runs here, in this tab.
No upload, no account, no server round-trip — the compiler that produces every verdict on the desk is Apache-2.0 and it is running in your browser right now.
- Do the classic Pine Script strategies actually work?
- Measured across 18 market and timeframe cells, out of sample, with a 0.05% fee on every fill: none of the 20 best-known public techniques earned a Verified verdict. 16 came back Conditional, meaning an edge appeared in part of the grid but not across it, 4 showed no alpha at all, and 17 of 20 finished behind simply holding the asset.
- Is the Golden Cross strategy profitable?
- It is the highest-scoring script in the catalog at 49 out of 100, with a composite Sharpe of 0.78 across the 18 cells and 17 of them positive. Its verdict is Conditional, not Verified, and it beats buy-and-hold by 783.1 percentage points over the same window.
- How many public Pine scripts survive a fair test?
- In a sweep of 12,424 public scripts, 4,595 measurements remained after eligibility filtering and duplicate collapse. Four guards were applied — positive out-of-sample Sharpe, a 20 basis point cost floor, beating buy-and-hold, and a deflated Sharpe of 0.95 — and two cleared all four. Neither has fills that daily bars can confirm.
- Can you run Pine Script outside TradingView?
- Yes. resin is an open-source (Apache-2.0) compiler and runtime for Pine v5 and v6 that turns a script into a plain JavaScript module, so it runs in a browser, in CI, or in a backtest of your own. It is verified bar-by-bar against an independent implementation on 251 of 257 corpus scripts, and it is what executes every script measured on this site.
- Why do backtests look better than live results?
- Three reasons dominate, and each is a rule here rather than an opinion. Parameters fitted on the same window they are scored on will flatter themselves, so the measured window is out of sample. Costs compound, so a fee is charged on every fill rather than subtracted at the end. And beating zero is not an edge, so the comparison is against holding the asset over the identical window.
It means an edge appeared in part of the grid — which is also what overfitting looks like from the outside.
Orders follow TradingView’s next-bar-open rule, but have not been matched trade-by-trade against it. Treat absolute returns as provisional.
The reference catalog is the classics. The wider corpus sweep is where the real distribution lives.
The rubric, the 18-cell grid and the exact source of every audited script are public, and the engine that ran them is open source. Rerun any row and disagree with a number — a grade nobody can check is just an opinion with a font.
Engine pine2js (next-bar-open fills, TV rule) · scored UTC 2026-08-08 04:53 · oracle parity 251/257 · deterministic rerun — same data, same numbers.