Everything about Alpha score.

wavealgo is a neutral measurement system for trading strategies. It backtests Pine Script strategies across 6 markets (BTCUSD, ETHUSD, SPX500, QQQ, GOLD, EURUSD) and 3 timeframes (1H, 4H, 1D) — 18 cells in all — producing an alpha score from 0 to 100 and one of four verdicts: Verified, Conditional, Repaints, or No alpha. It is not a signal generator, broker, or indicator seller — it is an independent grader that applies the same rules to every script.

The alpha score runs 0 to 100 and is built from four measured parts: out-of-sample edge (0-40), regime stability across the 18 cells (0-25), a flaw penalty (0 to -25) and reproduction accuracy against an independent engine (0-10). A strategy must work across markets to score well — a single-symbol overfit that shines on Bitcoin and fails everywhere else loses almost all of the stability component. Alongside the number sits one of four verdicts: Verified, Conditional, Repaints, or No alpha.

wavealgo tests every strategy against 6 markets: BTCUSD (via Binance BTCUSDT), ETHUSD (via Binance ETHUSDT), SPX500 (via SPY proxy from Yahoo Finance), QQQ, GOLD (via GLD proxy), and EURUSD. This covers crypto, equities, commodities, and forex — ensuring strategies are evaluated across diverse market conditions.

Three timeframes: 1H, 4H and 1D. Cells are deliberately not equal length — each uses whatever history its source provides, so daily cells run 3,278–4,000 bars, 4H runs 1,707–19,653 and 1H runs 5,090–78,545. That is why the composite Sharpe is the median of the 18 cells rather than the mean.

Each of the 18 cells produces a Sharpe ratio. The composite is the median of the 18 — a median, not a mean, because cell windows differ by an order of magnitude. The alpha score is then built from four measured parts: out-of-sample edge (0 to +40, from the composite Sharpe), regime stability (0 to +25, from how many of the 18 cells are positive), a flaw penalty (0 to -25 for static repaint or look-ahead findings) and reproduction accuracy (0 to +10, from engine parity against an independent implementation). The full formula and the four verdict rules are documented at wavealgo.com/methodology.

Not yet. Scoring requires a strategy() declaration with strategy.entry and strategy.close calls, because there must be entry and exit rules to measure. Indicator-only scripts (those with only indicator() and plot calls) still compile and chart in the playground, but they cannot be graded.

Yes — all of it. The leaderboard, the methodology, the corpus sweep and the playground are free with no account and no run quota. There are no paid plans today; if that changes, the terms will be updated before anything is charged.

TradingView's Strategy Tester evaluates one symbol and one timeframe at a time. wavealgo tests across 18 cells simultaneously (6 markets × 3 timeframes) with uniform fee assumptions. This cross-market validation reveals overfit strategies that look profitable on a single chart but fail everywhere else — something TradingView's tester cannot show.

Backtrader and QuantConnect are execution frameworks — you write Python code and run backtests. wavealgo accepts Pine Script directly (no manual porting required) and provides a standardized, comparable grade. wavealgo is a measurement tool for evaluation and triage, not an execution platform for live trading.

All strategies are tested with 0.05% commission per side, which is the midpoint between Binance spot fees (0.10%) and a discount stock broker (0.005%). Initial capital is 100,000, sizing is 100% of equity per entry (long-only), and market orders fill at the next bar open — the rule TradingView documents. Slippage is not modeled in v1. Full assumptions are documented on the Methodology page.

The v1 reference catalog is the twenty best-known public techniques, and none of them earned Verified: 16 came back Conditional, 4 had no alpha at all, and the highest score was 49 of 100. Seventeen of the twenty finished behind simply holding the asset. A wider sweep of 12,424 public scripts is published separately on the corpus page.

Single-symbol overfit is the most common failure mode. A strategy tuned on one market (typically BTCUSD) and never tested elsewhere will often score well on that market but collapse on the other five. The 18-cell matrix is designed to catch exactly this pattern.

Not running yet. The plan is a Model Context Protocol server that exposes the same 18-cell grading to Claude Desktop, Cursor and other MCP clients, so a strategy can be evaluated without leaving the editor. Until it is hosted, the execution engine underneath is open source (github.com/nullarch/resin) and can be wired into an MCP server of your own today.

Yes. The Pine Script compiler and runtime that executes every audited script is resin, Apache-2.0, at github.com/nullarch/resin. It is verified bar-by-bar against an independent implementation on 251 of 257 corpus scripts, and it is the same engine running in your browser on the playground. A companion adapter renders its output on TradingView’s open-source lightweight-charts.

Yes. The full methodology — including the 18-cell matrix, the alpha formula, the verdict rules, fee assumptions, and known limitations — is documented at wavealgo.com/methodology. The execution engine is open source at github.com/nullarch/resin. There is no hidden model, no stochastic component, and no per-user tuning.

Not automatically yet. The playground compiles and charts any Pine script in your browser, but it does not grade — grading is an offline pipeline that runs the whole 18-cell grid under fixed rules. To have a strategy measured, submit it through the audit request form.

wavealgo supports Pine Script v5 and v6. The engine handles strategy declarations, indicator functions, UDFs with var/if/for/while blocks, and per-bar execution. Known limitations include inline cross calls inside if blocks (bind to a variable instead) and some advanced request.security patterns.