The method, on one page
How Thinridge decides whether a trading strategy passes.
In one sentence: Thinridge backtests a mechanical futures strategy exactly as written, on one-minute data from 2008, with commission and slippage on every trade, and gives it a pass only if it beats 5,000 copies of itself that pick the side by coin flip, on the years before 2023, and holds up on five other checks.
- The strategy is locked before it runs. What runs is what was written. A change is a new test, and every test of the same idea is counted: the count goes into the coin-flip result, so the best of thirty tries is not shown as if it were the only one.
- The data. One-minute bars from FirstRate Data, bought, not given. Most markets start in 2008; the shelf runs on data to October 30, 2026. All times are New York time. The data.
- Costs on every trade. $4.20 commission a round turn, plus one tick of slippage against the trade on the way in and on the way out. Orders at the market fill at the next bar’s open; a target must be traded through, not touched. Then the same test at two and three ticks.
- 5,000 coin flips. Copies of the strategy trade the same days, at the same times, with the same costs, but pick long or short by coin flip. The result is how many of the 5,000 did as well or better. A strategy that only buys, or only sells, is compared with copies that keep its side and trade on random days instead.
- The years before 2023 decide. The coin flips and two of the checks are scored on 2008 to 2022, kept apart from the years since. A strategy whose money comes only from 2023 on, or that made money before 2023 and has lost since, fails whatever its score.
- Six checks, 100 points: clean data (5), the years before 2023 (25), year to year (20), the coin flips (15), costs (20), and whether a 10% nudge to a setting breaks it (15).
- Three verdicts. Passed: 70 or more, no check under half its points, no automatic fail, the coin flips beaten by 1 in 20 counting every try, and its range of likely results on the years before 2023 above zero. Close: 40 to 69, or a pass by score that misses one of those last two. Failed: under 40, or failed automatically (too few trades, or the 2023 rules above).
- Published either way. Every verdict of the shelf is on the site, failures included, and none is paid for. Who pays for this.
What a pass does not mean. A pass says the strategy beat the coin flips on years kept apart. It does not say the edge is real, that it will last, or that you could sit through its long flat stretches. Only trading forward can say that: the strategies filed before it happened.
How much of it is luck. Across the shelf’s 535 tests with trades, 59 beat their coin flips by 1 in 20, against about 27 by luck alone; counted across all of them, none clears a 5% false-discovery rate. What that means, worked out: What luck alone would pass.
What it is not. Software testing and statistical analysis of mechanical strategies as submitted. Not advice, not signals, not a recommendation to trade. Hypothetical backtests: one contract, after costs, not real trading. Simulated results have limits: they are not actual trading, they are made with the benefit of hindsight, and no account is likely to achieve results like these. The standard statement.
To cite a result: Thinridge (thinridge.io), a division of Edgeproof LLC; the setup and the market; the verdict and the score; the data, FirstRate one-minute futures, 2008 to October 30, 2026. Every verdict, as a table: verdicts.csv. Our verdicts and the table are free to reuse with credit, under Creative Commons Attribution 4.0; the market data under them is FirstRate Data’s, not ours to license. The long version, with the reasons: How it works.
The table, column by column
verdicts.csv, version 1.1 (November 16, 2026): the first ten columns are as in version 1.0; the rest are new.
- setup, market, market_name: the setup as the shelf names it, and the market.
- trades, net_after_costs_usd: every trade from 2008 (or the market’s first year) to the data’s end, one contract, after costs.
- score, verdict: out of 100; Passed, Close, Failed or No trades.
- page, data, source: the result page; the data and its end; how to cite it.
- flips_as_good_before_2023: how many of the 5,000 coin-flip copies did as well or better on the years before 2023 (k).
- flips_kind: what the copies change: the side picked at random, or the same side at random times.
- p_before_2023: (k + 1) / 5,001.
- p_counting_every_try: the coin-flip result with every try of the same idea counted; a pass needs 0.05 or less.
- luck_share_across_shelf: Benjamini–Hochberg across every shelf test with trades: of results this good or better, the share luck alone would account for.
- likely_low_before_2023, likely_high_before_2023: the range of likely results on the years before 2023; a pass needs the low end above zero.
- net_before_2023, net_since_2023, trades_since_2023: the split the verdict uses.
- decided_by: what decided the verdict, in a few words.
- run_date, table_version: the day the test was run; the version of this table.