OCTAVIAN APEXTomorrow's news. Today's advantage.
Sign inGet access
Test a thesis

Anyone can backtest.
Almost nobody counts the tries.

State a condition, pick what it should move, and measure it against six years of daily data. The result arrives with its sample size, its confidence interval, a half-split test and a plain-language verdict.

The part that matters is the threshold. Every test ever run here makes the next one harder to pass. Test enough variations of anything and one will look significant — so the bar rises with the count, permanently.

Build it

How to read it

What the verdict means

  1. ROBUST

    Below the family-wide threshold, holds in both halves of the period, and genuinely beats its own baseline. All three, or it is not robust.

  2. SIGNIFICANT · NOT ROBUST

    Would pass the usual p < 0.05 on its own, but not once every other test is counted. Most interesting-looking results land here — including the example above.

  3. WEAK

    Above the baseline, but not far enough to separate from chance at this sample size.

  4. NO ROBUST EDGE

    Enough data to answer, and the answer is no: the condition does no better than an ordinary day.

  5. INSUFFICIENT DATA

    Fewer than ten signal days. Whatever the percentage says, it is not evidence yet — and it is re-checked on every later signal day.

Why the threshold moves

Run twenty tests at p < 0.05 and one will pass by luck alone. That is not a flaw in the tests; it is what the number means. The usual fix is to correct for how many tests were run — and the usual practice is to quietly forget the ones that failed.

Here every test is written to a permanent log, and the threshold is computed across all of them. A test you run today is measured against a bar that your earlier attempts helped raise. It is the difference between a search and a measurement.

Already measured

Twenty things everyone believes

Before this page existed we ran the most repeated claims in markets through the same engine — gold and the dollar, Bitcoin and the Nasdaq, buying extreme fear. Each one has its own page with the measurement behind it.

  1. Does gold rise when the dollar falls?

  2. Does Bitcoin follow the Nasdaq?

  3. Does Bitcoin rise when the dollar falls?

  4. Does gold rise when real rates fall?

  5. Should you buy Bitcoin when Fear and Greed shows extreme fear?

  6. Does silver follow gold?

  7. Does Bitcoin bounce after a 20-day low?

  8. Does a falling Nasdaq lift gold?

  9. Does Ethereum beat Bitcoin when crypto rallies?

  10. Does Bitcoin drop when bonds sell off?

  11. Does a strong dollar hurt oil?

  12. Does high Bitcoin volatility mark a bottom?

And then?

A result that survives does not become a position. It goes into shadow: measured on live data, with no deadline, and counted toward nothing until there is enough of it to say the relationship is still there. Instruments have sat in shadow for months. That is the point of it.

Recording a test — writing it to the permanent log, with its date and its invalidation rule — needs an account. Measuring is free and is not logged.

Get access