Financial astrology, actually tested

The thesis, stated plainly

There is no established evidence that planetary positions predict returns. That sentence appears throughout this product, and it is not a disclaimer bolted on for safety — it is the premise the whole thing is built to test.

The working position practitioners generally take is narrower and more defensible: astrology supplies a dated hypothesis, and price action supplies direction and confirmation. Astrology gives time; the chart gives the trade. That is a testable claim, and testing it is what this software does.

Why most market astrology research is wrong

Three failures recur, and all three inflate results.

First, no baseline. "The market rose 60% of the time after this aspect" means nothing until you know it rose 58% of the time generally. The number that matters is the excess over an unconditional baseline drawn from the same series, not the raw hit rate.

Second, no correction for multiple testing. Try a few hundred configurations across several horizons and some will clear any threshold by chance. Without a correction you are guaranteed to find something, and you will believe it.

Third, no out-of-sample split, or a random one. A random split leaks, because adjacent sessions share overlapping forward windows — the test set has already seen the answer. The split has to be chronological.

What this platform does instead

Every configuration is measured against an unconditional baseline from the same instrument. Significance is resampled rather than assumed normal. A Benjamini-Hochberg correction is applied across horizons, producing a q-value rather than a bare p-value. The sample is split chronologically and the effect must survive on the held-out portion. A minimum of thirty instances is required before anything is weighted at all.

Four gates, and a factor must clear all four. On most days nothing does, and the answer is FLAT — reported as a finding rather than padded into a view.

What a negative result is worth

A great deal, if you were about to trade on the opposite assumption. The useful output of this software is often "the thing you believed does not survive measurement", and that is worth more than a confident signal built on sixty-eight observations and no correction.