Does it work anywhere?
Only two of the thirty-six rules on the scorecard have enough bitcoin history to judge. Four cycles cannot make a sample for a once-per-cycle signal. The only honest route to more episodes is more assets: if a rule fails in bitcoin, ether, Solana, the Nasdaq and the S&P alike, it has failed as a rule. If it holds in all five, that is worth knowing. Six rules, five assets, one engine.
Computed from the daily snapshot for 10 September 2026
Of six rules run across five assets, 1 beats the pooled baseline in the same direction everywhere (Golden cross); 0 beat it pooled but reverse in at least one market; the rest are indistinguishable from chance or have too few episodes. Six rules tested means about 0.6 would look good by luck.
A hit is a one-year forward return above the asset’s own 80th percentile (bottom rule) or below its 20th (top rule), so the base rate is 20% by construction in every asset. Episodes, baseline and interval are the scorecard’s engine.
The assets, and what a hit means in each
| Asset | Days | From | Median 1y | Bottom rule hit: above | Top rule hit: below |
|---|---|---|---|---|---|
| Bitcoin | 4072 | 2015-07 | 76.4% | 218.1% | -21.7% |
| Ether | 3767 | 2016-05 | 37.5% | 600.7% | -37.3% |
| Solana | 1913 | 2021-06 | -15.8% | 348.4% | -57.0% |
| Nasdaq | 4450 | 2009-01 | 19.2% | 30.5% | 5.0% |
| S&P 500 | 2518 | 2016-09 | 14.9% | 24.6% | 3.4% |
The cuts show why one outcome definition cannot serve all five: a bottom-rule hit in ether needs a forward return above 600.7%; in the S&P, above 24.6%. Each rule is asked to beat its own market’s ordinary year.
| Rule | Bitcoin | Ether | Solana | Nasdaq | S&P 500 |
|---|---|---|---|---|---|
| RSI-14 below 30 | -615× | -1312× | +184× | +1516× | +69× |
| RSI-14 above 70 | -112× | +711× | +27× | +125× | -215× |
| Price 20% below its 200-day average | -244× | +46× | +243× | -191× | +811× |
| Price 40% above its 200-day average | -48× | +77× | -203× | –0× | –0× |
| Golden cross | +58× | +38× | +374× | +169× | +64× |
| Death cross | +218× | +198× | +105× | -119× | +34× |
Each cell is one rule in one asset: how far the hit rate sat above or below that asset’s own base rate. Depth of colour is the size of the gap, not confidence — a deep cell on four episodes is still four episodes. The number beneath each is how many times the rule fired there.
RSI-14 below 30
| Asset | Times it fired separate occasions | What followed share of them | Could plausibly be 90% interval | Happened anyway on days it did not fire | Difference percentage points |
|---|---|---|---|---|---|
| Bitcoin | 15 +2 open | 20.0% | 8.3–40.9% | 26.1% | -6.1 |
| Ether | 12 +2 open | 8.3% | 1.9–30.1% | 21.7% | -13.4 |
| Solana | 4 +2 open | 25.0% | 5.8–64.4% | 6.5% | +18.5 |
| Nasdaq | 16 +1 open | 31.2% | 16.1–51.8% | 15.9% | +15.3 |
| S&P 500 | 9 +1 open | 22.2% | 7.6–49.6% | 15.8% | +6.4 |
| Pooled, five assets | 56 | 21.4% | 13.8–31.7% | 18.2% | +3.2 |
Pooled, indistinguishable. And the sign flips: oversold RSI ran ahead of its baseline in the Nasdaq and the S&P and behind it in bitcoin and ether. The scorecard found it worse than the baseline in bitcoin on the doubling outcome; here, on bitcoin’s own percentile outcome, it is worse again. Whatever oversold RSI does in equities, it does not do in crypto.
RSI-14 above 70
| Asset | Times it fired separate occasions | What followed share of them | Could plausibly be 90% interval | Happened anyway on days it did not fire | Difference percentage points |
|---|---|---|---|---|---|
| Bitcoin | 12 +1 open | 16.7% | 5.7–39.9% | 17.7% | -1.0 |
| Ether | 11 +1 open | 18.2% | 6.2–42.7% | 11.2% | +7.0 |
| Solana | 7 +2 open | 14.3% | 3.3–45.2% | 12.6% | +1.7 |
| Nasdaq | 25 +1 open | 16.0% | 7.4–31.2% | 14.9% | +1.1 |
| S&P 500 | 15 +1 open | 13.3% | 4.5–33.4% | 15.2% | -1.9 |
| Pooled, five assets | 70 | 15.7% | 9.9–24.1% | 14.9% | +0.8 |
Seventy episodes across five assets, within a point of the baseline pooled, and within a few points in every asset. Overbought RSI predicts nothing, anywhere. This is the cleanest null on the site.
Golden cross
| Asset | Times it fired separate occasions | What followed share of them | Could plausibly be 90% interval | Happened anyway on days it did not fire | Difference percentage points |
|---|---|---|---|---|---|
| Bitcoin | 8 +1 open | 25.0% | 8.7–54.0% | 20.0% | +5.0 |
| Ether | 8 +1 open | 12.5% | 2.8–41.1% | 9.7% | +2.8 |
| Solana | 4 +1 open | 50.0% | 18.2–81.8% | 13.3% | +36.7 |
| Nasdaq | 9 | 33.3% | 14.2–60.2% | 17.0% | +16.3 |
| S&P 500 | 4 | 25.0% | 5.8–64.4% | 19.3% | +5.7 |
| Pooled, five assets | 33 | 27.3% | 16.6–41.4% | 16.3% | +11.0 |
This is the first rule on this site to beat its baseline, and it deserves its caveats before its credit. Thirty-three episodes clears the twenty-episode floor; the pooled interval’s lower edge sits 0.3 points above the baseline, which is thin. The sign is the same in all five assets, and none of the five is significant on its own. And a golden cross is a trend-following signal: it fires after price has already risen, so a forward return above the asset’s 80th percentile partly measures the persistence of a rise that had already begun. Momentum is the null hypothesis here, not chance. Six rules were tested; about 0.6 would clear a 90% interval by luck.
Death cross
| Asset | Times it fired separate occasions | What followed share of them | Could plausibly be 90% interval | Happened anyway on days it did not fire | Difference percentage points |
|---|---|---|---|---|---|
| Bitcoin | 8 +1 open | 37.5% | 16.1–65.2% | 16.6% | +20.9 |
| Ether | 8 +1 open | 37.5% | 16.1–65.2% | 18.1% | +19.4 |
| Solana | 5 +1 open | 20.0% | 4.6–56.5% | 9.6% | +10.4 |
| Nasdaq | 9 | 11.1% | 2.5–37.7% | 21.8% | -10.7 |
| S&P 500 | 4 | 25.0% | 5.8–64.4% | 22.3% | +2.7 |
| Pooled, five assets | 34 | 26.5% | 16.1–40.3% | 18.9% | +7.6 |
Pooled, the death cross ran 7.6 points ahead of its baseline, but its interval does not clear the baseline, so the verdict is indistinguishable. It also reverses by eleven points in the Nasdaq, where a death cross was followed by an unusually bad year less often than an ordinary day was. A pooled figure that averages a crypto effect against an equity reversal has not shown the rule works anywhere; it has shown it may work in some markets, on too few episodes to say.
Price 20% below its 200-day average
| Asset | Times it fired separate occasions | What followed share of them | Could plausibly be 90% interval | Happened anyway on days it did not fire | Difference percentage points |
|---|---|---|---|---|---|
| Bitcoin | 4 +1 open | 0.0% | 0.0–40.3% | 23.9% | -23.9 |
| Ether | 6 +1 open | 16.7% | 3.8–50.2% | 12.4% | +4.3 |
| Solana | 3 +1 open | 33.3% | 7.8–74.6% | 9.2% | +24.1 |
| Nasdaq | 1 | 0.0% | – | 19.1% | -19.1 |
| S&P 500 | 1 | 100.0% | – | 19.2% | +80.8 |
| Pooled, five assets | 15 | 20.0% | 8.3–40.9% | 18.6% | +1.4 |
Price 40% above its 200-day average
| Asset | Times it fired separate occasions | What followed share of them | Could plausibly be 90% interval | Happened anyway on days it did not fire | Difference percentage points |
|---|---|---|---|---|---|
| Bitcoin | 8 | 12.5% | 2.8–41.1% | 16.0% | -3.5 |
| Ether | 7 | 14.3% | 3.3–45.2% | 7.0% | +7.3 |
| Solana | 3 | 0.0% | 0.0–47.4% | 20.3% | -20.3 |
| Nasdaq | 0 | – | – | 20.9% | – |
| S&P 500 | 0 | – | – | 22.0% | – |
| Pooled, five assets | 18 | 11.1% | 3.7–28.6% | 18.1% | -7.0 |
Six rules were tested
Six rules at 90% confidence means about one in ten — 0.6 — would clear its interval if none of them worked. One did, marginally, and it is the one where momentum rather than chance is the alternative explanation. That is not nothing. It is also not a discovery, and the page would be wrong to present it as one.
The methodology decision
The scorecard’s outcomes — a doubling or a 40% fall within a year — are bitcoin-scale. Equities never double in a year, so applying them across assets scores zero for every episode and zero for the baseline, which means nothing. Here a hit is a one-year forward return above the asset’s own 80th percentile for a bottom rule, or below its 20th for a top rule. The cut was fixed before any result was computed and is the same for every asset and every rule. Its consequence is that the unconditional base rate is 20% everywhere, and every result reads against that. The percentile is computed over each asset’s full history, including the forward windows the rules are then tested on; it is in-sample by construction, in the same way as the scorecard’s baseline, and equally for every rule.
The bitcoin row is not the scorecard’s test. Same rule, different outcome definition, and a series starting in 2015 rather than 2009. It answers “does the rule beat bitcoin’s own base rate” on the same footing as the other four assets. Where it disagrees with the scorecard, that is why.
What pooling assumes
Pooling sums episodes across assets and treats them as independent. Between equities and crypto that is defensible; between bitcoin and ether, which move together, it is not, and the pooled interval is narrower than it should be for that reason. A pooled result is downgraded to “mixed” if any asset with five or more episodes runs the other way by more than five points, because a rule that works in one market and fails in another has not been shown to work.
The same rules scored in bitcoin alone, on the site’s own outcome definition, are on the scorecard.
Recomputed daily by fetch/crossasset.py and published as crossasset.json. Method on the methods page.