Evidence & Limits
Last run: 19 October 2026 · 1,960 US sessions · 52,920 star-days per index
The short version: we tested whether a GO verdict marks a better session than an AVOID verdict. Over seven and a half years of US sessions, on SPY and QQQ, it does not. The measured difference is indistinguishable from zero, and the score has no correlation with returns. We publish that here because a framework you cannot audit is just a claim.
1. What was measured
- Verdicts: the same engine that serves this site, run for all 27 janma stars on every US trading day from 2 January 2019 to 19 October 2026.
- Returns: SPY and QQQ daily bars. Intraday is close divided by open, minus one. Next day is the following close over today's close.
- Volatility: daily range (high minus low, over the open) and absolute intraday move.
- Sample: 1,960 sessions × 27 stars = 52,920 star-days per index.
2. How to read the numbers
- bps (basis points) — one hundredth of one percent. 10 bps is 0.10%, roughly a tenth of a typical daily move. The comparisons below are in single bps, so they are tiny by construction.
- GO − AVOID — the average return on GO days minus the average return on AVOID days. Positive would mean GO days do better. Zero means the verdict does not separate them.
- CI 95% — the range the true value is likely to sit in. If that range straddles zero, the difference is not distinguishable from luck.
- P(>0) — the share of resamples where the difference came out positive. Near 0.50 means a coin flip.
- Correlation — whether a higher score tracks higher returns. 1.0 is perfect, 0.0 is none.
Clustering caveat: the 27 stars on one day all see the same market return. They are not 27 independent observations. Every interval here comes from a bootstrap that resamples days, not star-days. Pooling first would have shrunk the intervals by roughly the square root of 27 and manufactured confidence that does not exist.
3. Direction — a clean null
| Comparison | SPY | 95% CI (SPY) | QQQ | 95% CI (QQQ) |
|---|---|---|---|---|
| GO − AVOID, intraday | +0.16 bps | −4.5 … +5.0 | +0.24 bps | −5.5 … +6.0 |
| GO − CAUTION, intraday | −1.11 bps | −5.0 … +2.7 | −1.25 bps | −5.9 … +3.6 |
| GO − AVOID, next day | +2.88 bps | −3.1 … +9.1 | +2.21 bps | −5.0 … +9.9 |
| Score vs return correlation | 0.0001 | — | 0.0007 | — |
Every interval straddles zero. GO days are not better, not worse, and not more volatile in any way that survives clustering. Per star, the GO-minus-AVOID spread was positive for 15 of 27 stars on SPY and 16 of 27 on QQQ — a coin flip, with a spread of roughly ±1 to ±13 bps that is what noise looks like.
4. The one real effect — and it runs backwards
| Comparison | SPY | 95% CI | P(>0) | QQQ | 95% CI | P(>0) |
|---|---|---|---|---|---|---|
| AVOID − GO, daily range | −7.24 bps | −12.1 … −2.5 | 0.002 | −6.47 bps | −11.8 … −1.3 | 0.007 |
AVOID sessions are quieter, not wilder — by roughly six percent of a typical daily range, consistently on both indices. The effect is small but it is not noise. If the product keeps any volatility statement, this is the only direction the data supports, and it is the inverse of the everyday meaning of the word "avoid."
5. What this does not say
- The astronomy is not in question. Positions agree with JPL DE421 to 0.0004°, the Lahiri ayanamsa matches a Spica-derived value to 0.07°, and the 81-cell geometry matches the published matrix cell for cell.
- Classical Sarvatobhadra Chakra is not refuted. This tests one operationalisation of it, on two indices, over one period.
- It does not prove there is no edge. A few bps could hide in this sample. An edge the size the product once implied could not.
6. Reproduce it
The study is a script in the repository: scripts/outcome_study.py, with the method written up
in docs/outcome-study-2026.md. Run it against any ticker and window:
python scripts/outcome_study.py --ticker SPY --start 2019-01-01 --end 2026-10-20