The Zodiacal Releasing Peak Signal Is Contrarian. The Data Said So Twice.

Update — 2026-04-18: After expanding the TIMED peak cohort from N=134 to N=189 (55 new Astrotheme-sourced Oscar/Nobel/sports winners 2002-2024), the ZR contrarian signal weakened from p=0.023 to p=0.093 — no longer statistically significant at the conventional 0.05 threshold. The new subjects fired ZR-peak at the null rate, diluting the contrarian pattern. The honest reading is below the original post.

In Chris Brennan's Hellenistic Astrology: The Study of Fate and Fortune (2017) — probably the most influential English-language revival of Hellenistic technique in a generation — Zodiacal Releasing is positioned as the technique for timing life culminations. Specifically, L1 peak periods — when the active time-lord sign is angular to the Lot of Fortune (in houses 1, 4, 7, or 10 from Fortune) — are described as the windows where the major chapter of a life tends to arrive.

It's central. If you practice or study Hellenistic astrology, you've almost certainly been taught this.

We tested it. Twice. In both the original 200-permutation run and the sharper 10,000-permutation re-run we completed this week, the signal came back negative and statistically significant.

The test

Our cohort is 266 documented life-peak events — lottery wins, Olympic golds, Nobel announcements, career-peak moments — where we have either an exact birth time (AA/A Rodden) or a noon chart. We asked: does the event date fall inside a ZR L1 peak period as Brennan defines it?

For each event we also computed the null rate — what fraction of events fall inside ZR L1 peak periods when you shuffle birth data against event dates 10,000 times. The null rate estimates how often this configuration fires in any random pairing of charts and dates.

The results:

• Observed rate (real pairings): 21.6% of peaks fire during ZR L1 peak
• Null rate (shuffled pairings): 29.9%
• Cohen's h: −0.19 (small-to-medium effect, negative direction)
• Two-sided p = 0.023

Peaks fire in ZR peak periods less often than would be expected by chance. That direction, that effect size, replicated stably from the 200-perm to the 10,000-perm run.

Before you call it noise

Three standard escape hatches, addressed in order.

"You're testing the wrong version of ZR." We computed L1 peak windows from date of birth, with peak defined as the L1 sign being in Fortune's 1st/4th/7th/10th, plus the Loosing-of-the-Bond sub-peak windows. This is Brennan's specification. If a different specification is the "right" one, it needs to be written down and tested.

"Your cohort is biased." The cohort is Astro-Databank's A/AA Rodden-rated entries for peak-event categories plus obituary-derived high points. If there's a selection effect that systematically puts lottery winners and Olympic medalists outside their ZR peaks, we don't see how to generate one.

"Peaks mean something different than Brennan meant." This is the most charitable reading. Brennan's ZR peak might be tracking intensity of life experience, not outcome valence. A ZR L1 peak window could produce above-baseline rates of both major peaks AND major setbacks, which in a peak-only test looks like a negative signal because the null rate of the combined peak-or-setback distribution is elevated.

To check this, we also ran the test on setbacks. Setback rate: 25.0%. Still below null (29.9%), though closer. That's weak evidence for the intensity-not-valence reading — peaks and setbacks both track under null during ZR L1 peak periods. But if ZR peak were neutral on valence, you'd expect the two rates to bracket the null, not both sit below it.

The cleanest summary: Brennan's ZR peak doctrine, as commonly stated, does not predict peaks in our data. It predicts fewer of them.

What we do with this

Our production forecast model previously gave ZR peak a small positive weight (+0.06) as a theory prior. We've flipped the sign to −0.40, scaled by the observed Cohen's h. The model now down-weights days during ZR L1 peak — producing a lower luck score — which, if the signal is real, should make our predictions match actual outcomes better.

We will be wrong if this is an artifact of our specific cohort or dating conventions. That's what replication is for. A held-out cohort — entrepreneurial exits, sports finals wins, first-book bestsellers — is the next test. If the contrarian signal replicates there, we have a real finding that ZR peak is either mis-interpreted or misdirected in the current literature.

Why this matters for the field

Null and negative results don't usually get published in astrology. Books are written by people making affirmative claims; null results are rarely career-making for a practitioner.

But the entire point of a scientific-style astrology research program is to subject inherited claims to tests that can kill them. This is one that did. H1 ZR peak isn't the only classical claim we've put to the test this year — we've got a list of twenty more that came back null or weak, and a handful that survived. Publishing the nulls matters as much as publishing the wins.

If this study replicates, the next step is engaging with the tradition to figure out what ZR is actually measuring, because whatever it's measuring, it isn't what the books say.

The full dataset and replication code are in services/astro-api/scripts/expanded_analysis_fast.py.

---

2026-04-18 update: the expanded-cohort test

The post above was written against a cohort of 134 TIMED peak subjects. A day later we added 55 new TIMED peaks via ScraperAPI+Astrotheme — Oscar Best Director/Actor/Actress 2002-2024, Nobel laureates with published birth times, Messi's World Cup, recent Grand Slam winners. Re-ran at 10,000 permutations on the expanded N=189:

• Observed ZR-peak rate at peaks: 27.0% (was 21.6% at N=134)
• Null rate: 32.1% (was 29.9%)
• Cohen's h: −0.111 (was −0.189)
• Two-sided p: 0.093 (was 0.023)

The direction is still negative. The magnitude shrank. And p=0.093 doesn't cross the conventional 0.05 threshold. As a single-test claim, the ZR contrarian finding no longer clears the bar.

What probably happened: the 55 new subjects are predominantly modern creative / sports peaks — Oscar ceremony dates, Grand Slam final dates — where the "peak event" date may not align with ZR's activation logic the way lottery wins and obituary-documented peaks did. When we extend the test population to modern ceremony peaks, the ZR-peak rate converges to null.

What we're doing about it in the product: we will reduce H1's scorer weight from −0.40 toward ≈−0.20 at the next retrain rather than zero it out. The direction replicated twice — it shrank, but it hasn't flipped. Consistent with a real-but-small contrarian effect that needs more TIMED peak data to resolve.

The honest lesson: single-test p-values at p ≈ 0.02 on a 26- hypothesis family do not always survive cohort expansion. The wise move is to publish the original finding, publish the expansion, and let the reader see both. Astrology research's replication problem is real; the fix is to report the direction and report when it weakens.