Pre-Registration — Solar Phase → Civilizational Crisis Clustering

Registered: 2026-08-16, before any analysis was run. Why this document exists: the existing solar/crash material (HISTORICAL_CRASHES_UET_SYNTHESIS_2025-11-03.md) is not a result. It cites three events, with inconsistent lags, under a rule permissive enough that both solar maxima and minima appear to precede crises. A rule that cannot fail is not a finding. This registers the test in advance so the parameters cannot be tuned after seeing the answer.

Status: RUN 2026-09-17 — H1 not supported. Result: SOLAR_PHASE_CRISIS_TEST_RESULT_2026-09-17.md. Nothing below was revised. Revisions require a new dated document and both remain public.

---

1. The hypothesis, stated one direction only

H1: Civilizational crisis events cluster in the declining phase of the solar cycle — specifically in the window beginning at solar maximum and ending at the following minimum — at a rate above chance.

The prior work's fatal flaw was claiming both maxima and minima. We commit to declining-phase only. If crises cluster at minima or rise-phase instead, H1 is refuted, not amended.

H0: Crisis events are distributed across solar cycle phase no differently than chance, given the uneven length of phases.

2. Solar data — fixed source

• Series: SILSO Version 2 International Sunspot Number, monthly smoothed, Royal Observatory of Belgium.
• Cycle boundaries: SILSO's own published minima and maxima dates. We do not re-date cycles ourselves — that would be a free parameter.
• Range: Solar Cycle 1 (1755) through the most recent completed cycle. Cycles with incomplete or contested dating are excluded and listed, not silently dropped.

3. Crisis data — externally compiled, not ours

This is the load-bearing design choice. We do not assemble the crisis list. Using our own list would let us select events that fit.

• Primary: Reinhart & Rogoff banking-crisis and sovereign-default chronology (This Time Is Different dataset), which covers the full period and was compiled with no reference to solar activity.
• Secondary, for the US subperiod: NBER business cycle contractions (1854–present).
• Both are used as published. No additions, no exclusions, no re-dating.

If the two sources disagree on an event, we report the test under each separately rather than merging.

Solar phase → crisis clustering — RESULT of the preregistered test

Registration: PREREG_SOLAR_PHASE_CRISIS_TEST_2026-08-16.md (16 Aug 2026), unchanged. First run: 2026-09-17. Script: scripts/research/prereg_solar_phase_crisis_test.py (re-runs from the four public sources in one command; raw output in prereg_solar_phase_crisis_test.out.md). Verdict under §6 of the registration: H1 is NOT supported. One crisis source shows a modest declining-phase excess, the other shows none, and the excess does not survive the conservative count.

---

What was run, exactly

• Solar phases: SILSO's published minima and maxima, cycles 1–24 (cycle 25 excluded as incomplete, as the registration requires). Declining = [maximum, next minimum). Declining phases are 60.8 % of the time in scope, so under H0 about 61 % of crises fall in them by construction.
• Crises, as published, no edits: Reinhart & Rogoff banking-crisis and sovereign-external-default chronology (HBS Global Crises panel, 70 countries, 1800–2016; onset = first year a flag turns on) and, for the US, NBER contraction onsets (peaks, 1857–2007).
• Two choices the registration left open, fixed in code before the first run: onsets are dated to mid-year (the panel is annual); each source's time-at-risk opens where its coverage opens (1800 / Dec 1854).
• Two counts, both reported: every country-onset as published, and distinct onset-years (one per year with ≥ 1 onset), because default and banking waves (1931, 1982, 2008) are not independent draws.
• Tests: one-sided binomial P(X ≥ observed) against the phase-length expectation, threshold p < 0.05; two-sided alongside; the declared secondary window (maximum + 0–24 months) alongside, never substituted.

Result

Test	n	in declining	expected	ratio