# Verdict: PASS on conditions (a) and (b); (c) checked only at the endpoints (2026-11-09 00:15 UTC; see Corrections)

Pre-registered test (PREREG.md + Amendment 1), `scripts/killtest.py` → `killtest_result.json`.

- Overlap years 2001–2005, 2007–2017 (16).
- Earthshine (BBSO Earthshine.txt, annual mean of nights, ×0.30×340.2): **−0.35 W/m² per decade**.
- CERES EBAF-TOA Ed4.2.1 global reflected SW anomaly: **−0.60 W/m² per decade**.
- Same sign: yes. Ratio 0.57 (window 1/3–3): yes. **PASS.**
- Sensitivity: earthshine as mean of monthly means −0.27/decade → ratio 0.44, still passes.
- Secondary: annual correlation r = 0.62 (16 years); earthshine 2015–17 minus 1999–2001 = −0.52 W/m² (paper: ~0.5); CERES kept falling after the earthshine record ended: 2023 −1.81, 2025 −1.62 W/m² vs the 2005–2015 climatology; CERES slope 2001–2025 −0.81 W/m²/decade.
- Reconstruction check: our earthshine 1999–2017 change −0.52 W/m² matches the paper's "0.5".

## Secondary (Amendments 2–3): does the day's satellite picture predict the night's earthshine?
Weak. Primary rule (date D, sunset+2 h / sunrise−2 h): r = 0.12, one-sided p = 0.009, n = 422 (waxing r = 0.22, n = 175; waning r = 0.04, n = 247).
Sensitivity: D−1 r = 0.14 (waxing 0.28); D+1 r = 0.01; ±1 h r = 0.10–0.11. Model night-to-night spread ~9 % vs observed anomalies 1.4 %.
Reading: single nights carry little shared signal (as Pallé et al. 2003 found, "more muted"); the shared decline over the overlap years is the main result. The page says so and makes no nightly claim beyond this.
Run 1 (timestamps wrongly read as times) is void: r = −0.06.

## Corrections after review (2026-11-09)
- Run 1 (void) overall r was +0.016 (n in `nightly_result_run1_wrongtimes.json`); −0.06 was its waning subgroup. Both lines above that say −0.06 refer to that subgroup. Run 1 stays void.
- Condition (c), reconstructability to 0.1 W/m² per annual point, could only be checked at the endpoint: the reconstructed 1999–2001 → 2015–2017 change is −0.52 W/m² against the paper's "about 0.5". The paper gives no table of annual values, so per-point accuracy is not verified.
- The first-3 vs last-3 comparison uses 1999–2001 (1998 has 7 nights and is dropped by Amendment 1).
- Secondary 3 (share of the earthshine drop in 2015–2017): 48 % (1999–2001 → 2012–2014: −0.27; → 2015–2017: −0.52).
- CERES timing: 2001–05 → 2010–14 −0.20 W/m²; 2010–14 → 2021–25 −1.36 W/m² (87 % of the fall).