Moon-phase evidence case v1.0 — 2026-10-21
Focused secondary review, not a systematic review, field trial or meta-analysis.
Full primary main papers: Pulver 2017 (Gulf commercial) and Millar et al. 1997 (New Zealand recreational).
Stevenson & Millar 2013: author university abstract only; no full-text effect-size extraction. Potential record overlap prevents independent-replication counting.
Pulver supplements not independently acquired. No raw fishing-set or Louisiana redfish records obtained.

Factual input: evidence.json, manually transcribed from Pulver Tables 1–3, checked against rendered pages 538,540,545; target sets/captures/mean CPUE Table1, P and AIC change Table2. Pattern descriptions read from results/discussion, not digitized curves. AIC without lunar minus AIC with lunar. AIC is model fit, not percent catch gain. Full-model explained deviance is NOT variance explained by moon alone; omitted from our comparison chart for that reason.
Total observed sets = 7,520 + 37,247 = 44,767. Species rows overlap all-species rows; multiple targets may share a set. Never sum all rows or interpret 5/8 detected associations as the probability moon matters on a trip. No pooled test or multiple-testing correction supplied.

Original contrast arithmetic: log-linked periodic regression, new theta=0, full theta=pi, quarters theta=pi/2 or 3pi/2. Published coefficients are EXPONENTIATED (methods p539). For longline red grouper one retained cos(theta) term: new/full = exp(beta*(1-(-1))) = 1.073^2. Limits 1.050^2 and 1.097^2. For vertical red grouper one cos(2theta) term: quarter/new = exp(beta*(-1-1)) = .964^-2. Limits .988^-2 and .940^-2 (order reversed). Hold other covariates fixed; contrasts differ between gears. No covariance matrix required for these single-coefficient linear contrasts. No intervals for tilefish or total catch assembled from multiple terms without joint covariance. Conditional published model limits from rounded values do not add uncertainty for selection, unmeasured confounding or transfer to new fisheries. Not a raw-data refit or held-out forecast.

Rights: public outputs contain factual extraction, original site interpretation/calculation and original figure. Source papers, figures and university HTML retained internally for audit, not redistributed as public downloads. Attribution retained.
Reproduction: Node24, save evidence.json/reproduce.mjs together, run node reproduce.mjs. Recreates analysis.json and two CSVs offline. Repository figure command: MPLCONFIGDIR=/tmp/ldsf-matplotlib /tmp/ldsf-noaa-plots/bin/python site/scripts/plot-moon-evidence.py (Matplotlib3.11.2). Chart is AIC reduction, no uncertainty bars implied.
Review: independent fisheries/statistical review pending. Recheck on source correction, new relevant primary research, or before production publication. Do not infer redfish behavior, original trips, independent review or universal solunar scores from this review.
Blank field log is a user template only. Empty records are not observations. Separate segments for method/angler changes; preserve missingness and actual effort. No upload/collection service configured.