Gauquelin g5 Historical-Sector Validation — 2026-07-20

Decision

The supplied g5 temporary-sector archive is admitted as an external, non-bundled validation witness, not as runtime truth and not as an algorithm oracle. Moira's current Gauquelin algorithm is unchanged.

The clean numerical tranches are CFEPP and the explicit-LMT subset of Müller. CFEPP has UTC date/time and coordinates; Müller LMT converts to UTC directly from longitude. CSICOP is retained as a two-policy sensitivity audit because its g5 conversion path has ambiguous 12-hour-clock semantics. Ertel and the 167 non-LMT Müller rows remain inventoried but not yet admitted for numerical comparison.

Source identity and custody

• User-supplied archive: g5-tmp-g-sectors.zip
• Bytes: 243939
• SHA-256: 889B27999D787574F9CE0771BEB8BA41AC91D1B1362D6A03D12442022821B0BC
• g5 repository inspected at commit bf0db345b58127a438121b74ebf4ad843243a573
• g5's source code is GPL-3.0. No g5 source code is copied into Moira.
• Written correspondence supplied with the archive states that CC-BY-SA is acceptable for g5 and Open Gauquelin attribution. Because the archive contains personal records and an exact redistribution attribution line has not been fixed, the CSVs remain outside this MIT repository.

The committed fixture contains only aggregate counts and the source hash. It does not contain names, dates, places, or coordinates.

Archive inventory

The archive therefore contains 6,995 source rows, not the 6,648 implied by the counts in the accompanying email. These are not 6,995 independent people or experiments: Ertel includes records from Gauquelin and skeptic-test groups, including CFEPP and CSICOP.

The g5 source definitions resolve the CFEPP discrepancy. Rows 1–1066 are the official 1996 CFEPP group; rows 1067–1120 are 54 supplementary Nienhuys cases. The g5 Ertel definition and conversion code both identify 4,384 rows. The email's “4091” is retained as an unresolved likely count typo, not used as a filter. Ertel's MA12 equals floor((MARS - 1) / 3) + 1 for all 4,384 rows.

The archive also contains a LibreOffice lock file. The harness ignores it and never treats it as evidence.

The archive member names and the four sanctioned row counts are fail-closed invariants. A same-named archive with added or missing rows is rejected rather than reported with stale canonical corpus sizes.

CFEPP computation

For every CFEPP row, the harness:

1. parses UNIV_DATE plus UT as an aware UTC instant;
2. negates g5's west-positive LONG into Moira's east-positive longitude;
3. computes apparent topocentric Mars RA/declination and local sidereal time through the same chart service used by the public Gauquelin route;
4. computes Moira's canonical geometric-horizon 36-sector position; and
5. maps it to 12 sectors with floor((sector36 - 1) / 3) + 1.

No source sector is supplied to the computation, and no matching rule can change Moira's output.

Results

All 13 non-exact results differ by exactly one adjacent 12-sector bin. Their Moira positions lie between 0.009740° and 0.534203° of the nearest 12-sector boundary. This localizes the disagreement to boundary-sensitive cases. It is consistent with differences in source time precision, ephemeris, reference frame, or historical calculation method, but this pass does not adjudicate which cause applies to any record.

The result corroborates the current numbering, longitude convention, and diurnal-sector implementation. It does not justify changing the algorithm to force 100% agreement with a source whose maintainer explicitly warns that some original sectors may be erroneous.

Müller explicit-LMT result

For 916 rows, MODE=LMT explicitly identifies local mean solar time. The harness parses directional longitude and applies UTC = LMT - east-positive-longitude / 15 hours. It also normalizes source times written as 24.xx into the following civil day. The 167 rows with a blank mode are excluded; their KORR codes are not interpreted.

The five larger circular differences are 7, 8, 13, 14, and 16 of 36 sectors. They remain source-adjudication cases rather than inputs to an algorithm change. Of the seven adjacent differences, the aggregate report retains continuous boundary distance without exposing personal records.

CSICOP sensitivity result

The 2019 g5 raw2tmp.php converter constructs local time by adding 12 hours for P and P1 but leaving A and A1 unchanged. Applied literally, that turns 12 P into 24:xx and leaves 12 A at 12:xx. The archive has 32 records at hour 12, so the harness reports both the literal g5 rule and conventional 12-hour-clock semantics. The special g5 ZEITZONE=0,5 rule is preserved as UTC-10:30 under both policies.

The conventional policy removes every half-wheel discrepancy caused by the hour-12 reversal. It is still a sensitivity result, not an admitted authority choice: confirmation from the data maintainer or an original CSICOP codebook is needed before selecting it. The remaining two outliers differ by 8 and 10 of the 36 sectors and should be checked against the original records. The other 80 non-exact conventional results differ by one or two narrow 10-degree diurnal bins, which can reflect clock/zone precision as well as calculation method. No tolerance is used to relabel them exact.

Reproduction

With the external archive present and a DE441 kernel installed:

$env:MOIRA_AUTO_DOWNLOAD = '0'
$g5Archive = 'C:\path\to\g5-tmp-g-sectors.zip'
.\.venv\Scripts\python.exe scripts\validate_g5_gauquelin_sectors.py `
  $g5Archive `
  --output .\tmp\gauquelin-g5-full.json

The committed aggregate result is tests/fixtures/gauquelin_g5_aggregate_2026_07_20.json.

Deferred tranches

• CSICOP: obtain an authority decision for hour-12 AM/PM semantics and inspect the two large outliers against the original records/codebook.
• Müller: adjudicate the 167 non-LMT correction codes and inspect the five large explicit-LMT outliers against the original booklet/list.
• Ertel: join its source identifiers to records that carry birth time and place. Many Ertel rows omit time and the raw sector file has no coordinates.

Each deferred tranche needs a separate, tested adapter. No guessed conversion may be admitted merely to improve agreement.