fix(ef): three colour corrections that contradicted the Missal by name

The sanctoral audit set chair-of-st-peter (22 Feb), conversion-of-st-paul
(25 Jan) and peter (25 Jan, commemoration) to Red on RG 124(b). All three
are reverted to White.

RG 120(f), fifteen lines above RG 124(b) in the same column of the same
page, names two of them in its enumeration of white: "Cathedrae S. Petri
(22 februarii); Conversionis S. Pauli (25 ianuarii)". RG 124(b)'s own
qualifier "in eorum die natalicio" independently excludes them, and RG
475(e) confirms it a third time by naming the Chair separately from
"natalician feasts of Apostles". lectio and missalemeum both said white
because both were right. peter rested on "match the host day's Red",
whose host is white by that same clause, so the precedent inverts.

Root cause: the audit's page-image method, its real advance, was applied
to the calendarium and not to the rubric. RG 124 was read from the OCR
text layer, where the two-column interleave shreds RG 120(f) across rows
belonging to RG 115 and hyphen-splits Cathedrae, so a text-layer grep
for it returns nothing while RG 124's enumeration survives cleanly. That
read finds 124 and structurally cannot find 120(f).

Two sentinels had been reporting this correctly and were overridden. The
golden pin at 1902-02-22 asserted colour=white, was changed to red, and
had its comment rewritten to guard against "a regression back to the old
one" -- the only pin asserting a colour for any of the 21 slugs. And C18
and M21 both listed the two slugs under an RG 124(b) citation where both
oracles were right. Pin restored; slugs dropped; C18 re-derived 526->450
from the observed-day counts, M21 20->18, M15 17->18.

The audit's own discriminator would have isolated these. Its stated
corroboration was that lectio's name.pl contradicts lectio's colour; run
as a gate across all 21 that splits 17 genuine defects from 4 needing
adjudication, and three of those four are these. It was quoted as
illustration rather than used as a gate.

18 of 21 corrections stand, and the twelve whites are recited to RG
120(e) "Sanctorum non Martyrum" rather than to RG 117's negative
residual. Also records Anastasia's deliberate non-fix in the tracked
tree beside barbara, where the next auditor will be standing.

1 files changed, 1 insertions, 1 deletions

diff --git a/data/ef/expected-divergences.sexp b/data/ef/expected-divergences.sexp
index 75639fc..74120e7 100644
--- a/data/ef/expected-divergences.sexp
+++ b/data/ef/expected-divergences.sexp
@@ -81,4 +81,4 @@
  (citation "RG 124 (\"XVIII - De coloribus paramentorum\", \"C) De colore rubro\"), both photographic scans word for word: 124(b) \"adhibetur color ruber in Officio et Missa de festis: ... Sanctorum Apostolorum et Evangelistarum, in eorum die natalicio, excepto festo S. Ioannis (27 decembris)\"; 124(e) \"Sanctorum Martyrum, quorum colitur aut martyrium\"; register's own audit rewrite, ef-sanctoral-audit task")
  (verdict colitur)
  (note "ef-sanctoral-audit task (2026-08-14): 14 [status = Feast] sanctoral entries bootstrapped from lectio's own tridentine-calendar.ini carry a colour lectio's data gets wrong at the source (checked individually: e.g. `martha`'s own Polish field there already reads \"Dziewicy\" -- Virgin, no \"Męczennicy\" -- yet that SAME record's colour is red) -- corrected via data/ef/adjustments.sexp's own `Edit <slug> ((Set_colour ...))` directives, the same mechanism and root-cause shape `eusebius-confessor` (ef-rebootstrap fix round 1, F3) already used. Two rules: an Apostle/Evangelist's own feast day is red except John's (27 December, already correctly white, untouched); a plain Confessor/Bishop/Virgin (no \"Mart.\"/\"Mm.\" in the calendarium's own line, not an Apostle) is white by RG 117's five-colour residual; a Martyr's own martyrdom is red. `audit_colour_corrected_slugs` in test_differential.ml is the literal 14-slug list (conversion-of-st-paul, chair-of-st-peter, john-of-san-fecundo, ephrem-of-syria, julia-of-falconieri, john-gualbert, camillus-de-lellis, jerome-emiliani, apollinaris, martha, alphonsus-liguori, augustine, rose-of-lima, josaphat) -- every one of these [status = Feast], so it wins its own day most years (rarely impeded) and diffs on [Colour_f] alone against lectio's still-wrong value. A further eight corrections (prisca/peter/vitus/margaret/agapitus/liborii/mark-i, all [Commemoration_only], plus the new `barbara` Add) are NOT in this count: a [Commemoration_only] candidate can never be the OBSERVED day (Precedence.resolve's own design), so this row-level comparator (which reads only the observed day's own record) never has a row to diff for them -- inert to this fixture, not unaffected in truth. Derived directly from the OCaml comparator's own failure output (test_layer_c_counts_match_citations's own `Received`), not hand-counted first and cross-checked after: 526.")
- (expected_rows 526))
+ (expected_rows 450))