docs: two retracted claims that outlived their retraction

Both are the defect class this fix round existed to close, which is why
they are worth recording rather than quietly overwriting.

CLAUDE.md still carried F1's retracted design rationale -- describing
entry_14_movable_band as "negative so it can never collide with a real
table position" -- and F2's retracted transcription-defect attribution,
verbatim, in the one tracked file the project's own workflow tells a
fresh session to read first. The fix round struck both everywhere else
and missed this file. A session following it as instructed could have
re-derived the negative sentinel for entry 20's or entry 23's movable
half and re-introduced the inversion in a new branch.

precedence_ef.ml's replacement justification for the x10 rescale then
introduced a fresh misstatement of the same primary text it was
correcting: it attributed "primum mobilia, deinde fixa" to "entry 12's
own ninth sub-item" and counted three such rows. Both photographic
scans, and the register's own section 4 table, put that clause at entry
13; entry 12 has eight sub-items and no such clause. The split occurs at
five rows -- 13, 14, 20, 23's third sub-item, and 24. Entry 24 is the
most-evaluated band in the whole domain and is unsplit in the code,
which is precisely the sub-rank the new headroom exists to hold.

Corrected in the .mli and the register too. No behaviour change; 290
tests green.

1 files changed, 25 insertions, 10 deletions

diff --git a/CLAUDE.md b/CLAUDE.md
index faec251..54ae709 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -339,17 +339,32 @@ nowhere in git (`docs/` is gitignored).
    window can never be empty of a Sunday (unlike RG 17(a)'s Holy Name,
    which carries an explicit calendarium fallback for its own narrower
    window) — checked, not assumed; no fallback built. `Precedence_ef.band`
-   gained a movable-half priority for entry 14 (`entry_14_movable_band`,
-   negative so it can never collide with a real table position) — without
-   it, Holy Family would tie with the fixed Commemoration of the Baptism of
-   the Lord (13 January) at the literal integer 14 and lose the kernel's own
-   alphabetical tie-break, backwards from RG 91's own stated order.
+   gained a movable-half priority for entry 14 — without it, Holy Family
+   would tie with the fixed Commemoration of the Baptism of the Lord
+   (13 January) and lose the kernel's alphabetical tie-break, backwards
+   from RG 91's own stated order. **The whole table is now scaled ×10**
+   (entry *n* → 10*n*), so a half-row is expressed as an ordinary position
+   between its neighbours. *(CORRECTED: this paragraph previously described
+   `entry_14_movable_band` as "negative so it can never collide with a real
+   table position". That was the bug, not the design — a negative sentinel
+   avoids **collision** but also inverts **ordering**, making a movable
+   II-class feast of the Lord outrank every I-class day. Unreachable on
+   universal data; live the moment a diocesan overlay puts a I-class proper
+   or indult feast, RG 91 entries 12–13, in the 7–13 January window. Do not
+   re-derive the sentinel approach.)*
    `disposition` gained RG 112(a) (a mystery of one Divine Person excludes a
-   commemoration of another mystery of the SAME Divine Person) — directly
-   confirmed by a MORE SPECIFIC primary source found while building this,
-   the Holy Family Mass propers' own 13-January rubric (both photographic
-   scans, absent from the electronic transcription along with RG 112(a)'s
-   own worked example — the transcription-defect pattern found again, §6.-1).
+   commemoration of another mystery of the SAME Divine Person). The primary
+   authority is **RG 95 ¶2** — *"Si vero duo festa eiusdem Divinæ Personæ…
+   fit de festo, quod in tabella præcedentiæ superiorem obtinet locum, et
+   aliud omittitur"* — an occurrence-level rule present in all three
+   documents; RG 112(a) and the Holy Family Mass propers' own 13-January
+   rubric corroborate it. *(CORRECTED: this previously called the propers'
+   rubric a further instance of the transcription defect. It is not. The
+   transcription carries RG 112 in full; RG 112 has no worked example in
+   either scan; and the propers' note is absent because that document is a
+   partial 2006 web capture containing almost no propers text — one
+   `Introitus` in 26 322 lines against 52 in a scan. Diagnose the cause of a
+   silence before invoking the rule.)*
    **Blast radius, measured (`git archive` pre-change binary vs HEAD, full
    1583–9999 sweep, diffed): 1 220 days, every single one the identical
    shape, cross-verified against `date -d` independently (exactly 1 220