Accuracy: burn down the missalemeum known-differences baseline (421 rows) #428

New issue

Open

opened 2026-07-02 17:50:51 +00:00 by manlycucumber · 1 comment

manlycucumber commented

2026-07-02 17:50:51 +00:00

(Migrated from github.com)

The validation harness (#45 / #48) pins the engine ↔ missalemeum divergences as a
categorised baseline (tests/Validation/fixtures/missalemeum/known-differences.ndjson).
missalemeum is the primary base-1962 authority. Two systematic engine bugs it caught
are already fixed (#422 Triduum, #424 commemoration-grade offices); this tracks
reconciling the remaining 421 field-level differences, most of which cluster into
a few systematic causes. Fix each as its own cited PR and re-freeze the baseline
(php bin/freeze-oracle-baseline.php) so the diff records the improvement.

Categories (2024–2026 fixture)

corpus-rank (~82) — saints our sanctoral ranks IV/commemoration that
missalemeum ranks III class (e.g. Holy Name of Mary, St Lawrence Justinian, Ss
Cornelius & Cyprian, St Januarius). Correct the cited sanctoral ranks (data, not
engine). Split corpus-rank-underranked (71) / corpus-rank-overranked (11).
commem-sunday (78) — the engine commemorates on some Sundays where the
oracle shows none; adjudicate the 1960 Sunday commemoration rules.
commem-feast (85) / commem-feria (37) — commemoration-count differences on
feast/feria days; adjudicate co-occurring commemorations.
colour-christmas-epiphany-tide (23) — ferial colour between Epiphany and the
Baptism / in Christmastide (white vs green).
colour-penitential-second (6) — Palm Sunday / Good Friday two-colour
(red+violet, black+violet). Likely a contract colour-alternate representation.
colour-other (110) — remaining colour differences, many coupled to the
corpus-rank rows above (a mis-ranked saint drags colour + commemoration with it).

Notes

Where missalemeum and our reading of Rubricae 1960 genuinely differ, keep the row in
the baseline with a note rather than conforming blindly — the baseline is the record.
SSPX particular-calendar differences are out of scope here (they belong to the v0.2
overlay, tracked separately via the #49 SSPX oracle).

The validation harness (#45 / #48) pins the engine ↔ missalemeum divergences as a categorised baseline (`tests/Validation/fixtures/missalemeum/known-differences.ndjson`). missalemeum is the primary base-1962 authority. Two systematic engine bugs it caught are already fixed (#422 Triduum, #424 commemoration-grade offices); this tracks reconciling the remaining **421** field-level differences, most of which cluster into a few systematic causes. Fix each as its own cited PR and re-freeze the baseline (`php bin/freeze-oracle-baseline.php`) so the diff records the improvement. ### Categories (2024–2026 fixture) - [ ] **corpus-rank (~82)** — saints our sanctoral ranks IV/commemoration that missalemeum ranks III class (e.g. Holy Name of Mary, St Lawrence Justinian, Ss Cornelius & Cyprian, St Januarius). Correct the cited sanctoral ranks (data, not engine). Split `corpus-rank-underranked` (71) / `corpus-rank-overranked` (11). - [ ] **commem-sunday (78)** — the engine commemorates on some Sundays where the oracle shows none; adjudicate the 1960 Sunday commemoration rules. - [ ] **commem-feast (85) / commem-feria (37)** — commemoration-count differences on feast/feria days; adjudicate co-occurring commemorations. - [ ] **colour-christmas-epiphany-tide (23)** — ferial colour between Epiphany and the Baptism / in Christmastide (white vs green). - [ ] **colour-penitential-second (6)** — Palm Sunday / Good Friday two-colour (red+violet, black+violet). Likely a contract colour-alternate representation. - [ ] **colour-other (110)** — remaining colour differences, many coupled to the corpus-rank rows above (a mis-ranked saint drags colour + commemoration with it). ### Notes Where missalemeum and our reading of Rubricae 1960 genuinely differ, keep the row in the baseline with a note rather than conforming blindly — the baseline is the record. SSPX particular-calendar differences are out of scope here (they belong to the v0.2 overlay, tracked separately via the #49 SSPX oracle).

manlycucumber commented

2026-07-02 20:19:17 +00:00

(Migrated from github.com)

Burndown progress — the corpus-rank category is done

Three PRs merged, each verified (not conformed to an oracle blindly — a 6-agent
multi-source check against the 1962 General Roman Calendar / Rubricae 1960 backed
every change):

#436 — Advent temporal ranks: Sundays of Advent → I class; Advent ferias (to
Dec 16) → III class.
#437 — 14 sanctoral rank corrections (11 saints → III class, St Joachim + Seven
Sorrows → II class, St Hilarion → commemoration) + 4 co-commemorations the fix
surfaced (Euphemia/Lucy/Geminianus, Maurice, Thecla, Ursula).
#438 — the Commemoration of the Baptism of the Lord (Jan 13, II class, white).

Result

Oracle	Before	After	Resolved
missalemeum	421	303	−118
SSPX	85	23	−62

Zero lasting regressions; the golden-year digest and both oracle baselines are re-frozen.

Verification refuted two oracle rows (kept in the baseline as noted discrepancies)

St Didacus (Nov 13) and St Ubaldus (May 16) are genuine III-class feasts.
missalemeum alone under-ranks them (no SSPX witness); the engine is right, so these
stay as a noted corpus-rank-overranked discrepancy rather than being conformed away.
2026-08-09 (overranked) is a missalemeum role-inversion artifact (it surfaces the
Vigil of St Lawrence as the headline and demotes the ranking Sunday); DO + SSPX + the
engine all concur on the II-class Sunday. Kept, noted.

Spun out into their own issues (feature-sized / distinct bug)

#439 — September Ember days (II class) — missing computed temporal office.
#440 — Vigil of the Ascension (II class) — missing computed temporal office.
#441 — Code of Rubrics n. 33 (omit II/III-class vigils on Sundays/I-class days or
when their feast is reduced) — a precedence bug found during the burndown (explains
the Vigil-of-St-Lawrence divergences; the vigil data itself is correct).

Remaining baseline (303 rows) — the ongoing worklist

commem-sunday (74, likely one systematic 1960 Sunday-commemoration rule), commem-feast
(83) / commem-feria (19), colour-christmas-epiphany-tide (18, the Epiphany-tide ferial
colour white-vs-green), colour-other (84, many coupled to the above), colour-penitential-second
(6, the Palm Sunday / Good Friday two-colour convention → likely a contract representation),
and the ~structural rows tracked by #439/#440/#441.

## Burndown progress — the corpus-rank category is done Three PRs merged, each verified (not conformed to an oracle blindly — a 6-agent multi-source check against the 1962 General Roman Calendar / Rubricae 1960 backed every change): - **#436** — Advent temporal ranks: Sundays of Advent → I class; Advent ferias (to Dec 16) → III class. - **#437** — 14 sanctoral rank corrections (11 saints → III class, St Joachim + Seven Sorrows → II class, St Hilarion → commemoration) **+ 4 co-commemorations** the fix surfaced (Euphemia/Lucy/Geminianus, Maurice, Thecla, Ursula). - **#438** — the Commemoration of the Baptism of the Lord (Jan 13, II class, white). ### Result | Oracle | Before | After | Resolved | |---|---|---|---| | missalemeum | 421 | **303** | **−118** | | SSPX | 85 | **23** | **−62** | Zero lasting regressions; the golden-year digest and both oracle baselines are re-frozen. ### Verification refuted two oracle rows (kept in the baseline as noted discrepancies) - **St Didacus (Nov 13)** and **St Ubaldus (May 16)** are genuine III-class feasts. `missalemeum` alone under-ranks them (no SSPX witness); the engine is right, so these stay as a noted `corpus-rank-overranked` discrepancy rather than being conformed away. - **2026-08-09** (`overranked`) is a missalemeum role-inversion artifact (it surfaces the Vigil of St Lawrence as the headline and demotes the ranking Sunday); DO + SSPX + the engine all concur on the II-class Sunday. Kept, noted. ### Spun out into their own issues (feature-sized / distinct bug) - #439 — September Ember days (II class) — missing computed temporal office. - #440 — Vigil of the Ascension (II class) — missing computed temporal office. - #441 — Code of Rubrics **n. 33** (omit II/III-class vigils on Sundays/I-class days or when their feast is reduced) — a precedence **bug** found during the burndown (explains the Vigil-of-St-Lawrence divergences; the vigil data itself is correct). ### Remaining baseline (303 rows) — the ongoing worklist `commem-sunday` (74, likely one systematic 1960 Sunday-commemoration rule), `commem-feast` (83) / `commem-feria` (19), `colour-christmas-epiphany-tide` (18, the Epiphany-tide ferial colour white-vs-green), `colour-other` (84, many coupled to the above), `colour-penitential-second` (6, the Palm Sunday / Good Friday two-colour convention → likely a contract representation), and the ~structural rows tracked by #439/#440/#441.