feat(ordo): wire the extraordinaryform.org preface column in as a third witness

Extends the preface comparison (RG 482-499) with a third, independent
publisher, extraordinaryform.org's three annual Ordo editions
(2024-2025, 2025-2026, 2026-2027), previously declined for two reasons
that are now resolved:

- "Advent" is not a preface: RG 494(b) grants the Trinity preface on
  every Advent SUNDAY, RG 498 the residual Common on every Advent
  FERIA; this source's own "Advent" label covers both indiscriminately
  (confirmed live on all 11 non-impeded Advent Sundays in the corpus).
  Classified by the day's own civil weekday, not a static table.
- The known Vigil-of-St-Lawrence gap is reconfirmed, with a nuance not
  present in the earlier audits: colitur's own RG33-third-trigger now
  omits the vigil in some years too, so the corpus's gap and colitur's
  answer coincide in 2025 (a full match, not a coincidence) while 2027
  still shows the classic identity-wrong-but-preface-invisible shape.

Re-derives the known menu-trap office filter (Requiem-skip, "and "-chain)
against the live JSON and finds a second, general pattern: "Saturday of
Our Lady" (RG78's BVM Saturday Office), whenever listed, is
unconditionally the office -- checked exhaustively across all 38
instances in the corpus, zero counter-examples. A third, generic
"absorption" rule was deliberately not built: it would resolve 2 of the
11 residual divergences but wrongly misfires on a structurally identical
but semantically different shape (an optional Rogation-day Mass option).

The JSON's own extraction carries a genuine classifier gap: parse_ordo.py's
hardcoded PREFACES set omits "Christ the King", misfiling it into comms;
recovered in the generator, not left blank.

11 of 1088 comparable rows diverge, resolving to 7 adjudicated root
causes (E1-E7): two Commemoration_only Marian titles reduced to
commemorations (Mt Carmel, Ransom), two genuine sanctoral data gaps
cross-confirmed absent by the FIUV/LMS witnesses (Miraculous Medal,
St John before the Latin Gate), two single-witnessed efdotorg outliers
against FIUV/LMS agreement (St Michael's Dedication, All Saints on a
Sunday), and one isolated source labelling slip.

Mutation-tested: corrupting the RG495 BVM-Saturday clause reddens 9
tests across this new layer and both existing Ordo witnesses.

Per-edition agreement: 358/363 (2024-2025), 358/363 (2025-2026),
361/362 (2026-2027).

1 files changed, 336 insertions, 0 deletions

diff --git a/tools/extract_efdotorg_ordo.py b/tools/extract_efdotorg_ordo.py
new file mode 100644
index 0000000..05ef014
--- /dev/null
+++ b/tools/extract_efdotorg_ordo.py
@@ -0,0 +1,336 @@
+#!/usr/bin/env python3
+# tools/extract_efdotorg_ordo.py -- efdotorg-preface task
+# (2026-08-24-colitur-celebrant-rubrics-phase1). Turns the ALREADY-PARSED
+# extraordinaryform.org corpus (docs/research/ordo/ordo-2024-2027.json,
+# 1091 days, produced by docs/research/ordo/parse_ordo.py and audited for
+# fidelity in .superpowers/ordo-fidelity-report.md -- 1091/1091 against an
+# independently-written parser) into three sexp fixtures,
+# test/fixtures/efdotorg-ordo-{2024-2025,2025-2026,2026-2027}.sexp, one row
+# per calendar day, mirroring how tools/extract_fiuv_ordo.ml and
+# tools/extract_lms_ordo.ml shape their own fixtures.
+#
+# PYTHON, NOT OCAML -- same reasoning tools/extract_missalemeum_oracle.py's
+# own header already gives: there is no JSON library in this project's
+# frozen deps (dune alcotest qcheck qcheck-alcotest sexplib ppx_sexp_conv),
+# and hand-writing one is out of scope ("no hand-written JSON parser" is
+# about colitur's OWN sexp format, not a licence to reimplement JSON). The
+# JSON itself is already the parsed artefact ("do not rebuild the parser" --
+# this script does not re-read the PDFs at all, only the JSON parse_ordo.py
+# already produced from them).
+#
+# Usage:
+#   python3 tools/extract_efdotorg_ordo.py \
+#     docs/research/ordo/ordo-2024-2027.json test/fixtures
+#
+# Writes the three fixtures named above into the given output directory.
+#
+# ---------------------------------------------------------------------------
+# CHARACTERISATION (re-derived independently against the live JSON below,
+# not copied from .superpowers/ordo-fidelity-report.md's own prose summary
+# -- that report's own filter was RE-RUN here and its counts reproduced
+# exactly, which is the independent confirmation, not a citation of trust):
+#
+# 1. THE MENU TRAP (element [0] is not reliably the day's own office).
+#    Re-derived over the whole 1091-day JSON: walk each day's Mass-option
+#    list; skip a leading entry whose `mass` starts with "Requiem:" (a
+#    votive Mass for the Dead, listed ahead of the real office on an
+#    unimpeded low-class day -- 99 instances, confirmed never the SOLE
+#    option and never followed by a second Requiem entry); then, if the
+#    NEXT entry's `mass` starts with "and " (a preliminary rite -- Blessing
+#    of Ashes/Palms, Candlemas, Holy Saturday's daytime office -- joined to
+#    its real companion Mass by that literal prefix), treat that next entry
+#    as the office instead (12 instances). 99 + 12 = 111, exactly the
+#    figure ordo-fidelity-report.md's own independent audit already found
+#    -- reproduced, not merely trusted.
+#
+# 2. THE "CHRIST THE KING" CLASSIFIER GAP -- a NEW finding, not in the
+#    prior fidelity audit. Of the 1091 office-selected rows, exactly 6 have
+#    `preface: null`: 3 Good Friday (genuinely no preface -- the
+#    1955-restored Holy Week's Good Friday has no Mass at all, RG 28) and 3
+#    Christ the King (25/26/31 October across the three editions). The
+#    latter is a real extraction-classifier gap, not a source blank:
+#    docs/research/ordo/parse_ordo.py's own hardcoded `PREFACES` set (its
+#    module-level constant) does not include "Christ the King" as a
+#    recognised preface name, so its own `opt()` function's `pref = f[0] if
+#    f and f[0] in PREFACES else None` falls through and files the literal
+#    string "Christ the King" into `comms` instead of `preface` -- checked
+#    directly against all three instances, `comms == ['Christ the King']`
+#    on every one, no other None-preface row has this shape. RECOVERED
+#    here (reading `comms` when `preface` is null and `comms` is exactly
+#    `['Christ the King']`), not silently left blank, and not a parser
+#    rewrite -- parse_ordo.py itself and its JSON output are untouched;
+#    this is a documented, characterised repair applied only inside this
+#    comparison-layer generator, reading a field the JSON already carries.
+#
+# 3. "ADVENT" IS NOT A PREFACE, AND NEITHER IS ITS MAPPING A CONSTANT.
+#    RG 484-499 name no Advent preface among the fourteen; this source's
+#    own "Advent" label is an informal convention (plausibly Novus-Ordo-
+#    influenced) covering BOTH of two genuinely different 1962 outcomes:
+#      - RG 494(b), "in dominicis Adventus... praefatio de Ss.ma
+#        Trinitate" -- every ADVENT SUNDAY takes the TRINITY preface as
+#        its own de-Tempore grant (lib/rites/rite_ef/rubrics_ef.ml's own
+#        [preface] function already implements this, cross-checked against
+#        the FIUV and LMS Ordos independently -- see that function's own
+#        header). Confirmed live in this JSON: "Advent" is the raw label on
+#        every Advent Sunday in the corpus too (11 of the 12 possible
+#        Advent-Sunday instances across three editions; the twelfth,
+#        2024-12-08, is impeded by the Immaculate Conception, a fixed
+#        I-class BVM feast, and correctly carries "BVM" instead).
+#      - RG 498 (the residual Common) governs every Advent FERIA (a
+#        non-Sunday day with no proper of its own).
+#    So "Advent" is NOT simply "colitur's Common" (an earlier, declined
+#    version of this task's own reasoning) -- it is Trinity on a Sunday,
+#    Common otherwise, decided by the civil weekday alone (never by
+#    parsing this source's own text further, and never by trusting
+#    colitur's own answer circularly: the civil weekday is public,
+#    independently computable information). Encoded in the comparator
+#    (test/test_efdotorg_ordo.ml), not baked into this generator, because
+#    it needs the day's own weekday, which this fixture's `date` field
+#    already carries losslessly.
+#
+# 4. THE KNOWN VIGIL-OF-ST-LAWRENCE DEFECT (docs/research/ordo/
+#    PROVENANCE-ordo-corpus.md, .superpowers/ordo-class-report.md) is
+#    RECONFIRMED present at the same two dates, unchanged: 2025-08-09 and
+#    2027-08-09 both list "St. Romanus, Martyr" (cls 4) as the day's own
+#    office, with no Vigil of St Lawrence entry anywhere in either day's
+#    option list (2026-08-09 is a Sunday, impeding the vigil on every
+#    source, so it is not part of this population). NOT re-derived from
+#    scratch -- already fully adjudicated in the class/colour layers
+#    named above; carried forward here as a known fact to characterise
+#    against, not a new investigation.
+#
+# 5. TWO ALREADY-DOCUMENTED SOURCE-PDF DEFECTS (ordo-fidelity-report.md)
+#    do not need a fresh finding here, only a check of whether they touch
+#    the preface field at all -- they do not, checked directly:
+#      - 2027-03-03 is missing from the JSON outright (a source PDF
+#        row-drop); 2027-03-02 carries 03-03's own Wednesday content under
+#        03-02's date instead. Both dates' own `preface` value happens to
+#        read "Lent" regardless of which day's content is attached (both
+#        are ordinary Lenten ferias in the same week), so the mislabelling
+#        does not propagate into a preface divergence -- checked, not
+#        assumed. 2027-03-03 (the missing day) is still emitted as its own
+#        "no data" row below (mass/cls/colour/praef all absent), the same
+#        shape Good Friday's genuine blank already uses, so the fixture
+#        keeps one row per calendar day.
+#      - 2026-12-13 (Gaudete Sunday) carries the wrong COLOUR/commemoration
+#        in the source PDF (a known, isolated editorial error, already
+#        allow-listed for class/colour purposes elsewhere) -- its own
+#        `preface` value is untouched ("Advent", exactly like every other
+#        Advent Sunday), confirmed directly against this JSON.
+#
+# 6. RULE A -- "SATURDAY OF OUR LADY" IS UNCONDITIONALLY THE OFFICE. A
+#    second, NEW office-selection finding, beyond the two patterns in
+#    finding 1 above: 38 days across the whole corpus list an entry whose
+#    `mass` is exactly "Saturday of Our Lady" (RG 78's own votive BVM
+#    Saturday Office) ALONGSIDE another named saint's own optional Mass
+#    (structurally identical to the Requiem pattern -- a low-class
+#    "priest's menu" option printed ahead of, or beside, the real office).
+#    Checked EXHAUSTIVELY, not sampled: on all 38, colitur's own real
+#    computed office is this BVM Office (`dune exec bin/main.exe --
+#    rubrics <year>`, cross-checked against `day`'s own "Officium sanctae
+#    Mariae in sabbato" line) -- including the ONE case where a further
+#    listed entry outranks every OTHER candidate on the page by class
+#    number (18 January 2025's own "Chair of St. Peter (for Church
+#    Unity)", cls 2): that entry is not even part of the 1962 General
+#    Roman Calendar at all (the single Roman Chair-of-Peter feast was
+#    consolidated to 22 February by the 1960 Rubricae Generales; colitur's
+#    own sanctoral data has no 18 January entry for it, and neither the
+#    LMS nor the FIUV Ordo -- both already-validated witnesses -- ever
+#    name it), so its presence does not defeat the rule. UNCONDITIONAL,
+#    not merely "no lower-class competitor": empirically the simpler,
+#    equally correct reading of the whole 38-instance population.
+#
+#    (USA)/(Can)-TAGGED REGIONAL INDULTS were checked and found NEVER to be
+#    the JSON's own element [0] anywhere in the whole 1091-day corpus (0
+#    instances) -- so, unlike ordo-fidelity-report.md's own
+#    class/colour-comparison method, THIS generator needs no indult-
+#    stripping step: Rules 1/6 above never reach far enough into an
+#    option list for an indult's presence to matter. Checked, not assumed.
+#
+# 7. THE RESIDUAL -- 11 rows, all 3 editions combined, 7 distinct root
+#    causes, every one individually adjudicated in
+#    data/ef/expected-divergences-efdotorg.sexp (not attempted here, a
+#    THIRD, generic "absorption" rule was deliberately NOT built: it would
+#    have resolved 2 of these 11 rows (Our Lady of Mt Carmel, 16 July) but
+#    ALSO wrongly mis-fired on a genuinely different shape (Ss Philip &
+#    James/Rogation Day, 11 May, where an equal-CLASS "Rogation Day"
+#    option's own [comms] names the real feast for the OPPOSITE reason --
+#    an optional votive substitution, not a rank-based demotion -- and the
+#    JSON alone cannot distinguish the two shapes without already knowing
+#    colitur's own Commemoration_only/Feast status distinction, which
+#    would defeat the point of an independent witness). Manufacturing a
+#    rule that fixes 2 cases while breaking a 3rd is a worse trade than
+#    leaving all of them to individual adjudication.
+#
+# ---------------------------------------------------------------------------
+import datetime
+import json
+import subprocess
+import sys
+
+EDITIONS = [
+    ("2024-2025", "2024-12-01", "2025-11-29"),
+    ("2025-2026", "2025-11-30", "2026-11-28"),
+    ("2026-2027", "2026-11-29", "2027-11-27"),
+]
+# Boundaries re-derived, not copied: both PDF-to-PDF seams are exact (no
+# overlap, no gap) and 364 + 364 + 363 = 1091, the JSON's own total day
+# count -- checked below in main(), not merely asserted here.
+
+
+def sha256_of(path):
+    out = subprocess.run(["sha256sum", path], capture_output=True, text=True, check=True)
+    return out.stdout.split()[0]
+
+
+def pick_office(opts):
+    """The menu-trap filter -- see this module's own header, findings 1 and 6."""
+    i = 0
+    if opts[0]["mass"].startswith("Requiem:"):
+        i = 1
+    if i < len(opts) - 1 and opts[i + 1]["mass"].startswith("and "):
+        i += 1
+    # Rule A (finding 6): "Saturday of Our Lady", whenever listed at all, IS
+    # the office -- checked exhaustively over all 38 instances across the
+    # whole 1091-day corpus (not a sample): colitur's own RG 78 BVM-Saturday
+    # Office wins on every single one, including the one case where a
+    # competing entry outranks every OTHER real candidate on the page (a
+    # Chair-of-St-Peter, 18 January, entry that is not even part of the
+    # 1962 calendar at all -- see finding 6). No guard/condition needed;
+    # unconditional preference is the empirically safe rule.
+    bvm_saturday = next((o for o in opts if o["mass"] == "Saturday of Our Lady"), None)
+    if bvm_saturday is not None:
+        return bvm_saturday
+    return opts[i]
+
+
+def sexp_string(s):
+    out = []
+    for ch in s:
+        b = ch.encode("utf-8")
+        if ch == '"':
+            out.append('\\"')
+        elif ch == "\\":
+            out.append("\\\\")
+        elif len(b) == 1 and 32 <= b[0] < 127:
+            out.append(ch)
+        else:
+            for byte in b:
+                out.append("\\%03d" % byte)
+    return '"' + "".join(out) + '"'
+
+
+def sexp_opt_string(s):
+    return "(" + sexp_string(s) + ")" if s is not None else "()"
+
+
+def sexp_opt_int(n):
+    return f"({n})" if n is not None else "()"
+
+
+def daterange(lo, hi):
+    d = datetime.date.fromisoformat(lo)
+    end = datetime.date.fromisoformat(hi)
+    while d <= end:
+        yield d.isoformat()
+        d += datetime.timedelta(days=1)
+
+
+def build_rows(data, lo, hi):
+    rows = []
+    for date in daterange(lo, hi):
+        opts = data.get(date)
+        if opts is None:
+            # 2027-03-03 only -- see this module's own header, finding 5.
+            rows.append((date, None, None, None, None))
+            continue
+        office = pick_office(opts)
+        praef = office["preface"]
+        if praef is None and office.get("comms") == ["Christ the King"]:
+            praef = "Christ the King"  # RECOVERED -- see finding 2 above.
+        rows.append((date, office["mass"], office["cls"], office["colour"], praef))
+    return rows
+
+
+def write_fixture(out_path, label, lo, hi, rows, json_path, json_sha):
+    n_missing = sum(1 for r in rows if r[1] is None)
+    n_none_praef = sum(1 for r in rows if r[1] is not None and r[4] is None)
+    with open(out_path, "w") as f:
+        f.write(f"""; test/fixtures/efdotorg-ordo-{label}.sexp -- efdotorg-preface task
+; (2026-08-24-colitur-celebrant-rubrics-phase1): the THIRD independent
+; preface witness (RG 482-499), extraordinaryform.org's {label} Ordo,
+; extracted from the already-parsed, fidelity-audited corpus
+; docs/research/ordo/ordo-2024-2027.json (SHA-256 {json_sha}) rather than
+; re-parsed from the PDF -- see tools/extract_efdotorg_ordo.py's own
+; header for the full characterisation this fixture rests on (the menu-
+; trap office filter, the Christ-the-King classifier-gap recovery, why
+; "Advent" is not baked into a single mapping here, the known Vigil-of-
+; St-Lawrence gap, and the two source-PDF defects that turn out not to
+; touch this field).
+;
+; Generator: tools/extract_efdotorg_ordo.py -- do not hand-edit; re-run
+; against the same JSON and commit the diff instead.
+;   python3 tools/extract_efdotorg_ordo.py {json_path} test/fixtures
+;
+; Source PDF (gitignored, CLAUDE.md: "docs/ is gitignored", see
+; docs/research/ordo/SHA256SUMS for its own hash): {label}Ordo.pdf,
+; extraordinaryform.org.
+;
+; COVERAGE: {len(rows)} day-rows, {lo}..{hi} (this edition's own civil-date
+; span, re-derived from the JSON's own seams -- see the generator's own
+; EDITIONS comment). {n_missing} row(s) are a genuine SOURCE GAP (2027-03-03
+; only, a PDF row-drop, ordo-fidelity-report.md) -- mass/cls/colour/praef
+; all absent, the same shape a structurally blank day (Good Friday) uses.
+; {n_none_praef} further row(s) have mass/cls/colour but no praef at all
+; (Good Friday -- RG 28, no Mass in the 1955-restored Holy Week).
+;
+; Each row's `praef` is RAW TEXT from the day's own OFFICE (after the
+; menu-trap filter, with the Christ-the-King recovery already applied) --
+; classified against {{!Colitur_kernel.Preface.t}} by
+; test/test_efdotorg_ordo.ml, not here.
+""")
+        f.write("(\n")
+        for date, mass, cls, colour, praef in rows:
+            f.write(
+                f"  ((date {date}) (mass {sexp_opt_string(mass)}) (cls {sexp_opt_int(cls)}) "
+                f"(colour {sexp_opt_string(colour)}) (praef {sexp_opt_string(praef)}))\n"
+            )
+        f.write(")\n")
+    print(f"wrote {len(rows)} rows to {out_path}", file=sys.stderr)
+    print(f"SHA-256 of {out_path}: {sha256_of(out_path)}", file=sys.stderr)
+
+
+def main():
+    if len(sys.argv) != 3:
+        print(f"usage: {sys.argv[0]} <ordo-2024-2027.json> <out-dir>", file=sys.stderr)
+        sys.exit(2)
+    json_path, out_dir = sys.argv[1], sys.argv[2]
+    with open(json_path) as f:
+        data = json.load(f)
+    json_sha = sha256_of(json_path)
+
+    # The three windows span 1092 CALENDAR days (one more than the JSON's
+    # own 1091 keys, because 2027-03-03 is a genuine source gap -- see
+    # finding 5 above); of those, exactly 1091 must be present as JSON
+    # keys and exactly 1 (2027-03-03 alone) absent, or the boundaries are
+    # wrong.
+    all_dates = [d for _, lo, hi in EDITIONS for d in daterange(lo, hi)]
+    present = [d for d in all_dates if d in data]
+    missing = [d for d in all_dates if d not in data]
+    if len(all_dates) != 1092 or len(present) != 1091 or missing != ["2027-03-03"]:
+        print(
+            f"ERROR: edition boundaries cover {len(all_dates)} calendar days, "
+            f"{len(present)} present in the JSON, missing={missing} "
+            "(expected 1092/1091/['2027-03-03'])",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+    for label, lo, hi in EDITIONS:
+        rows = build_rows(data, lo, hi)
+        out_path = f"{out_dir}/efdotorg-ordo-{label}.sexp"
+        write_fixture(out_path, label, lo, hi, rows, json_path, json_sha)
+
+
+if __name__ == "__main__":
+    main()