Greek lectionary: Aug 28 is Slavic-only, oca.org cross-check, and the commit #220 dropped

[brianglass]

Consolidates #217, #222 and #223 into one branch. Those three are closed in
favour of this; nothing from any of them is lost.

Six commits, each independently meaningful:

commit	what it does	came from
Add the 2020 Antiochian ordo rows	re-lands a commit the #220 squash dropped	#223
Aug 28's vigil set is Slavic only	data fix + tests	#223
Cross-check the Greek Menaion rows against oca.org	docs only	#222
Salvage the Great Vespers source trap	docs only	#217
Keep ordo pks stable across loader runs	tooling only	new
Regenerate the fixture with main's ordo pks	fixture only	new

1. Re-lands what the #220 squash missed

main is currently missing the last commit of #220. Two things were lost:

• the 2020 Antiochian ordo rows — 2020-01-24 is a fifth date where the
  two jurisdictions genuinely differ (antiochian pdist=166 vs greek 147)
• the doc note recording that antiochian.org returns HTTP 500 for 2027
  because their chart is not published yet, not because the API can't reach
  that far. Without this the horizon looks like a hard limit and several
  Part III questions stay wrongly closed.

Worth knowing for next time: because #220 was squash-merged, git merge-base --is-ancestor reports every commit from that branch as absent from main,
so it cannot tell you what landed. This was found by comparing file contents.

2. Aug 28's vigil set is Slavic only

Aug 28 carried a full vigil package tagged common — three Vespers OT
lessons, a Matins Gospel, and a proper Epistle/Gospel — for Job of Pochaev,
a 17th-century Ukrainian saint kept in the Russian church. Greek was being
served a saint it does not commemorate.

#217 logged this and deliberately left it alone: the evidence then was
inconclusive, and this project does not change reading data on inference. It is
now settled.

source	Aug 28
antiochian.org	ordinary daily cycle in 8 of 8 years (2019–2026), always titled with the plain cycle label
goarch.org	agrees for 2026: 2 Cor 11:5-21 / Mark 4:1-9
oca.org	confirms the set is right for Slavic

The absence is evidence rather than merely uninformative, which is what the
earlier pass could not establish. That same feed demonstrably does surface a
saint's proper readings when one applies — it is how most of the Greek Menaion
data in the doc was found, and on Jul 5 this very Epistle appears for Athanasius
of Athos.

Six rows retagged common → slavic. Verified both directions:

2026-08-28 slavic: Wis 3.1-9; Wis 5.15-6.3; Wis 4.7-15; Matt 11.27-30;
                   2 Cor 11.5-21; Gal 5.22-6.2 (St Job); Mark 4.1-9;
                   Luke 6.17-23 (St Job)
2026-08-28 greek : 2 Cor 11.5-21; Mark 4.1-9

3. The ten Menaion rows are correctly greek

Scoped greek in #220 because no OCA source was to hand. oca.org turns out to
be plainly fetchable, so this is now checked rather than assumed: nine of ten
are correctly greek — oca.org gives either a different reading (Apr 25,
Apr 30, May 7, Jul 5's Gospel) or none at all (Jul 13, Aug 31, Sep 24, Dec 17).

The tenth is genuinely shared but still stays as two rows: both traditions read
Gal 5:22-26; 6:1-2 on Jul 5, but the attributions differ and both are right.
Slavic's says "either Saint" because OCA also commemorates the uncovering of
Sergius of Radonezh's relics that day; Greek's names Athanasius. Collapsing to
one common row would force one tradition's attribution on the other.

Incidentally: on all eight dates the app's Slavic output matches oca.org
exactly, Vespers and Matins included — the first direct check of the Slavic
data against its own source anywhere in the doc.

4. Salvaged from #217

#217's conclusion is superseded, but one finding in it is durable and was
recorded nowhere else — it is precisely what made the evidence look
inconclusive:

antiochian.org's Aug 28 page links a "Great Vespers" service text that is
actually Aug 29's Beheading of John the Baptist, served the evening
before and cross-linked from both dates.

Now a Part II source trap: a vigil-looking service linked from a page is not
evidence that that day carries a vigil. Read the citations, not the
attachments.

5. Loader no longer renumbers every ordo pk

Noticed while reviewing this branch's own diff: it touched 119 pk lines to
add one row. load_ordo.py emptied OrdoReading and rebuilt it on each run,
so every harvest renumbered the table and buried the actual change.

The up-front delete() is replaced by a prune of what the sources no longer
produce; update_or_create then keeps a surviving row's pk.

The pks were safe to move, which is worth stating explicitly: OrdoReading is
the target of no foreign key (the schema's only relation is
Reading.pericope), and both the runtime lookup and the tests address these
rows by their natural key (jurisdiction, year, month, day, source). This is a
legibility fix, not a correctness one.

Also corrects the docstring's dump command from --indent=1 to --indent=2.
The fixture is stored at 2 and load_shared_menaion.py already documented that;
dumping at 1 reformats all ~62,000 lines, which has happened once already.

Verified: load the fixture, re-run the loader, dump — empty diff, all 60
rows keeping their pks — and a row the sources no longer produce is still
removed.

6. …and the churn already committed is undone

Fixing the loader stops future churn but does not undo what was already
committed in #223, so the branch still renumbered every ordo row (main's
60-118 → 119-178). Rebuilt on top of main's own pks with the corrected
loader.

The fixture diff against main is now exactly the six Aug 28 retags and the
one added ordo row — 1 pk line touched instead of 119, and the file's
diffstat drops from 143 changed lines to 25. No data changed; 189 tests still
pass.

Verification

• 189 tests, OK (1 skipped) — same count as before, no tests dropped
• fixture diff contains exactly the intended changes and nothing else: 6
  common → slavic retags and 1 added ordoreading row, verified line by
  line; calendarium.reading row count unchanged at 1807
• both behaviors spot-checked live (above, and 2020-01-24 correctly renders
  both jurisdictions with labels)

Closes

• Document open question: does Aug 28's vigil reading set apply to Greek? #217 — question resolved; its durable finding salvaged above
• Cross-check the Greek Menaion rows against oca.org #222 — superseded, commit carried over verbatim
• Aug 28's vigil set is Slavic only; re-land a commit the #220 squash missed #223 — superseded, both commits carried over verbatim