The commit that reordered the chain did not carry the version bump. The
bump was chained after check.sh with &&, the commit after a semicolon, so
a failing check skipped the bump and let the commit through. The tag then
pointed at a tree still calling itself 0.3.0, which the release workflow
asserts against and would have refused.
check.sh was failing because a test names which register is asked first
about an ISBN, which is exactly what changed. It now names the index, for
the reason the reorder happened.
The chain asked three registers and returned whichever answered first
with a title and an author, discarding the rest. The registers are
complementary rather than redundant: OpenLibrary's edition record
carries the publisher, the pagination and the language and names its
authors by key, while its search index carries the authors and not the
publisher. One book, two half-answers, and the whole of it only if they
are put together.
They are merged now, first answer winning each field, so the order of
the chain is the order of trust. The record says which registers spoke.
It also carries what was being thrown away: subjects, language and a
page count, from whichever register has them. The index will not return
those unless asked for by name, which is how a language it held went
unread. Subjects are capped at twelve — the index lists every heading
every edition of a work has ever been given, which on a well-known book
is hundreds.
An edition record keeps its subtitle in a separate field, so a title
came back without it.
Read out of the review queue a real run produced, where each of these was
sitting as a proposed title waiting for a person to say no:
1. General Remarks a numbered heading
2 James W. McAllister a numbered heading
5222 What Is This a folio that ran into the line
827 North Fourth Street an address
Page xxxvi / Chapter I a marker
title a form label, printed literally
Sidebottom, Harry. Ancient a bibliography entry
To Mackenzie, a dedication
The numbers are the delicate part. A leading number that could be a year
is left alone, because 1984 Revisited is a book and 5222 What Is This is
a folio, and nothing else distinguishes them. A dedication has to end in
a comma, so To Kill a Mockingbird survives.
The copyright notices in the same pile are already fixed: those
proposals were written before 0.2.2 and are stale rather than wrong.
title_case counted small words from the start of the whole string, so
every subtitle that opened with one was set in lower case: 'Aesthetic in
Kant: a Critique', 'Born Fighting: the Scots-Irish'. A colon starts a
title of its own and the first word of one is capitalised however small
it is.
Found by book-pipeline, which had been proposing a title page's capitals
verbatim into a catalogue and needed this to stop.
'(c) Alfredo Ferrarin 2004' was read as a title. The notice pattern
wanted the year to follow the symbol directly, and a copyright line puts
the holder's name in between, which is the ordinary way to set one.
The Python surface gained enough this week that three different builds
were installed on two machines all calling themselves 0.1.0: moria had
read_colophon through merge_colophon, mirkwood had a partial from the
middle of the afternoon, and PyPI had neither. A version that does not
move is a version nobody can depend on, which is how book-pipeline ended
up running a wheel scp'd into place by hand.
New since 0.1.0: colophon --shelf and catena.shelve, which file a book's
printed Library of Congress headings onto a closed list; catena
.read_colophon; catena.fetch_identifier, the registers in process; and
catena.colophon_prompt / merge_colophon, the question about what the
pages did not settle and the reply folded back in.
A catalogue's worst records are the ones whose file says nothing true about
itself: a PDF titled `Untitled` by an author called `Oxford`, which is a
publisher, from a filename that was never anything but a filename. What
those books do have is pages. The title page carries the title as the
publisher set it; the copyright page carries the imprint, the year, the
ISBN, and in most English books since the 1970s the cataloguing block, which
is a catalogue record a librarian already made, subject headings and all.
So this reads the pages. Every field names the page it came from, and a book
that printed nothing comes back empty with its fields in `unsettled` rather
than filled with something plausible — an extraction that starts at chapter
one has no front matter in it, and saying so is the answer.
Two pages look like a title page and are not, and both are how a catalogue
comes to name the wrong book. A series advertisement lists other titles in
the shape a title page uses, over as many as four pages with only the first
announcing itself. A translation's cataloguing block prints the original
title in brackets above the real one. Both are found and refused.
Three things the pages give that no file does: the subject headings a
librarian assigned, an ISBN that exists only in print — which turns a book
with no identifier into one `bibfetch` can look up — and a title in the case
the publisher meant, because a half title prints unbroken what the title
page breaks across two lines and sets in capitals.
`--prompt` and `--merge` are citefind's: a question about exactly the fields
the pages left unsettled, composed here and answered wherever the caller
keeps a model, with the advertising taken out of what the model is shown.
A field a model answered is marked `model`, so a record always says which of
it was read and which was inferred.
Measured against 600 books of the home corpus: the cataloguing block
answered for 21% of them and agreed with the catalogue 89% of the time, most
of the residue being the catalogue's error rather than this one's. Title
pages answered for another 41%. The remaining 38% print nothing readable,
and are what the prompt is for.
The harness compared these tools against the code they replaced — Python
imported from book-pipeline, podcast-pipeline and knowledge-mcp, and
JavaScript copied out of the podcast media server and indexer, each pinned to
a commit. All of it has since been deleted, so the live half cannot run and
the golden half pins answers against a source nobody can consult.
What was worth keeping was never the comparison. `KNOWN.md` argued, at
length, why a dozen of these tools deliberately do something a reader would
not expect: why pdfgutter measures the crop box throughout, why avplan finds
the encoder rather than reading one out of the environment, why mdscrub will
not manufacture a page boundary. Those arguments are about this code and
stand on their own, so they are now `docs/decisions.md`, rewritten to say
what each tool does and why rather than what it does differently.
Binding coverage is unaffected: the bindings carry 24 tests of their own, a
node smoke test against the built package, and a typecheck.
Cuts a reference list into entries — at the note numbers where there are any,
at the layout where there are not — and reads each as far as its punctuation
allows. Fields that would not parse are named in `unsettled` rather than
guessed at, and `--prompt` asks a model about exactly those and no others.
Over 202 real reference sections from the corpus: 8,566 references, 95% with
a title, 82% with a year, 46% settling every field with no model involved.
A title is refused where the entry was never divided, so no record reports
its own text back as the work it cites.
`ibid.` is flagged and left alone — it points at a work, and the entry does
not say which. A rule of dashes is filled in, because the style defines it as
the author above.
chunkmd gains `note_starts` and `cites_something`: recognising a page of
notes and splitting one are the same knowledge, and two copies of that
pattern would drift.
funkpower