The rules of the frame, each on one short passage, opened by the reader itself. For a year these were drawings: a page that imitated the reader's card with typed Hebrew inside it, which demonstrates a drawing. Each of these is a real zone, built by the builder every served book goes through and opened by the same reader with the same card, and branded an instrument so that no gate can mistake it for a work.
The C0 never moves. Whatever a reader selects, the ink stands exactly as the source wrote it. Selection governs what is defined, never what is displayed.
0absenceM · no witness, and the page says so rather than filling it incarried
The text never depends on the catalog. A word with no reading is still the word, drawn exactly as the source wrote it; what is missing is the English, and the page is required to say which of several different nothings this is.
Most of what can go wrong here is a reader looking at nothing and not knowing why. There are several different nothings and they are not the same: a form the catalog has never carried, a verse the edition itself prints empty, a reading held back because its license forbids display. The page draws none of them as a blank box, because a blank box claims there is a reading and then fails to show it. The line stays quiet and the card says the reason.
Carried from the served store and the corpus lane's restore, in the source's own vocabulary: a form absent from the zone's gloss table; EMPTY_VERSE as the edition's own mark. the counts are computed over the thirty-nine files this edition ships, not recalled.
no_cardthe line under the word stays empty and takes no border — an empty box would claim a reading; the card carries the sentence1623 sites in 39 books
amos 3:11— two of the fourteen words in this verse have no reading; the card says “no reading in the catalog for this form” daniel 1:4— in the Hebrew opening of a book that turns to Aramaic amos 9:6
empty_versethe verse keeps its number and its address, and carries the mark instead of words2 sites in 1 books
joshua 21:36— printed empty by the edition itself — the absence is the source's, not ours joshua 21:37
None of these is a mark a reader may switch off. A missing reading is a fact about the catalog and an empty verse is a fact about the edition; hiding either would be the page deciding what a reader is allowed to notice.
0the language underneathM · a source admits Hebrew and Aramaic, and says whichcarried
The text is what it is. A catalog answers about a FORM, not about a sentence — so a form that is a name in this verse and a common noun elsewhere gets the reading the catalog carries, and the order is the reader's.
Two verses show the whole problem. At Daniel 2:4 the text changes language mid-verse and says so in its own words — three words of Hebrew, then the word Aramaic, then Aramaic to the end of the chapter. At Genesis 31:47 Laban and Jacob name the same heap in two languages, and the catalog answers the Aramaic from Hebrew. Nothing here is a fault in the readings: each is attested and licensed, and each is answering about the form rather than the sentence. It is the clearest argument on the site for the order being the reader's.
Carried from the served store, in the source's own vocabulary: the readings as the catalog carries them. what each card says here was read off the built zone, not recalled.
mid_verse_switchno boundary is drawn: the words simply continue, and the text names the language itself1 sites in 1 books
daniel 2:4— the fourth word is the word Aramaic, and the language changes on it
two_languages_in_a_verseevery word carries whatever reading the catalog holds for its form1 sites in 1 books
genesis 31:47— the opening verb reads as the name of a book, the name Laban reads as a common noun, and the Aramaic reads from Hebrew — three different ways for a form to outrun its sentence
Nothing here is hidden or corrected. The site does not overrule a licensed reading; it lets a reader put the pool in another order.
1qere and ketivC0 · one for the sitecarried
One occurrence, one C0. The complex is the word; the breakdown happens a layer down.
Where the reading differed from the letters the scribes had received, they changed nothing: the written form was kept and the read form recorded beside it. The site is one word of the book, counted once, and its ink carries both forms in the source's own order.
The passage is tanakh/nahum tanakh-nahum-kq-1, printed as moses-ledgers/kq-reseal-v1/streams/tanakh-nahum-kq-resealed-v1.tokens.csv.gz wrote it.
1the rare shapes of qere and ketivC0 · one occurrence, more than one way incarried
A pair is one site of the text with more than one record behind it. Choosing a branch changes what the card is about and moves no character of the line.
Most pairs have two branches and both are written. These are the ones that are not. Sometimes only one side exists at all — the scribes wrote something they did not read, or read something they did not write — and the card still has to explain both halves of a pair one of which is absent. Sometimes one written word answers to two read words, or the reverse, so the branches are not the same length. And where a pair sits inside a maqaf, rule 1 and rule 2 both fire on one occurrence.
Carried from the corpus lane's restore, in the source's own vocabulary: the kq, ketiv and qere columns. the shapes were counted over all thirty-nine files, not sampled.
one_sidedthe one branch that exists, with the card naming which side is missing17 sites in 6 books
pair_with_a_letter_markrule 1 and rule 12 on one word1 sites in 1 books
job 7:5— the only place in the corpus where a pair also carries a marked letter
A pair is the text, not a note about it.
2maqafC0 · one per wordcarried
Two words, two C0s, two cards. Nothing is grouped over them: the count follows the word, not the typography.
A maqaf joins words into one breath and one accent. Each word it joins is its own word of the book with its own card, and the joiner rides as ink on the word before it, which is the word that gave up its accent.
The passage is tanakh/ruth ruth-1-16, printed as moses-ledgers/maqaf-reseal-v1/streams/tanakh-ruth-maqaf-split-v1.tokens.csv.gz wrote it.
2a maqaf pair the catalog answers unevenlyC0 · two occurrences that read as one breathcarried
A maqaf occurrence is two C0s, each counted, each with its own card. Nothing about the join changes what either piece is.
Because the pieces are separate, the catalog can answer for one and not the other — and it does, at a hundred and seven places. The run then prints one reading line with a gap in it, which is honest: the piece with no reading has none, and the page does not borrow its neighbour's.
Carried from the served store against the corpus lane's restore, in the source's own vocabulary: a key absent from the zone's gloss table. 42,325 pairs have a card on both pieces; these are the rest.
one_piece_cardedthe run's line carries the reading that exists and leaves the other empty107 sites in 20 books
The gap is the catalog's, and hiding it would make the page claim a reading it does not have.
3abbreviationC0 · one, oncarried
One C0 for the abbreviated form, always on. The words it stands for get no C0 at all.
An abbreviation is one token standing for several words. All of them are read, but their letters are not in the ink, so the token is the word of the book and the words it stands for are a grouping of their own. The mark between the letters is part of the key, whichever character the source wrote it with.
The passage is chasidut/maor-vashemesh maor-vashemesh-bo-bo-15, printed as the verified July body, shard cleanroom-c0-ledger-map-rebuilt-active-000000000000000003740849-000000000000000003840848.csv.gz wrote it.
One C0, off by default, keyless. It marks structure, not a word.
A closed section mark is the scribes' bookkeeping: it says a section ends here and the next begins on the same line. It is not a word of the book and it opens no reading, so it is off until a reader turns the marks on.
The passage is tanakh/exodus exodus-20-13, printed as cleanroom-c0-ledger-map-rebuilt-active-000000000000000069794244-000000000000000069808717.csv.gz wrote it.
One C0, off by default, keyless. Its own kind, beside the closed section.
An open section mark says a section ends here and the next begins on a new line. Like the closed section it is the scribes' bookkeeping rather than a word of the book, and it is off until a reader turns the marks on.
The passage is tanakh/genesis genesis-35-22, printed as cleanroom-c0-ledger-map-rebuilt-active-000000000000000069828900-000000000000000069846706.csv.gz wrote it.
6inverted nunC0 · one per word, every one cardedcarried, mark typed
Every word inside gets its own C0 and its own card. The brackets are pressable and are never toggleable.
Two inverted letters stand around a short passage, and the tradition reads them as a statement about the passage itself rather than as bookkeeping. So the words inside are ordinary words with ordinary cards, the brackets can be pressed to say what they are, and no toggle may hide them: they speak about the text.
The passage is tanakh/numbers numbers-10-35+numbers-10-36, printed as cleanroom-c0-ledger-map-rebuilt-active-000000000000000070166543-000000000000000070180865.csv.gz wrote it. The mark itself is typed, and the page says so on the word it is typed on: the two inverted nunim are absent from every source this project holds: the probe byte-scanned all 4,646 shards of the verified body and every reseal stream and found no U+05C6. The words between them are the body's own ink; the brackets are typed so the rule can be shown at all.
Included, with its own C0, behind the toggle the section marks share.
The scribes counted, and wrote their counting down: the middle of a book by verses or by words, and the tallies at its end. Those notes are text about the book rather than the book, so they are carried, they are pressable, and they sit behind the same toggle as the section marks.
The passage is tanakh/haktav-vehakabalah tanakh-haktav-vehakabalah--haktav-vehakabalah-postscript, printed as cleanroom-c0-ledger-map-rebuilt-active-000000000000000149689492-000000000000000149718405.csv.gz wrote it.
No C0 of its own. The words are ordinary words; what is recorded is that a run of them is set apart.
A commentary opens by quoting the words it is about, and the printers set that quotation apart. Stripping the setting apart as though it were transport loses a distinction the source drew, so it is recorded as an overlay over the run before anything is stripped: the ink stays what it was and the run says what it is.
The passage is midrash/matnot-kehunah-on-ruth-rabbah matnot-kehunah-on-ruth-rabbah-petichta-petichta-1-6, printed as the verified July body wrote it.
10 · the brick gap of a song · approved 2026-09-11
10the brick gap of a songC0 · a gap is its own row, and it is not a wordcarried
The scribes set the songs differently from prose — brick over brick, half-brick over half-brick — and the space is part of how the passage was written. It gets its own record and is counted as a mark, never as a word.
In the songs the layout carries meaning, so the gaps are transmitted rather than styled. Each is a row of its own, so the words on either side keep their own addresses and the setting can be drawn or dropped without moving a letter of the text.
Carried from the corpus lane's restore, in the source's own vocabulary: the gap column of the restore rows. counted over the thirty-nine files this edition ships.
brick_gapa gap glyph standing in the line, taking no key and no card328 sites in 8 books
The gap speaks about the text rather than about the page, so it stands with rule 6 rather than with the section marks.
11 · paseq and sof pasuq · approved 2026-09-11
11paseq and sof pasuqC0 · each is its own INKOFF rowcarried
Neither is a word and neither carries a key. Each is a record of what the scribes put between or after words, counted on the written axis and never on the read one.
The sof pasuq closes a verse and the paseq stands upright between two words the scribes wished to keep apart. They are the most common marks in the corpus by a wide margin, which is exactly why they are worth a demonstration: a rule that fires twenty-three thousand times is one nobody looks at.
Carried from the corpus lane's restore, in the source's own vocabulary: the paseq and sof_pasuq columns. every mark is its own row; none is counted as a word.
paseqthe upright stroke, standing in the line as its own row2293 sites in 37 books
These two the marks switch does hide — they are bookkeeping about the page. The switch names them: 1,717 things, verse ends and paseqs among them.
12 · letter marks · approved 2026-09-11
12letter marksC0 · an overlay on a letter, not a word of the bookcarried
The mark is a pointer, never a character. It names a letter by its position and says how the scribes wrote it; it supplies no glyph, changes no surface, and leaves the word's key exactly what it would be without it.
In four places the scribes wrote a letter differently from its neighbours — larger, smaller, with a dot above it, or lifted above the line — and every copy since has kept the difference. The record says which letter, counted alef through tav with the points not counted, and what was done to it. The page draws the letter with every point that rides on it, so a mark and its vowel move together instead of the vowel being left behind at normal size beside an enlarged consonant.
Carried from the corpus lane's restore, scribal_detail column, in the source's own vocabulary: OTIYOT_GEDOLOT, OTIYOT_KETANOT, OTIYOT_MENUKADOT, OTIYOT_TELUYOT — the source's own terms, mapped one to one. tools/serve-from-restore-v5.mjs refuses the whole build on a scribal token it cannot name (LETTER_MARK_UNREAD), so a mark this project does not understand stops the line rather than being dropped.
largefont-size 1.45em, line-height 0.927 sites in 17 books
leviticus 11:42— where rule 12 meets rule 2 — the source marks the fifth letter counting through the maqaf, and the split re-bases it onto the third letter of the second piece deuteronomy 6:4— the opening word of the Shema, its last letter written large daniel 6:20— a large and a small letter standing in one verse
The source names a letter by counting through a maqaf, and rule 2 cuts that occurrence into two C0s. So the index must be re-based onto the piece that holds the letter, or the mark lands on the wrong glyph. Leviticus 11:42 is the case: the fifth letter of the joined run is the third letter of the second piece.
The marks switch hides 1,717 things — verse ends, paseqs, section marks — and no letter mark is among them. Measured: pressing it leaves the letter drawn. The same reasoning as rule 6: bookkeeping about the page may be hidden, something written in the ink may not.