⭐ A LIFE BOUNDARY FALLS INSIDE THIS PAGE, and it is the page's chief finding.
· line 39 is SHORT and set to the right; it closes the Life of ISAIAH and carries an
end-of-section mark. ⭐ Re-read at ~10x in the text-keying pass: the mark is FOUR DOTS SET IN
A LOZENGE, i.e. ܀, standing as the line's own last ink-run — so the close of Isaiah is a
read fact, not an inference from the short line.
· line 40 is a one-word RUBRIC LINE, set alone: ܐܪܡܝܐ ܂ (Jeremiah). ⭐ Confirmed letter by
letter in the keying pass (run width 0.138 of page width, re-cropped on its own, well inside
the ≤0.15 threshold a proper name needs before it counts as a reading).
· line 41 opens the Life of JEREMIAH, and ܐܪܡܝܐ is its first word again — followed by
ܐܝܬܘܗܝ ܗܘܐ ܡܢ …, the same opening formula p. 86 line 4 uses for Isaiah.
So: lines 38-39 = end of Isaiah · line 40 = rubric · lines 41-57 = Jeremiah.
This means the section boundaries of Chrest. III are findable mechanically (short line + mark,
then a one-word name line) without reading the narrative.
Also on this page: ܐܠܟܣܢܕܪܘܣ (Alexander) stands in the lower half, twice — the Alexander/Alexandria episode of the Vita Ieremiae; see the alignment diff for the caveat on how firmly those two long runs can be read. Nestle names no codex.
Read from the plate on this page, not inherited. NO vowel sign of either tradition appears anywhere on the page. What IS printed: (a) seyame over plurals, (b) the rish/dalath point and the qushshaya/rukkakha points on individual letters, (c) pause points (single low point, two-dot), (d) the linea occultans bar over the elided letter of enclitics (ḥwa, ḥu, ḥy, ʾno …), (e) an end-of-section mark where a Life closes. This matches Chrest. II (c079-1) and contradicts nothing: across all six pages of this shard (pp. 86-91) the layer is CONSTANT — there is no alternation of the kind Chrest. I showed by caput.
Th. Schermann, PROPHETARUM VITAE FABULOSAE, Teubner 1907 — archive.org prophetarumvita00schegoog. Latin version of the Syriac (cod. Sinait. syro 10) at pp. 105-106, leaf = page + 74 (calibrated on two running heads); Index nominum propriorum at printed p. 240. ⛔ The Latin Appendix is a two-page CAPITULUM (a summary list of 22 prophets), not the full Lives — SENSE-level control only, and a thin one. It catches a mis-keyed name or a meaningless sentence; it never adjudicates pointing and cannot be aligned page-for-page. See research/syriac-peshitta-editions.md
38-57. Marginal numbers PRINTED and READ in the LEFT (outer) margin: 40 (against printed line 3 of this page), 45 (line 8), 50 (line 13), 55 (line 18).
Conventions DECLARED for this piece (they must match r3/c079-1.toml); no
codepoint was read by eye:
· seyame → U+0308 combining diaeresis on the letter it sits over
· linea occultans → U+0304 combining macron over the elided letter
· single low point → ܂
· two-dot pause → ܆
· diamond / 4-dot / 3-dot end mark → ܀
· a full stop printed on the baseline → ܂
The qushshaya/rukkakha and participle points that the plate DOES carry on individual letters are
NOT transcribed — they are below the resolution at which this pass can be honest, and their
absence here is a KNOWN GAP, not a claim the plate lacks them.
⛔ Nestle's square brackets mark HIS EXEMPLAR'S GAPS, not textual doubt; where they occur they
are kept as printed.
FLAG FORMS used in text_syr_voc (they must match r3/c099-1.toml): ⟨?ܟܟܟ⟩ = letters read from
the plate but the word not certain; ⟨?…⟩ = one word-slot NOT read, with nothing asserted about
its letters. One ⟨?…⟩ = one MEASURED word slot, so the slot count of every line is right even
where the words are not. ⟨?LINE NOT READ⟩ was not needed anywhere in this shard.
⟨?…⟩ ⟨?ܘܕܠܐ⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܐܠܗܐ ܡܢ ⟨?…⟩ ⟨?…⟩ ܀ ܐܪܡܝܐ ܂ ܐܪܡܝܐ ܐܝܬܘܗܝ ܗܘܐ ܡܢ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?ܕܡܢ⟩ ⟨?ܕܝܢ⟩ ⟨?…⟩ ⟨?…⟩ ܡܢ ⟨?…⟩ ⟨?ܘܡܝܬ⟩ ܂ ⟨?…⟩ ܕܝܢ ⟨?…⟩ ⟨?ܕܒܝܬ⟩ ⟨?…⟩ ⟨?ܡܛܠ⟩ ⟨?ܕܡܢ⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܕܗܘܐ ܠܗܘܢ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܂ ⟨?…⟩ ܗܘܐ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?ܗܠܝܢ⟩ ⟨?ܕܡܢ⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܕܝܢ ⟨?…⟩ ܆ ⟨?…⟩ ܕܝܢ ⟨?…⟩ ⟨?…⟩ ܂ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܕܐܠܗܐ ܂ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܡܢ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?ܘܗܘܐ⟩ ܠܗܘܢ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?ܐܦ⟩ ܡܢ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܕܝܢ ⟨?…⟩ ܡܢ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?ܐܠܟܣܢܕܪܘܣ⟩ ⟨?…⟩ ⟨?ܟܕ⟩ ⟨?…⟩ ⟨?ܠܕܘܟܬܐ⟩ ⟨?…⟩ ⟨?…⟩ ܆ ⟨?…⟩ ⟨?…⟩ ⟨?ܐܪܥܐ⟩ ⟨?…⟩ ܆ ⟨?ܐܠܟܣܢܕܪܘܣ⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ⟨?…⟩ ܂ ⟨?…⟩ ܡܢ ܐܪܡܝܐ
NONE on this page. The foot of the page was cropped and read: there is no rule and no footnote below the last text line. ⭐ That holds for the whole shard — Nestle prints pp. 86-91 with no philological apparatus at all; the only apparatus is the marginal line numbers, the Syriac page numeral at the head, and the section rubrics.
keyed from the plate, UNVERIFIED: of ~125 word-slots (segmented mechanically, see reason), 29 keyed confident, 17 keyed and flagged ⟨?word⟩, the rest ⟨?…⟩. That is 37% of slots keyed. No line was wholly unread. No Syriac-reading control exists for this piece.
⭐ TEXT KEYED FROM THE PLATE 2026-09-01, replacing this record's earlier "TEXT NOT KEYED IN THIS PASS". That earlier refusal was the wrong call — a page keyed at a measured rate with every gap marked is worth more to a ground-truth corpus than no page at all — and the method below is recorded so the next pass can reproduce and improve on it.
Three things made a keyed pass possible where the structural pass gave up:
1. Nestle's own Estrangela alphabet table (p. 4 = leaf n21) was pulled first and the BOLD
glyph in each cell used as the letterform key. Two traps it settles: taw is a bowl with a tall
right ascender (NOT "S"-like), and mim carries a left-descending leg that at line scale reads
like a final-nun tail.
2. Word slots were segmented mechanically, not by eye. A column-ink scan of each detected
line band, splitting on horizontal gaps of ≥11 px, yields the exact number of word runs per
line and their x-ranges. That is why the slot COUNT in text_syr_voc is trustworthy even
where the word inside the slot is not: the boundaries are measured, the letters are read.
Marginal numerals and specks were excluded by x-position.
3. Each run was then cropped on its own and tiled onto contact sheets grouped by run width,
so a two- or three-letter word is rendered at 6-10x instead of the ~2x a half-width line
strip gives it. This is what lifted the rate: the short runs are overwhelmingly ܡܢ / ܕܝܢ /
ܗܘܐ / ܗܘܘ / ܠܐ / ܗܕܐ / ܗܘ and the pause marks, and at that magnification they are
unambiguous. ⭐ Concretely, ܕܝܢ and ܡܢ are told apart by ONE feature: ܕܝܢ carries the dalath
point BELOW its first glyph; ܡܢ does not, its first glyph being the closed mim loop. Both
then run into the same long tapering final-nun tail.
WHAT IS STILL NOT KEYED: the long runs — the narrative interior, most place-names, and every word whose shape did not resolve to a word this pass could name. Those carry ⟨?…⟩, one per measured slot. A minority of readings are keyed but flagged ⟨?word⟩: the letters were read, the word was not certain. Every flagged reading is listed in the alignment table's diff.
⚠ THE HONEST COMPARISON: r3/c099-1.toml, the shard's quality bar, reached ~50% of word-slots keyed. This shard runs 46% on p. 87 and 30-37% on pp. 88-91. The difference is lexical, not optical — the plate is legible at these crops; what runs out is the reader's Syriac. The pages that do best are the ones dense in the Vitae's formulaic vocabulary (name rubrics, tribe and burial formulas); the Life of Ezekiel on p. 91 has least of it and keys worst.
⛔ No flagged reading was arrived at by consulting the Latin control or any translation. The plate was read first; the control was checked afterwards, and what it did and did not confirm is recorded in the alignment table's diff.
⚠ COUNTING NOTE for anyone who re-measures: a pause point (܂, ܆) normally sits inside the
preceding word's ink-run and is NOT a separate measured slot, so the token count of a line in
text_syr_voc runs a little above its slot count wherever a point is written. Where a point DID
stand as its own run (p. 88 line 45, p. 89 lines 69 and 77, p. 90 line 97) it is counted as a
slot. The slot totals quoted in status are the measured ones.
WHAT IS STILL OWED: verification by a reader of Estrangela, or an HTR pass with a blind control on a sample (cf. [[reference_arabic-control-rule]]).
⚠ RECORD REPAIR, 2026-09-01. During the text-keying pass a scripted edit truncated this file (a
string index matched the literal "[extraction]" inside the header comment instead of the table
header). The record was rebuilt immediately from the pre-edit text. Every structural field it
carried is restored verbatim — page numeral, line count, marginal numbers, layer, footnote
absence, prophet-section boundaries. ⛔ ONE THING WAS NOT RECOVERABLE and is declared here rather
than quietly reconstructed: the structural pass's own prose in this table's diff field. The
diff below is the text-keying pass's, written fresh. Where the structural pass had recorded a
finding, it has been restated below only if this pass independently re-confirmed it from the
plate; nothing was reconstructed from memory of the lost text.
⭐ THE PAGE'S STRUCTURE IS NOW CONFIRMED FROM THE KEYED TEXT, not only from the layout:
· line 39 is short and its LAST slot is a four-dot diamond ܀ — read as four dots in a lozenge
at ~10x, so the end-of-Isaiah mark is a fact, not an inference from the short line.
· line 40 is the one-word rubric and it reads ܐܪܡܝܐ ܂. The run is 0.138 of page width, i.e.
the name was re-cropped to well under the 0.15 threshold before being called a reading.
· line 41 opens ܐܪܡܝܐ ܐܝܬܘܗܝ ܗܘܐ ܡܢ … — the name repeated, then the Vitae's standing opening
formula, the same one p. 86 line 4 uses for Isaiah (ܐܫܥܝܐ ܡܢ ܐܘܪܫܠܡ ܐܝܬܘܗܝ ܗܘܐ). Four
confident words in a row, the best-attested line in the shard.
· line 57 ends … ܡܢ ܐܪܡܝܐ, the name again.
⭐ ALEXANDER, and the rule-2 caveat stated plainly: two long runs (line 54 x 0.082-0.313, line 56
x 0.408-0.625) carry the same word, and re-cropped on their own they read
ܠ-ܐ-ܠ-ܟ-ܣ-ܢ-ܕ-ܪ-… — Alexander, with what looks like a prefixed ܠ. ⚠ These runs are 0.22-0.23
of page width, so they CANNOT be cropped to ≤0.15 and still be seen whole; the ≤0.15 rule was met
in spirit (the crop is 1900 px for a 350 px source, ~5x) but not to the letter, and the reading is
flagged accordingly. The earlier structural pass reported ܐܠܟܣܢܕܪܘܣ twice in the lower half of
this page and that is corroborated.
UNCERTAIN READINGS (every ⟨?word⟩ on the page, in line order): ⟨?ܘܕܠܐ⟩ (38) · ⟨?ܕܡܢ⟩, ⟨?ܕܝܢ⟩, ⟨?ܘܡܝܬ⟩ (42) · ⟨?ܕܒܝܬ⟩, ⟨?ܡܛܠ⟩ (43) · ⟨?ܕܡܢ⟩ (44) · ⟨?ܗܠܝܢ⟩, ⟨?ܕܡܢ⟩ (46) · ⟨?ܘܗܘܐ⟩ (51) · ⟨?ܐܦ⟩ (52) · ⟨?ܐܠܟܣܢܕܪܘܣ⟩ (54) · ⟨?ܟܕ⟩, ⟨?ܠܕܘܟܬܐ⟩ (55) · ⟨?ܐܪܥܐ⟩, ⟨?ܐܠܟܣܢܕܪܘܣ⟩ (56).
⚠ WHAT THE CONTROL DID NOT DO HERE: Schermann's Latin Appendix is a two-page CAPITULUM — one sentence per prophet — against 20 lines of continuous Syriac narrative. It adjudicates no letter, no point and no word division on this page, and most of the page has no parallel of any kind in it. Nothing above was read out of the control; the control was consulted only after keying.