Heinrich Weber, Lehrbuch der Algebra I-III - complete German source transcription (working draft)

Start here: the first file is a single 1,463-page reader containing the complete German source transcription of all three volumes of Weber's Lehrbuch der Algebra, with a bookmark tree. The three volumes follow individually, then the buildable LaTeX for each.

What this is. A transcription made page by page from the printed volumes (Bd I/II: the two-volume edition, Braunschweig 1895–1896; Bd III: the Zweite Auflage, 1908). Volume I pages 1–148 come from an earlier hand-verified pass; the remaining 2,066 pages were transcribed by an agent swarm reading 500 dpi scan images and then checked for page identity, page-boundary duplication, coverage against the scans' own OCR text layer, and LaTeX compilation. All three volumes compile.

What this is not. It is not a verified edition. Only a small sample of the 2,066 swarm-transcribed pages has been collated against the scans by hand. On those the text, mathematics, orthography and figure handling were exact; letter-spaced emphasis (Sperrung) and paragraph breaks around displayed formulas were not. Explicit uncertainty markers remain in the LaTeX wherever the transcriber could not read something — each marks a place to consult the scan. Treat this as a working draft for correction, not as an authority.

Known systematic defect: letter-spaced emphasis. The transcription instructions used for this corpus wrongly stated that unnumbered set-off statements are not letter-spaced. They are: Weber sets such statements in Sperrung in full, and the spacing continues across a displayed equation. Because the instruction was wrong rather than merely absent, the omission is consistent across the corpus rather than sporadic, and so does not show up as disagreement between pages. Every page of Band I checked against the scan at high magnification so far has carried at least one unmarked spaced passage. The instructions have been corrected and the passages are being restored page by page, each with the magnification and the evidence recorded in the source; assume any page not yet marked as checked still under-reports emphasis. Nothing else about those passages — wording, orthography, mathematics — is affected.

Source defects, recorded rather than silently repaired. The Band II scan lacks printed pages 318–319 and 732–733, recovered here from a second scan of the same volume, and photographs two gatherings twice. The Band III scan photographs printed 384–393 twice, and one leaf between printed 708 and 711 is absent from the only available scan of that volume and could not be recovered.

What changed in this version (2026-08-07). An English facing-edition translation pass was run over Band I. Its real purpose is as a second pair of eyes: the translator reads the German fragment and the scan side by side, and reports — never edits — anything that does not match. Each report is then adjudicated here against a high-magnification crop before any change is made. That channel found defect classes that no mechanical check in this project can see, because they leave the LaTeX well-formed and the compile clean:

• Lost printed material. The whole lower third of printed 306 (two sign tables and their connecting line), the closing equation of a numbered display on printed 282, the closing display of printed 373, a line of running text on printed 153. Coverage cannot catch these: a sign table is almost no letters, so the ratio does not move.
• Page misalignment. On printed 320/321 a theorem, a bridging sentence and a whole further theorem were printed on 320 but filed under 321. The continuation markers were internally consistent and both fragments carried the correct page number, so nothing mechanical could see it.
• Two boundary leaves at the Book Two and Book Three part-titles were described wrongly; in one case the part-title and its blank verso were filed on opposite leaves. The decisive evidence in both was that a show-through ghost reads backwards.
• Invented emphasis and invented text. Letter-spacing applied where the page sets ordinary roman — on one page by analogy with a neighbouring statement that genuinely is spaced — and one fabricated word in running prose. Emphasis lost and emphasis invented are equally wrong, and the invented direction is harder to notice.
• A silent correction. A section heading printed §. 136 had been “corrected” to 137. Counting the heading series settled it: the volume then carried two §. 137 and no §. 136. The page was right.

Two typographic rulings now applied corpus-wide, both properties of the fount rather than of any occurrence, and both implemented in the preamble so that no fragment text was altered: the Greek capitals in this setting are inclined (LaTeX's upright Δ is not what Vieweg set), and lower-case rho is the open-tailed form.

The English facing edition itself is not included in this version: it covers Band I only, and its audit queries are logged with verdicts still blank. It will be published when it is complete and adjudicated.