Mills 8A

Each reset is in tex/pages/, set with tex/bulletin1947.sty; ./compare-pages.sh rebuilds the images from the scans.

P. Erdős, Some asymptotic formulas for multiplicative functions, Bull. Amer. Math. Soc. 53 (1947)

J. L. Doob, Probability in function space, Bull. Amer. Math. Soc. 53 (1947)

M. Kac, On the notion of recurrence in discrete stochastic processes, Bull. Amer. Math. Soc. 53 (1947)

E. L. Post, Recursively enumerable sets of positive integers and their decision problems, Bull. Amer. Math. Soc. 50 (1944)

S. Wright, Statistical genetics and evolution, Bull. Amer. Math. Soc. 48 (1942)

Th. von Kármán, The engineer grapples with nonlinear problems, Bull. Amer. Math. Soc. 46 (1940)

1. Segment each 600 dpi page into glyph impressions, with Tesseract for a first reading and the page's lines (revival/segment.py, baselines.py).
2. Normalize the scans: each paper's ink weight and type size are measured against the Erdős paper and corrected (docweight.py).
3. Cluster identical sorts and label them (cluster.py, classify.py, assign.py). Rules do most of it; overrides.tsv holds the hand corrections, each naming one impression, so re-clustering does not invalidate it.
4. Average each sort's impressions at 4× resolution, aligned to ¼ px (masters.py). Old-style figures come from the years in the Bulletin's running heads (oldstyle.py).
5. Trace and space the masters into OpenType fonts (build_font.py). Side bearings are fitted from about 440,000 letter pairs in the text and land on Monotype's 18-unit grid. Then accents (accents.py), the math font on Latin Modern Math's tables (build_math.py), bold and 9 pt (build_sizes.py), and the pdfLaTeX fonts (pdftex/build_pdftex.py).
6. Check: check_fonts.py tests baselines, heights, side bearings and the script sorts of every font, and proof/ has proof sheets of everything the family sets.

revival/README.md describes each step, the measurements behind it, and where each glyph comes from.

Traced from the 1940s scans (each glyph the average of its printed impressions):

• every roman and italic letter and figure at 11 pt, the roman J from its 9 pt sort (66 impressions) scaled up;
• the old-style figures, from the years in the Bulletin's running heads;
• punctuation, with the 1947 quotes ‘ ’ “ ”, the question mark (10 impressions) and the en dash (a 9 pt sort, scaled);
• the accented ü á (roman) and ü Ü ä (italic), and the dieresis;
• 23 small capitals;
• bold, from the title lines and run-in heads (told by the word, not the letter): the capitals but J K X Z, the lowercase but q w z, the figures 0–7 and the hyphen;
• Greek (with Γ Σ Φ and θ), most math operators and relations, ∮ ∨, the display ∑ ∏ ∫, Fraktur 𝔄 𝔅 ℭ 𝔇 𝔊 𝔎 𝔔 ℜ ℑ 𝔖 𝔗 𝔛 and script 𝒜 𝒞 ℒ 𝒰;
• the 9 pt cut: 80 roman and 54 italic sorts;
• 43 script sorts for indices (40 first-order, 3 second-order), among them the short 1947 index arrow.

Filled in otherwise:

• from the 1922 Lanston Specimen Book (No. 8A at 9–18 pt): small caps J Q X and $;
• built from 1947 sorts: the fl/ffl and italic ffi/ffl ligatures, the em dash, the colon and ellipsis, \cdot and \cdots (from the period), bold J K X Z q w z 8 9 and bold italic (thickened, to the real bold's weight and x-height), and the script sizes that have no legible real sort (the text glyph scaled: among them the index 2 3 4 8, whose real sorts fill in);
• from Latin Modern, thickened to the type's weight: accents other than the dieresis (the acute of á and é is real only in those letters), rarer punctuation and symbols (# % @ † ‡ ¶ « » ß Æ Œ Ø Ł and so on), and the math symbols the scans lack.

Spacing is measured as well: side bearings from about 440,000 letter pairs in the text, the thin space 1947 set before : and ;, and the script positions and gaps of the Mills page. A few bearings that the pairs cannot fix (the small-cap A before a period) are set by hand from the scans.