Layout transcription from primary sources

Record of transcribing real manufacturer type catalogs into v4’s layout.rows presets: what was read, what was imported, what was deliberately left out, and — most importantly — why each judgement call went the way it did. Several imports rest on reasoning rather than on reading a glyph off a scan, and that reasoning is the part worth keeping.

All layout data lives in lib/layouts/<machine>_layout.py. This document is the narrative; those modules are the source of truth for the values.

For the complete shuttle-by-shuttle enumeration with import status, see CATALOG_INDEX.md — 258 of 346 catalogued Hammond shuttles and 73 of 84 sighted Blickensderfer shuttles are currently covered. Both tables are generated, so those counts cannot drift from the code.

Sources

Source	Location	Form
Hammond, 1920	`windows_tailscale:C:/Users/Leonard/Documents/Python/images to pdf/HammondTypeShuttleCatalogSmall.pdf`	6 PDF pages (4 are two-page spreads), organised by shuttle number
Hammond, 1915	`windows_tailscale:E:/Leonard/Typewriter/Books Manuals/Hammond_type_Catalog_1915.pdf`	35 pages, organised by language
Blickensderfer	`windows_tailscale:E:/Type Elements/Blickensderfer/Catalog/20230113_01{55..68}.jpg` (local copy: `~/Blickensderfer-Catalog`)	14 page scans, organised by language/market

None have a text layer — all are scans, read as images. 200 dpi is enough for the letter rows; individual ambiguous glyphs needed 300–600 % crops.

Higher-resolution originals exist if a glyph is ever contested: the Hammond 1920 source TIFFs (HammondShuttles143-148.tif, ~300 MB each) and the full-size HammondTypeShuttleCatalog.pdf (48 MB). The Blickensderfer directory’s BLICKENSDERFER FONTS_..._2023.zip is the same page images zipped, not a separate source.

These catalogs outrank v1/v2 .scad on layout content. That is a deliberate, narrow exception to CLAUDE.md’s “v2 is ground truth” rule: v2 was itself transcribed from these catalogs, so where they disagree the catalog is the earlier authority. It is only an exception for layout character content — v2 remains ground truth for geometry and dimensions.

Method

Four habits did most of the work. They generalise to any future catalog.

1. Read by column, not by row. Each column is one physical key with three levels: unshifted / shifted / figure. The three rows therefore constrain each other, and a reading that breaks the column is wrong even if the glyph “looks like” something. This turned several judgement calls into near-certainties (below), and let print artifacts be dismissed rather than transcribed.

2. Check the letter inventory. Compare the lowercase row’s letters against the uppercase row’s. A duplicated letter plus a missing one is the signature of the transcription-error class that turned up twice. This is mechanical, needs no scan, and found the Blickensderfer bug independently of the catalog.

3. Group by identical rows; typeface is not layout. Most catalog entries differ only in typeface — Small/Medium/Large Roman, Gothic, Italic, Clarendon, Script all print the same rows. Within a keyboard the letter rows are constant, so essentially all real variation is in the figures row. ~80 Hammond entries collapse to a handful of layouts. One preset per distinct layout; the typeface is the Font tab’s job.

4. When it can’t be settled, don’t import it. A guessed glyph silently builds a wrong shuttle — there is no error, just a wrong part. No layout beats a wrong layout. Every exclusion below is recorded at its definition site with the specific reason.

Preset naming

One convention, all ten machines:

<Keyboard>[, <Language>][ (<Variant>)]

Title Case. A comma introduces the language, or the variant when there is no language; parentheses hold the variant once a language is named, with multiple variants comma-separated inside the one pair. So Universal, Universal, Math, Ideal, Spanish (¢), DHIATENSOR, British (Fractions, Mimeograph).

“Keyboard” is whatever axis that machine’s own sources divide its range along — Ideal/Universal for the Hammonds, the letter arrangement (DHIATENSOR/QWERTY/CHARIENSTU) for Blickensderfer. A machine with only one keyboard starts at the language: Bennett’s British, Selectric’s United States.

This was a Hammond-only rule while six other machines spelled their presets SCREAMING_SNAKE (QWERTY_BRITISH, UNITED_STATES, GERMAN_MOD). Unified in one pass, verified by checking that every renamed key still maps to byte-identical rows and that no unrenamed preset moved.

Two consequences worth knowing:

A SCREAMING_SNAKE name in a code comment is now a quote of a v1/v2 .scad array, not a v4 preset. blickensderfer_layout.py, helios_layout.py and the two Selectric modules all carry such quotes deliberately and say so.
Renaming is safe because no config stores a preset name, only the rows it produced — the picker is a template chooser, not persisted state. The one place names are stored is blickensderfer_catalog.py’s preset column, and gen_catalog_index.py exits non-zero when one stops resolving.

Bugs found and fixed

Two real defects, both the same shape, both in the original source rather than the v4 port, and both silently dropping a letter an English typewriter cannot do without.

Hammond Ideal — missing `b`

v1/Hammond/HammondSplitShuttle.scad:24   IDEAL=["?zxqkjgdmpcfld,…
v2/hammond_split.scad:76                 Ideal_Element=["?zxqkjgdmpcfld,…

d where the catalog reads b — leaving b absent entirely and d duplicated. Every Ideal entry in both catalogs reads ?zxqkjgbmpcfld, / !ZXQKJGBMPCFLD;. Confirmed at 200 dpi.

Blickensderfer CHARIENSTU — missing `Y`

v1/Blickensderfer/Blickensderfer2.scad:82,85,88   "XQZV&PFLOCHARIENSTUGMDB:WKJU"
v2/lib/layouts/blick_layouts.scad:18,21,24        "XQZV&PFLOCHARIENSTUGMDB:WKJU"

U where it should be Y, in both CHARIENSTU arrays (v4’s CHARIENSTU, German and CHARIENSTU, German (Modified)). Found by habit 2 — the uppercase row’s inventory was a–z with U duplicated and Y missing, while its own lowercase row was clean. The catalog then confirmed it (GMDB:WKJY, Bohemian 426/443).

The same inventory check now passes on every Blickensderfer preset. (DHIATENSOR, Hebrew-English’s row 0 is Hebrew, so the Latin check correctly does not apply to it.)

What the catalogs confirmed unchanged

Hammond Universal — an independent transcription of the 1920 catalog’s Universal rows, reversed into this machine’s storage order, came out byte-identical to the shipped preset on all three rows.
Hammond Ideal standard / fractions — the 1915 catalog’s English entries 37/10 and 1/2 match the 1920-derived layouts character for character, five years apart. Two independent printings agreeing is much stronger evidence than one.
Blickensderfer DHIATENSOR — fourteen English Scientific entries (365, 407, 409, 455, 457, 356, 362, 374, 474, 440, 499, 201, 223, 308) all print it identically, differing only in typeface. QWERTY gets the same treatment from eight Universal entries (325, 406, 418, 364, 359, 497, 304, 216).
Blickensderfer WKJY — German 423’s row 1 was a third independent confirmation of the WKJU→WKJY correction above, after the letter-inventory check and the Bohemian scan.

Judgement calls

The decisions that were reasoned rather than simply read. Each records what would have to be true for it to be wrong.

ƒ not f (Dutch Ideal, position 29). Row 0 already carries a lowercase f; a figures row does not repeat a letter that has its own key. The guilder sign is also the one currency mark a Dutch machine needs that £/$ don’t cover. The 200 dpi scan shows the hooked italic form.

ç not ¢ (Spanish Ideal, position 4). At 600 % the glyph has a cedilla hook and no vertical stroke; the ¢ in the CENT variant at that same position clearly has one. The two were compared side by side.

1ó2 not 162 (Spanish Ideal, positions 7–9). The accent is visible, and a second 6 would collide with the 6 already at position 16. The resulting accent set — á é í ó ú ñ ¡ ¿ plus ¨ as the diaeresis dead key for ü — is complete and self-consistent for Spanish.

& re-homed onto the shifted . key (Universal Fractions). Nine figure slots became fractions, displacing &; it reappears at the shifted . position, which is the one key printing a bare dot at all three levels on the catalog’s own keyboard plate — i.e. the only spare slot. A mechanical diff confirms exactly those 9 + 1 positions move and nothing else.

ſ not ß (German, old orthography). Held for a while, then settled by finding both printed on the SAME page in the same typeface and size — Universal 119A (old orthography) beside Universal 117 (New Orthography), catalog p.44. Side by side they are plainly different glyphs: 119A a single descending stroke with a top hook and no bowl, 117 a clear bowl. So old-orthography German shuttles carry long-s and New Orthography ones carry eszett — and the same old/new split appears independently in the Ideal pair 36 vs 36B. Two independent pairs agreeing is what made it callable; one ambiguous glyph on its own would not have been.

The payoff was disproportionate: Ideal, German turns out to differ from Ideal, Dutch in exactly ONE position (ƒ vs ſ), and resolving that single character moved 18 catalogued shuttles from held to covered.

⅌ — the one the scans could not give up. Universal position 5 resisted three faces and every zoom: a p-like form with a bowl and a stroke. It is the PER SIGN (U+214C), identified by the machine’s owner rather than from the page. That single character was gating 19 catalogued shuttles — Universal German, Universal Dutch, and the German fractions family — which is why it was worth stopping to ask rather than guessing.

It also corrected something already written down. An earlier note here called v2’s ⅌ and § in Qwerty_Element transcription errors, “neither of which any catalogued Universal entry shows”. They are at positions 5 and 20 — exactly the two positions where Dutch and German Universal differ from English Universal, and both characters appear there in the catalog. v2’s array is English Universal with the Char_Mod character swapped in, not a botched transcription. Corrected at the definition site.

An italic shuttle settled what the roman ones could not. The guilder ƒ had been argued for from structure (row 0 already has an f) rather than seen cleanly. Dutch Universal 119E is an ITALIC face, and at that size the hooked form is unmistakable in the same slot — so the reading is now observed, not just inferred. Worth remembering as a tactic: when a glyph will not resolve, look for the same layout in a different TYPEFACE before reaching for a better scan.

Greek is held on the diacritics, not the letters. The Greek shuttles (7, 75, 82, 8, 112A, 112B) are polytonic: alongside the alphabet they carry standalone breathing and accent marks, and what appear to be iota-subscript vowel forms. The letters read fine; assigning the marks to codepoints is the problem — acute vs tonos, perispomeni vs a plain tilde, psili vs a left quote are all distinguishable in principle and not at this scan resolution. This one wants someone who reads polytonic Greek, not a better crop.

Cap height is evidence. Hungarian Universal read as ó twice in its figures row, which looked like the duplicate signature. At 700% the first reaches the full height of the adjacent digits and the second sits at x-height: Ó and ó, not a duplicate at all. The inventory agreed — Ideal Hungarian carries both cases, and reading both as lowercase would have left Ó absent from the shuttle entirely.

A missing letter is only evidence if the letter should be there. The same entry appeared to be missing í, which reinforced the false duplicate reading. It isn’t missing: no Hungarian shuttle on either keyboard carries í, per the machine’s owner. The check is only as good as the alphabet you compare against, and for a 1915 shuttle that alphabet is a question about the machine, not about the language.

Derive a near-identical variant, don’t retype it. The three Mathematical shuttles differ in exactly two figures-row positions — 135 */|, 135B {/}, 135C ν/μ. 135B is built from 135 by substituting those two characters programmatically, so the only difference between the two presets is the one actually observed. Retyping four 30-character rows of Greek and mathematical symbols to change two of them is precisely how a transcription error gets introduced — the thing this whole document exists to avoid.

This also confirmed the shipped Universal, Math layout, which came from v2 rather than the catalog: it matches shuttle 135 character for character.

Quote marks are a real character, and I got five of them wrong. The Universal German/Dutch figures row has “ (U+201C) at position 4, not ” — at 600% the raised pair curls the same way as the „ beside it (6-shaped, not 9-shaped). Five layouts had been imported with the wrong one. Corrected from that single verified instance, since all five share that row half.

The IDEAL family’s quote was deliberately NOT normalised to match. German and Dutch genuinely diverge here — German pairs „…“, Dutch pairs „…” — so “make them consistent” would be exactly the wrong instinct. It stays as read, flagged as unverified, rather than tidied into a plausible uniformity.

Bulgarian: solved by accounting, not by reading. Three glyphs on this shuttle would not resolve visually at any zoom. What settled them was counting the alphabet. With ѫ at position 0, row 0 holds 30 distinct letters and no duplicates; pre-1945 Bulgarian has 33; the three absent are х, щ and ѭ; х/Х and щ/Щ are plainly in the figures row — leaving exactly ѭ unaccounted for and exactly one unidentified pair left to hold it. The layout is forced, without ever reading the glyph.

The same accounting rejected a plausible-sounding alternative. Ѣ (yat) was proposed for position 0, and the form is stylised enough to invite it — but yat is already on this shuttle at position 28, in a plainly different shape (stem, top crossbar, bottom bowl), and putting it at position 0 too would have duplicated a letter while leaving the alphabet short. The count says which readings are possible, not just which are plausible.

The inventory check can indict the SOURCE, not just the port. It was built to catch v1/v2 transcription errors, but on Bulgarian 31C it fired against the catalog itself: ж is printed at both position 0 and position 12 of row 0, while ѫ (big yus) — which pre-1945 Bulgarian cannot do without — is absent from the shuttle entirely. A duplicated letter alongside a missing one is precisely the signature. Either the catalog has a typo or that glyph is not ж; either way it is not importable, so all four Bulgarian entries are held. Worth knowing that “primary source” does not mean “error-free” — it means earlier authority.

The letter-inventory check works on Cyrillic too, and is what made a Russian read trustworthy rather than a guess. Of the six pre-reform letters absent from shuttle 143’s row 0 (ж х ц э ѳ ѵ), five reappear in its figures row — the shuttle is complete except ѳ (fita), which was already rare in 1915 and abolished three years later. A misread letter row would almost certainly have left that accounting broken. Note Ѵ (izhitsa) genuinely is V-shaped, so a Latin-looking V there is the correct character, not a Latin letter.

Read the siblings, not just the entry. Two holds were cleared this way rather than by better optics. Roumanian Ideal 92’s unresolved closing mark was settled by Universal Roumanian 32C, which carries a bare ¸ (U+00B8) outright — once the language is known to have a standalone cedilla, and the shape matches, no other reading is left. And the Scandinavian pair settled each other: Swedish-Finnish and Danish-Norwegian differ in exactly ONE letter position, å against æ, so the catalog demonstrably had both and chose — neither is a misreading of the other. Danish-Norwegian carrying Æ Ä Ö rather than Æ Ø Å is historically odd, and was held on exactly that basis until its sibling made the reading certain. It is recorded as read, not as one would expect it to be.

Two print artifacts dismissed, not transcribed. Shuttle 26’s row 0 p:- and 41’s row 1 P:: are both the ;/: key, whose unshifted and shifted forms are fixed by the keyboard. Transcribing them literally would have invented characters that key cannot produce.

“LARGE FRACTIONS” is a typeface, not a layout. Shuttle 40 prints the same rows as 26/52 in a larger fraction face.

“Caps and Small Caps” is a typeface, not a layout. Row 0 stores lowercase, because on a small-caps face the lowercase codepoints are the small capitals — a correctly-designed font then renders the row as the catalog prints it, with the cases matching. Storing capitals would hard-code the appearance into the layout and fight whatever font is selected. Consequence: Universal, Caps and Small Caps (27, 27E) collapses into Universal, Standard and is now defined by reference so the two cannot drift. Spanish 5A does not collapse — its figures-row accented letters really are full capitals (ÑÍÉÚ where plain Spanish has ñíéú), and those are reached by the figure shift rather than the case shift, so that is a genuine character difference.

Two spellings kept rather than one picked. v1/v2 read 9[0] where every catalogued Ideal entry reads 9(0). Both survive as separate presets — Ideal (£)/Ideal (⅌) preserve the shipped source history, Ideal, Fractions preserves the printed catalog. Same treatment as the pre-existing £/⅌ pair.

Storage order is load-bearing and differs per machine. hammond stores rows reversed; hammond_split stores them in catalog reading order (its TextAssemble does its own per-half [14-i]/[29-i] reversal). Reversal is applied programmatically, never by retyping.

What was imported

Hammond (both machines — 10 presets on `hammond`, 11 on `hammond_split`)

Universal is qwerty; Ideal is Hammond’s own proprietary arrangement, not a qwerty remap. Both are real for both machines, so both are offered on both.

Layout	Shuttles covered
Universal, standard	23/23B, 24/24B, 25/25A/25B, 158/158A, 180, 96, 134, 170, 68, 169, 97B, 28, 80, 145
Universal, fractions	26, 40 (LARGE), 52, 80A, 97
Universal, caps & small caps	27, 27E (same rows as standard — typeface)
Ideal, standard	10/10A/10B/94, 37A, 51/51A/3B, 60, 118, 70, 144A
Ideal, fractions	1/48/48A, 2, 3/3A, 4, 5, 6, 9
Ideal, Dutch	36A, 11A, 12A, 76A, 13A/102B, 91A, 78A
Ideal, Spanish	65, 16, 46
Ideal, Spanish (¢)	65B, 16B, 46B
Ideal, Spanish caps & small caps	5A
Ideal, Croatian	58, 12C
Ideal, Danish (fractions)	87
Ideal, Portuguese	63
Ideal, French	61, 14, 62, 15
Ideal, German (New Orthography)	36B, 114
Ideal, German (old orthography)	36, 11, 12
Universal, German (New Orthography)	117, 117A, 119C, 124, 145A
Universal, French	69, 32, 32A, 67, 104, 104A, 85, 111, 111A, 134E, 145B
Universal, French-German-English	32E
Universal, Esperanto	135A, 135D
Universal, Italian	134F, 73, 32B, 67A, 69A, 104B, 150B, 85A, 111B
Universal, Spanish	33, 66, 66B, 47, 89, 28A, 68A
Universal, Portuguese	150, 103, 103A, 103B, 159
Universal, Roumanian	32C, 32D, 97A
Universal, Swedish-Finnish	93A, 90, 101, 101A, 101B, 134B
Universal, Danish-Norwegian	55D, 90A, 93B, 101C, 134C
Ideal, Roumanian	92
Ideal, Bohemian	54, 86, 120, 153C, 154A, 155A, 156A, 157A
Ideal, Polish	155, 77, 154, 156, 157, 153B, 121
Ideal, Hungarian	152, 44, 57, 151, 74
Universal, Chilian	47A, 89A, 16A, 17A, 46A, 65A, 66A, 33A
Universal, Russian	143
Universal, Bohemian	116
Ideal, Italian	61A, 14A, 62A, 15A
Universal, Dutch (Fractions)	34D, 50C, 71F, 84C, 119E
Universal, German (New Orthography, Fractions)	119C, 117D, 117E, 55F, 101E, 71E
Universal, Dutch	71A, 117B, 34A, 50A, 119B
Universal, German	119A, 84, 34, 50, 71, 117, 119
Universal, Russian (Old Style)	29, 29A, 29B
Universal, Bulgarian	31C, 29C, 42C, 115C
Ideal, Chilian	65A, 16A, 46A, 17A, 83A
Universal, Polish	141, 30, 150C
Universal, Math (braces)	135B
Universal, Math (Greek)	135C
Universal, Hungarian	134A, 56, 64

hammond_split’s UNIVERSAL also revives v2’s Qwerty_Element (Layout_Selection=1), which was complete in the source but never wired into the picker — with two characters corrected against the catalog (v2 had ⅌ and § where every catalogued Universal entry has × and ^, which hammond.yaml’s own Universal row already spelled correctly).

Blickensderfer (28 presets, 20 of them from the catalog)

All 14 pages have now been read: 73 of the 84 shuttles they contain are covered. See CATALOG_INDEX.md for the per-entry table.

Layout	Shuttles covered
`DHIATENSOR, British (Fractions)`	Imperial 212, and Scientific E458, 412, 428, 435, 405, 363, 393, 357, 454, 300, 205
`QWERTY, British (Fractions)`	Elite 350, Small Roman 494, Large Roman 379, Italic 371, Script 337, Vertical Script 217
`DHIATENSOR (Fractions)`	English Fractional 436, 424, 425
`DHIATENSOR, British (Literary)`	Elite Literary 381, Small Roman Literary 462, Extra Large Roman Literary 307, Italic Literary 383, Script Literary 395, Vertical Script Literary 213
`CHARIENSTU, German (Pfennig)`	Small Roman 404, Large Roman 423, Large Roman 378, Italic 489
`QWERTY, British`	Small Roman 441, Large Roman 442
`DHIATENSOR, British-American`	Large Roman 432, Small Roman 433
`DHIATENSOR, English-Japanese`	Small Roman 332, Large Roman 333
`DHIATENSOR (Chemical)` / `QWERTY, British (Chemical)`	385 / 222
`DHIATENSOR, Cosmopolitan` / `QWERTY, Cosmopolitan`	328 / 367
`DHIATENSOR, British-India` / `DHIATENSOR, British` / `QWERTY, British (Literary)`	458 / 407½ / 203
`QWERTY (Fractions)` / `DHIATENSOR (Fractions, Alternate)`	494½ / 447
`CHARIENSTU, German (Eszett)` / `CHARIENSTU, German (Fractions)`	303 / 204
`CHARIENSTU, Danish` / `DHIATENSOR, Hungarian`	420 / 415

Much of this catalog is one layout in many typefaces, and the same few one-position swaps recur across markets:

$ ↔ £ is the American/British split, and it is only that one position, four times over: QWERTY / QWERTY, British, DHIATENSOR / DHIATENSOR, British, DHIATENSOR (Fractions) / DHIATENSOR, British (Fractions), and QWERTY (Fractions) / QWERTY, British (Fractions).
DHIATENSOR, British-American is the wheel that refuses to choose — the only one carrying both $ and £, paid for with three moved slots.
DHIATENSOR, British-India is the British fraction wheel with ₨ for ⅞ and ⅓ for ⅛.
DHIATENSOR, English-Japanese carries ¥, $ and £ at once.

Two of the four wheels that aren’t a fraction bank are symbol banks: CHEMICAL_* swap the figures row for subscript digits (₁₂₃₄₅ before the full-size digits, ₆₇₈₉₀ after, for formulae like H₂SO₄), and the two Cosmopolitan wheels spend nine slots on free-standing accents plus æ/œ, paying for them by dropping the digit 1 entirely.

What was NOT imported, and why

Recorded here in summary; the authoritative per-entry reasons live in CATALOG_SHUTTLES in lib/layouts/hammond_layout.py and in the comment block at the end of lib/layouts/blickensderfer_layout.py.

Hammond 1920

41 Small Roman Fractions — a genuinely different second fractions scheme (diagonal fractions, and it keeps & in the figures row rather than moving it). Numerators not separable at this resolution.
162 Medium Gothic Fractions — right half matches the imported fractions layout exactly, but the left half’s glyph after 4% is scan-damaged.
184 Gothic Special Fractions — prints four lines, so it does not fit the three-row shape without deciding which line is the figures row.
23E / 23F / 23G / 136 — each differs from standard in only one or two figure slots, and those slots are exactly the unidentifiable ones.
Medical / chemical (43, 43A, 107, 179, 21, 18) — purpose-made symbol sets with no reliable Unicode reading from this scan.
Diacritical / library (113, 122, 48C) — bare combining accents printed in isolation; which precomposed/combining codepoint each means is a judgement call, not a reading.
Literary (192, 193, 194) — subscript/superscript digit banks.
Non-Latin / special — 195 Astronomical, 196/197 International Phonetic, 135/135B/135C Mathematical, 112C Greek, 59/20 German Text (fraktur), 165/167 Yiddish (Hebrew), 185 Check Writer (perforating, prints as dot matrices).

Hammond 1915 (per-language)

Not scan-quality problems — these entries are legible — but each needs its own character-by-character pass, and a wrong accent builds a wrong shuttle just as silently as a wrong glyph.

Latin, accents only: Croatian (58, 12C), Danish (87, 88), Portuguese (63, 63A, 63B, 106), Roumanian (92), Polish (156, 153B, 157), plus the language sections on pp. 10–20 not yet sampled.
Pre-reform Cyrillic: Russian (49, 35), Servian (125) — use letters dropped in the 1918 reform (ѣ, і, hard-sign ъ) and need a font carrying them.
Ornate faces: language-specific Vertical Script / Italic Script (e.g. 83A Chilian, 78A Dutch) whose forms make individual accent marks hard to separate even where the underlying layout is known.

Blickensderfer

Eleven of the 84 sighted shuttles, all for reasons of script or of type that has no codepoint — none for image quality, which is consistently better here than in the Hammond scans.

Bohemian 426/443 — CHARIENSTU letters with a Czech figures row whose doubled dead-key accents (´ ´ and ˇ ˇ) cannot be separated reliably.
Armenian 218, Ancient Greek 309, Hebrew 354/358/348/351 — whole scripts, each needing its own pass rather than being folded into a Latin-wheel batch. 358 needs no reading at all once 354 is done: the catalog says in prose that it is 354 with £ for $.
Bulgarian 452½ — Cyrillic; the same case as the Hammond Bulgarian shuttle, which alphabet accounting eventually solved.
Special British 387 — QWERTY, British (Fractions)’s row with its last five slots given to shilling numerators 1⁄ 3⁄ 5⁄ 7⁄ 9⁄, each cast as one piece of type. Only the first has a codepoint (⅟, U+215F), and layout.rows is strictly one character per position. The reading itself is not in doubt.
British Telegraph 376 — not the usual three-row 28-column shape.

Two entries in this list used to be wrong, and both failed the same way — a group judged on one glance instead of read.

“Each fraction entry packs a different bank.” It doesn’t. Twenty-one entries collapse to four layouts plus four genuine one- or two-position variants. This one deferral was hiding a sixth of the whole catalog.
“English-Japanese 332/333: kana.” There is no kana on it. It is a Latin trading wheel carrying ¥ $ £.

The lesson is narrow and worth keeping: a shared heading (“Fractional”, “English-Japanese”) predicts nothing about the rows underneath it. Read one entry per group before deferring the group.

Font implications

A layout is only buildable if the selected font actually has its glyphs. font_coverage.py --preset MACHINE:"Preset Name" reports this.

OCR-A II Regular, configured for both Hammond machines, lacks the fractions and every accent. This is not a regression from the imports — it equally lacks ¢/°/× from the already-shipped Universal, so every Hammond preset including the defaults is in the same position.
AverageMono Mod covers all four per-language layouts and the fractions set with no missing glyphs.
529 of 1377 fonts in the library cover the Universal fractions set.
The Caps and Small Caps entries need an actual small-caps face to look right — by design, since that is now the only thing distinguishing them.

Remaining work, prioritised

The test is character identification, not font availability. A layout gets catalogued as soon as every character can be identified and has a Unicode codepoint. Whether a font on this machine currently carries the glyphs is a separate, solvable problem — fonts can be found or made — and is explicitly NOT a reason to leave a layout out. An earlier version of this list got that wrong and parked the non-Latin scripts as “blocked on fonts”; they are not blocked, just unworked. What genuinely blocks an import is a character the scan will not resolve, since a guessed glyph silently builds a wrong shuttle.

Ordered by value per unit of effort.

1. Hammond 1915 Latin languages — biggest remaining vein, low risk. English, Dutch, Spanish, Croatian, Danish and Portuguese are done. Status of the rest already sighted:

Language	Shuttles	Page (PDF)	Status
Croatian	58, 12C	14	imported
Danish	87	14	imported (fractions)
Danish	88	14	held — one slot: blank, or an under-inked `_`?
Portuguese	63	20	imported
Portuguese	63A, 63B, 106	20	unread
Roumanian	92	20	held — the mark closing row 1 won’t separate from `¸`
Polish	156, 153B, 157	20	unread
French	61, 14, 62, 15	16	imported
French	39 (Special)	16	unread — a fractions variant
German (New Orthography)	36B, 114	16	imported
German (old orthography)	36, 11, 12	16	imported — the `ſ` was settled, see below
Esperanto	105	16	letter rows clear (ĥ ĵ ŝ ĉ ĝ ŭ); figures row not yet resolved
French-German	79	16	unread

The catalog has a full index on PDF page 12, listing every shuttle number, its description and the page it appears on — including languages not yet sighted anywhere above: Hungarian, Swedish-Finnish, Danish-Norwegian, Italian, Bohemian, Polish-German, Argentine, Braille, Turkey-Persian, Arabic, Arabic-Persian-Turkish, Bell Visible Speech, and Dutch-German-French-English. Start there rather than paging through the scans.

Pages 10–13, 15–19 and 21–35 have not been sampled at all — there are more languages there than the table above. Each section is one layout shared by several shuttles, so the yield per verified layout is high. Difficulty is ordinary accented Latin.

2. Blickensderfer — done, except for scripts. All 14 pages have been read; 73 of their 84 shuttles are imported. What is left is the eleven in “What was NOT imported” above, of which seven are whole scripts (Armenian, Greek, Hebrew ×4, Bulgarian) and the rest are Bohemian’s dead-key accents, 387’s shilling type and 376’s odd shape.

The bigger Blickensderfer gap is now the missing pages, not unread ones: the scans hold catalog pages 3–14, 18 and 21, so pages 1–2, 15–17, 19–20 are absent — roughly 42 more shuttles at this catalog’s steady six per page. Alphabetically, 15–17 fall between English-Japanese and German (so French and Esperanto sit there) and 19–20 between German and Greek. Nothing can be done about these without more scans.

3. Hammond 1920 leftovers — small and fiddly. 41, 162, 184, 23E/23F/23G, 136. Each is one or two unresolved figure slots, or a non-standard row shape (184’s four lines). Worth a pass with the higher-resolution TIFF originals rather than more zooming on the small PDF — that is the specific thing likely to settle 162’s damaged glyph after 4% and 41’s fraction numerators.

4. Non-Latin scripts — transcribable, just unworked. Hammond’s non-Latin sets (195 Astronomical, 196/197 Phonetic, 135/135B/135C Mathematical, 112C Greek, 165/167 Yiddish), the 1915 pre-reform Cyrillic (Russian 49/35, Servian 125), and Blickensderfer’s Armenian 218, Ancient Greek 309, Bulgarian 452½ and Hebrew 354/358/348/351. All of these have complete Unicode coverage — the pre-reform Cyrillic letters are ѣ U+0463, і U+0456, ъ U+044A, Ѳ U+0472 — so they meet the bar and can be catalogued by anyone willing to read them carefully. They need script familiarity, not a font.

One of these is easier than it looks: 59/20 German Text is Fraktur, a TYPEFACE. Its characters are ordinary German letters (umlauts and ß) in a blackletter face, the same way “Caps and Small Caps” turned out to be a face rather than an arrangement. It should be read as German, not as a separate script.

Not worth doing: Hammond’s medical/chemical (43, 43A, 107, 179, 21, 18), diacritical/library (113, 122, 48C) and literary (192, 193, 194) sets, and Blickensderfer’s British Telegraph 376. These are purpose-made symbol banks with no reliable Unicode reading from these scans; they would need a different, better source rather than more effort on this one.

Continuing this work

Pick a target from the priority list above.
Render the page at 200 dpi (pdftoppm -r 200 -png -f N -l N), crop the entry, and zoom to 300–600 % for any contested glyph.
Apply the four habits above — especially the column reading and the letter-inventory check.
Verify every shuttle claimed to share a layout actually prints the same rows; don’t extrapolate from one entry to its siblings.
Add the layout to the machine’s lib/layouts/ module, wire it into both Hammond machines if it’s a Hammond layout (respecting the per- machine storage order), and add a LAYOUT_PRESET_BASELINE_ROW_BY_MACHINE entry for hammond.
Run the hard gate (see CLAUDE.md) — adding a preset should leave every config’s mesh output byte-identical, since it doesn’t touch the active layout.rows.
Move the entry out of the “not imported” list and into CATALOG_SHUTTLES, with the shuttles it covers.

Note on `config/blickensderfer.yaml`

Its row 2 reads * where the DHIATENSOR preset has %. That is a deliberate hand-edit through the Layout tab (modify_glyphs: true), not drift — don’t “correct” it.