MIT CADR font usage audit

Conclusion

The pinned System 46 source snapshot does not establish a purpose for every recovered font. It does, however, support a complete evidence-graded audit:

• 15 of the 88 source-backed logical names are selected by executable CADR Lisp code for an identifiable screen or application role;
• three names are demonstrably selected by XGP or Dover document workflows, but the name match does not prove that those printer resources contain the same raster data as the recovered CADR screen-font sources;
• three more have a specific use described by contemporary documentation or a bug report, without a surviving executable selection site;
• one was reported as frequently used, but the report gives no purpose;
• six are loaded by the standard FONTS package without an identified consumer;
• 45 occur only in the source-archive compilation list, and 15 only as recoverable source representations.

Thus 21 source-backed names have a specific, evidence-supported role and 67 have an explicit TODO: no application purpose established. “No purpose established” does not mean “never used”: font names could be entered interactively, constructed dynamically, or referenced only from compiled programs absent from this tree.

The separate set of 17 compiled-only logical names is less complete. Executable System 46 source establishes roles for MEDFNB and SHIP. Maintained LM-3 System 303 source directly establishes the editor role of SEARCH, whose ten visible labels agree with the System 46 raster even though the two compiled artifacts are not byte-identical. The other 14 remain qualified runtime, load, or inventory observations with purpose TODO.

The machine-readable companion is the font-usage-catalog.json. It has one entry for every source-backed logical name and every compiled-only logical name. The 62 retained alternate source representations inherit the result of their logical name; the snapshot contains no evidence that an application selected one recovered source variant rather than another.

Scope and method

The audit uses the complete src/ tree of mietek/mit-cadr-system-software at commit 8e978d7d1704096a63edd4386a3b8326a2e584af, the same revision used by the source recovery. Targeted cross-checks use the LM-3 Fossil repository's System 303 check-in 4df393c68d7f083ce42d5c377039d26043cc18a9031ace28258dc97f4137eb91 only when the primary snapshot lacks an executable consumer. Those cross-version findings are identified explicitly and never treated as proof of byte identity.

For each of the 88 logical names in the tracked asset catalog, the audit searched case-insensitively for:

• exact symbols, including FONTS:<name>;
• ITS filenames such as LMFONT;<name> and QFASL load declarations;
• editor mode lines, document-font directives, generated XGP font tables, manuals, bug mail, and source comments;
• binary files treated as text, which exposed otherwise missed evidence in an archived editor environment and generated document output;
• plausible aliases only when another source established the mapping.

Every surviving context was read manually. Common-word collisions such as BUG, MOUSE, SEARCH, and SHIP were rejected unless the context was font-specific. The ST6, ST8, and ST10 strings in cadr/stat.drw are hardware signal labels, not font uses. The ordinary word “pronto” is not evidence for PRONTO.

The zero-hit boundary is necessarily limited. The window system can construct and load an LMFONT filename from a symbol, and ZWEI can read an arbitrary font ID from the minibuffer or load names computed from buffer font metadata. This is why the negative result is “no specific use established by preserved literal references,” never “unused.”

Two shared references define important negative boundaries:

• fcmp.xfile calls its list essentially the archive directory, copies out the AST members, and compiles them. Presence there proves a source-build operation, not an application role.
• pkgdcl.230 declares the standard FONTS package files. Presence there proves planned system loading, not why a font was loaded.

Evidence classes

The machine catalog additionally uses cross-version-direct-runtime for compiled SEARCH: executable System 303 source establishes its role, but no selector survives in the primary System 46 snapshot and byte identity is unproved. That status does not occur in the source-backed table above.

Direct runtime uses

These are the strongest findings because executable source selects the font object.

Document-output name matches

These references establish use of a printer or formatter resource with the same name. They do not establish that the XGP or Dover resource is byte-for-byte or glyph-for-glyph identical to the recovered LMFONT screen source.

Contemporary use without a surviving selector

No application purpose established

Every name below is an explicit negative audit result for this snapshot. No role is inferred from an abbreviation, apparent family resemblance, glyph shape, or metric.

Compiled-only fonts

These names are not among the 88 source-backed derivatives. They survive only as compiled font files in this snapshot and must not be silently merged with a similarly named source-backed font. All 19 physical artifacts, comprising these 17 names plus the two older compiled versions below, are now decoded and rendered in the compiled QFASL font recovery and its machine-readable catalog. Recovery establishes their runtime font data, not their application purpose.

The common compiled-file evidence is the preserved LMFONT media inventory. N43XMS and NTOG are older compiled files corresponding to the source-backed 43VXMS and TOG names. They occur only in that inventory; the established uses of the newer logical names are not automatically claims about those older artifacts.

Alias and identity decisions

• CPTFON is the source/QFASL artifact whose cold load provides runtime FONTS:CPTFONT. This is an operational artifact-to-object mapping, not evidence that two runtime Lisp symbols are aliases. This specific mapping is directly evidenced; it is not a general rule for similar names.
• HIPO10 is not equated with HIPPO10, which occurs in generated XGP font tables. The snapshot supplies no alias declaration, and the checked XGP bodies do not select the HIPPO10 slot.
• CPT-HL10, CPT-HL10B, and CPT-TR10I are distinct compiled logical names, not aliases for the source-backed HL10, HL10B, and TR10I rows.
• TR12B1 is not treated as an alias for TR12B; no mapping was found.
• MAT10A and MATH10 are separate archive entries.

Open questions

• Search additional period load bands, user files, source tapes, manuals, and application QFASLs for the 67 source-backed and 14 compiled-only names whose purpose remains TODO.
• Compare the running System 46 FONTS package and editor font maps with this static source audit. That can discover dynamically loaded names, but a single running world still would not prove every historical use.
• Establish whether the 25FR3, GACHA8, and MATH10 printer resources are derived from, or merely share names with, the recovered LMFONT raster sources.

Last verified: 2026-07-17.