The historicalsource repositories (2019) - title versions

In April 2019 the leaked Infocom source code was published on GitHub under the historicalsource account, one repository per game. Treated as a compilation, it is the largest one there is: 49 repositories contain ZIL source, and 46 of them also carry compiled Z-machine story files - 70 distinct builds at the current HEAD, plus 19 more that exist only in earlier commits (see the git-history section below), enumerated the same way as the disc tables. Captured 2026-09-03 by walking each repository’s git tree through the GitHub API and reading each story file’s z-code header; checksums.txt records every file’s git blob SHA, MD5 and SHA-1 so the exact bytes remain identifiable.

One thing this table does not claim: that a repository’s compiled files correspond to its source. The repositories’ own READMEs state the binaries “were there as of final spin-down of the Infocom Drive” - leftovers archived beside the source, not builds of it - and the data below proves the point where it can: witness pairs one source tree with six builds spanning sixteen months, and checkpoint pairs Checkpoint source with a Journey binary. With no surviving official compiler, source-to-build correspondence is unverifiable even where it is probable (the lone builds in zork1 and zorkzero fit their source trees well). The commit history settles how the binaries got there: they arrive in the initial source-import commits of 2019-04-14 (they are drive contents, not later additions), and the only archivist post-processing is the 2019-04-16 pair of reorganization commits, “Moved compilations to subdirectory” (creating COMPILED/) and “Extensions on Z-Machine Fixed” (adding the modern .z3/.z5-extension copies of the drive’s .zip files).

Column notes:

• Most repositories carry each build twice (a COMPILED/*.z* copy and a *.zip twin, “ZIP” being Infocom’s own name for Z-machine binaries); rows list every path that shares the one blob.
• “Built-in hints” uses the same test as the disc tables: the in-game hint system’s machinery in the decoded text, not the mere presence of a HINT verb. The one ? is a build txd cannot disassemble.
• “Same z-code as” marks builds byte-identical (up to the header-declared length) to a build on a cataloged disc: LTOI1 = The Lost Treasures of Infocom (1992), LTOI2 = volume II (1992), MP = Masterpieces (1996), each split by platform side. LTOI2’s fourteen titles are byte-identical to Masterpieces’ copies, so an MP match on those titles extends to LTOI2 implicitly; only Shogun, which Masterpieces lacks, is labeled LTOI2 directly. A match against MP PC extends transitively to the Zork Anthology (1994) and Zork Legacy Collection (1997) copies of the anthology titles, which are the same bytes.

What matches the discs

Twenty-two of the seventy builds are z-code-identical to a disc build, including every Solid Gold repository: hitchhikersguide-gold is LTOI1’s PC build, planetfall-gold is the Mac build on both hybrid discs, and leathergoddesses-gold is Masterpieces’ Mac build. shogun is byte-identical, full file, to the Lost Treasures II CD’s SHOGUN.ZIP (verified 2026-09-11) - the leak’s lone Shogun build is that disc’s PC build, while the disc’s Mac r292.890314 exists nowhere else. zorkzero holds r296.881019 - the Mac disc build; the PC side’s r393 with its reworked border graphics is not in the leak, consistent with the Masterpieces analysis that the leaked ZIL predates the r393 border rework. journey holds exactly Masterpieces’ PC build (r83.890706), sherlock exactly its Mac sound build (r26.880127), and seastalker’s reg.z3 is exactly Masterpieces’ odd PC-only r16.850603 - the repositories settle where that disc’s Seastalker split came from.

What the discs never shipped

• Post-release masters newer than any disc build: Zork I r119.880429, Zork II r63.860811, Zork III r25.860811, Cutthroats r25.840917, Deadline r28.850129, Sorcerer r18.860904, Moonmist r13.880501, Planetfall r39.880501, Hitchhiker’s r60.861002, Beyond Zork r60.880610, Trinity r15.870628, Bureaucracy r160.880521, Nord and Bert r20.870722, The Witness r23.840925, Ballyhoo r99.861014.
• The Lurking Horror r221.870918, the sound release (129,944 bytes). No cataloged disc ships this story file, but Masterpieces ships LHSOUND.ZIP, whose LURKING.CNV is an XOR patch (129,944 bytes, one byte per output byte; the algorithm is in the package’s own UPDATELH.C) that converts the disc’s r203 to r221. Verified 2026-09-03 by applying the patch to Masterpieces’ r203: the output is byte-identical to this repository’s h1.z3, with the patcher’s own checksums valid. The leak and the disc carry the same r221, one as a file and one as a delta.
• Unreleased games: abyss (James Cameron’s The Abyss, v6 r1.890320), restaurant (The Restaurant at the End of the Universe, v6 r184.890412), and zork-german (a v6 German translation of Zork I, r15.890613 - with built-in hints).
• Development snapshots: Beyond Zork alpha (r1.870412) and beta (r1.870715), Trinity alpha (r1.851202) and beta (r1.860221), a Journey beta (ojourney.zip, r54.890526), and five generations of The Witness (r13 through r23) of which only r22.840924 ever shipped on these discs.
• The samplers: three Infocom Sampler demos (r15.840330, r5.840512, r55.850823) and the three Mini-Zork releases, none of which appear on the compilations.
• Zork I Solid Gold (zork1-gold, v5 r52.871125, with hints): the one Solid Gold edition none of the cataloged discs carries in any form.

Oddities

• The checkpoint repository’s compiled file is not Checkpoint. The repo holds Stu Galley’s unreleased Checkpoint in source form, but its lone COMPILED/spy.z5 (v5 r46.880603, copyright 1988) is full of Praxix, Bergon, Esher and Astrix - it is an early, otherwise unknown v5 prototype of Journey, filed in the wrong game’s repository. No compiled Checkpoint exists in the leak.
• Two Suspects, one byte apart. suspect carries two copies of r18.850222 that differ in exactly one content byte (plus the header checksum) - a hand-patched build, and neither is the r14.841005 the discs ship.
• Two different “release 8” Suspendeds. suspended.z3 is r8.830521 and mac.z3 is r8.840521 - same release number, serials a year apart; only the later one matches the discs. The serial, not the release, is the discriminator.
• wishbringer carries an experimental build (nj2.z3) whose release field reads 32933 with serial 880609 - an EZIP-era test binary that txd cannot disassemble.
• Seastalker is the most-built game in the leak: seven distinct builds, including platform-targeted Atari, TRS-80 CoCo and Tandy variants that the compilations never carried.

A second layer in the git histories

The repositories were not imported as single snapshots. For most classic games the archivist committed an earlier drive state first - the commit is labeled “Revision NN (Original Source)” where NN is a shipped release - and then committed “Final Revision”, the drive’s shutdown state, on top; stationfall even carries “Beta Version” and “Gamma Version” layers, and beyondzork a “Revision 57” between its two. The HEAD trees enumerated above therefore hide a second census: 19 builds that exist only in earlier commits, recoverable with git cat-file at the commit shown.

Twelve of the nineteen are the exact shipped masters, and they close gaps the HEAD census left open: the shipped Zork I/II/III (r88/r48/r17), Beyond Zork r57, The Lurking Horror r203, Moonmist r9, Nord and Bert r19, Planetfall r37, Suspect r14, Trinity r12 and Cutthroats r23 - each z-code-identical to the disc copies - plus Seastalker’s atari.zip in the “Revision 15” layer, which is actually the Mac disc build r15.840522 (the platform labels shifted between layers). The other seven never shipped on the cataloged discs at all: Enchanter r24.851118, Hitchhiker’s r58.851002, Sorcerer r13.851021, Spellbreaker’s first release r63.850916, a second Seastalker r15 (serial 840501), and Stationfall’s beta (r63.870218) and gamma (r87.870326). So in these repositories the layer where source and binary do plausibly correspond is the “Revision NN (Original Source)” commit - shipped source paired with its shipped build - while HEAD pairs the final source state with whatever binaries the drive held last.

Internal codenames

The leak’s internal filenames - the compiled stories and their .serial companions - are Infocom’s project codes, and they are genre-keyed. Observed in the repositories:

Two codes are small revelations: restaurant’s story is h2 - Hitchhiker’s 2, the sequel tracked under its parent’s letter, not a genre - and the unlabeled early mysteries Deadline and The Witness sit where m1 and m2 ought to be, before the convention took hold.

The Mac type/creator codes in Masterpieces’ catalog listing corroborate the scheme from the packaging side: INm5 Moonmist, INZ4 Enchanter, INz6 Spellbreaker, INH1 Lurking Horror, INR1 Plundered Hearts - and extend it where the leak is silent: INZ5 Sorcerer completes the Enchanter trilogy as z4/z5/z6, INM9 makes Sherlock mystery number nine, INZ7 Trinity and INZ8 A Mind Forever Voyaging keep counting the Zork line, and INE1 files Border Zone under espionage.

The two systems also drift, which dates them as different eras of the same idea: Nord and Bert is j3 in the source but creator INC3 on the disc (with Bureaucracy INC2 - a comedy series?), Wishbringer is j2 in the source but INR2 on the disc, Leather Goddesses is x1 in the source but INS7 (science fiction) in the catalog, and by the v6 era the creators are per-game again: INJ1 Journey, INL1 Arthur, IN0Z Zork Zero.

The .serial files do not hold the release numbers

Thirty-two repositories carry a <codename>.serial file next to the source - a bare number. The natural guess is that it tracks the release, so here is the cross-check: every .serial value against the release of the same-codename compiled build in the same repository.

So: no. The deltas run from -397 to +118, in both directions, and witness.serial reads 4 in a repository that preserves six successive release builds - the value tracks neither the release nor the count of kept builds.

The one place the relationship is visible is seastalker, the only repository with several .serial files: all three sit at exactly release minus 15, in sequence (1 -> r16, 2 -> r17, 3 -> r18). Within one directory the counter and the release move in lockstep; across directories the offset is arbitrary. That is consistent with the .serial file being a per-directory counter maintained by the toolchain while the release word tracked the output story file’s TOPS-20 generation number: the two agree only up to whatever generation the directory’s story file started at, which each new working directory reset differently. The near-misses fit the same picture - young single-lineage trees land close (sherlock -1, nhitch -2, mini2 +1), long-lived ones drift far (beyondzork at 457 against r60).

Source-only and non-game repositories

arthur and wishbringer-gold contain ZIL source but no compiled story file, so the leak has no Arthur build at all (Masterpieces’ r54/r74 exist only on disc) and no compiled Solid Gold Wishbringer. infocom-zcode-terps (interpreter source for a dozen platforms) and zil (language documentation) round out the Infocom material; the account’s Zork history repositories (zork-mdl, zork-fortran, zork-1977-source, zork-german aside) predate ZIL and hold no Z-machine files.