A measurement of two CD-ROMs: the European PC edition of Resident Evil 2, published in 1999 by Virgin Interactive from a game CAPCOM made for the PlayStation in 1998.
The objects are RES2 LEON.bin, 758,583,504 bytes, and RES2 CLAIRE.bin,
722,626,128 bytes — the discs rather than copies of their files: 322,527 and
307,239 physical sectors of 2,352 with no remainder, sync patterns and
Reed-Solomon parity included. Neither was ever mounted. Everything here was read
out of them sector by sector, and every figure carries the command that
re-derives it.
This is the thirtieth object in this collection and the first that is two. That is not a formatting inconvenience: it creates six defensible denominators where every earlier object needed two, and the headline figure of the repository moves by twenty-six points depending which is used. Every table below names which.
The discs are not in this repository and never will be.
| images | 758,583,504 + 722,626,128 = 1,481,209,632 bytes; sha1 0f27e8f5…fb68 and 5deeb74d…ea3a |
| geometry | 322,527 + 307,239 = 629,766 sectors of 2,352, Mode 2 Form 1, remainder zero on both |
| frame, censused whole | sync 629,766/629,766 · EDC 629,766/629,766 · Form 2 sectors 0 · 0 mismatches of any kind |
| frame cost | 12.9252 % on both — which is 304 ÷ 2,352, proved last session and cited here |
| Mode 2 without XA | subheader 00 00 00 00 on 629,766 of 629,766; no CD-XA001; 0 of 4,490 records with an XA System Use area |
| filesystems | two ISO 9660 volumes, primary + Joliet on each, 36 directories, 2,194 / 2,224 files, 657,241,774 / 625,887,450 bytes |
| namespaces | 0 extents in one namespace only, delta 0 bytes, 20 names of 2,194 spelled differently — all of them third-party driver folders |
| descriptors | CeQuadrat 32bit ISO-9660 Formatter … 1995; both burned 1999-03-21, one hour two minutes apart; timezone bytes −66 and +113, outside ECMA-119's legal range |
| unclaimed space | 304 sectors in 2 runs on each disc and both runs contain something — the first non-empty holes in thirty objects, both signed by the burner |
| sector accounting | closes with residue 0 on both discs |
| the two discs | 1,696 shared hashes, 357,785,576 bytes = 54.4374 % of Leon and 57.1645 % of Claire |
| media | 32 films, IV50 320×160, 52,780 frames, 2,532.5314 s = 42.21 min, all named .BIN · 1,495 .SAP banks, 5,195 ADPCM streams, 18,450.146848075 s, closed to nine decimals |
| the port | 1,445 PlayStation TIM blocks validated, 10 TMD against 1997's 828, and a CAPCOM model container derived on 169 files of 169 |
| dates | 1,446 / 1,478 distinct timestamps over 835 days, 0 at midnight, and Tokyo's nine hours visible on 9 binaries across two discs |
| crossings | 222 — of which 12 with Resident Evil and 221 with Final Fantasy VIII, and every single one is Microsoft's |
| leftovers | 11,490,859 bytes, 1.2418 % — plus 357,785,576 duplicated bytes that are argued and not counted |
| slack | 4,624,504 bytes across both, 100.0000 % zero, 0 dirty files of 4,418 |
| personal data | 0 identifiers of this machine, 0 personal addresses, one false positive counted inside a TIM |
| predictions | 70 clauses written before either image was opened: 36 hit, 29 half, 5 miss = 50.5 of 70 = 72.1 % |
More than half of each disc is the other disc. 1,696 shared content hashes,
357,785,576 bytes, 54.4374 % of Leon and 57.1645 % of Claire, and four of the
seven layers — COMMON, ZMOVIE, GALLERY, DRIVERS, 1,142 files —
byte-identical between them with zero differing and zero present on one side
only. Outside the scenario directories exactly three files differ, and the
biggest of them is the game itself: same 1,001,984 bytes on both discs, 809,643
of them different, two compilations twenty-six days apart, one carrying
http://www.virgin.com and the other http://www.vie.co.uk. The duplication is
not waste — it is what a two-CD game does so the player never swaps discs
mid-scenario — and it is argued rather than asserted, with the reference
structure of the four duplicated layers as the evidence.
These are Mode 2 discs and nothing on them uses Mode 2. Both cue sheets say
MODE2/2352 and both tell the truth: mode byte 2 and Form 1 on 100 % of 629,766
sectors, 629,766 EDCs verified, zero errors. But the eight bytes that make Mode 2
Mode 2 — the subheader, whose whole purpose is to say what a sector is — are
00 00 00 00 00 00 00 00 on every sector of both discs, submode 0, without
even the data bit. Both modes cost the same 304 bytes a sector, so the choice
was free; the object writes, into the field Mode 2 reserves for meaning, exactly
the bytes Mode 1 reserves for nothing. A Mode 1 disc in a Mode 2 wrapper, and
the only consequence is that the payload starts eight bytes later — which is what
killed a tool written twenty-four hours earlier.
The burner signed the holes, and the volume descriptor counts them without
describing them. For twenty-nine objects the answer to what is in the space the
filesystem does not claim was all zero. Here it is two structures, both
derived completely: sector 19 holds CeQuadrat Joliet directory link table, a
count of 37, and 37 pairs of dwords mapping every Joliet directory extent to
its primary counterpart — all 37 of 37 resolved; and the last sector of the
volume space holds CeQuadrat ISO 9660 formatter information block, its own
address written both-endian in ISO 9660's own convention, and AA 55 55 AA.
The sector accounting closes on them with residue 0. So the collection's oldest
question gets its sharpest answer in eighteen: this descriptor does not merely
stop short of the lead-out, it counts two sectors into its own volume size that
its own author wrote and provides no way to reach them.
Last session's proof about dates does not work here, and finding out why is
the result. The 1997 disc proved its dates came from a FAT source by a
contrast — every file record's seconds even, the mastering program's directory
records even only by chance. Here both populations are even, 4,418 of 4,418 and
72 of 72, so the contrast is gone. The replacement test settles it in the other
direction: all 36 directory records on each disc are dated the burn day,
inside a five-minute window, so they are CeQuadrat's own work — and Claire's
volume descriptor carries an odd second, 14:35:11.95, proving CeQuadrat's
clock is not restricted to even ones. The program has two code paths at two
resolutions, the parity of the file dates is fully explained by the writer, and
last year's conclusion is not inherited. What survives without needing parity:
835 days of development visible in a directory table, and Tokyo's nine hours on
nine binaries — including, on Claire, the game executable that the 1997 session
could not explain, because Leon's copy is the only file of 2,194 whose internal
clock is newer than its own directory record.
Twelve files in common with its own predecessor, and 221 with Final Fantasy
VIII. Swept against 58 repositories and 28,382 hash tokens, these discs cross
222 times and every crossing is Microsoft's DirectX. Resident Evil 2 shares
352,267 bytes with Resident Evil and 10,537,993 with a Squaresoft role-playing
game — thirty times more, by the measurement this collection's Saga column was
defined on. A rule whose honest output is Resident Evil 2 belongs to the Final
Fantasy saga is measuring the wrong thing, so the rule was rewritten and both
Resident Evil rows filled. Kinship measured structurally runs the other way:
seven traits shared with the 1997 disc and zero with Final Fantasy VIII,
including one that byte-identity is constitutionally unable to see — both games
carry exactly one .ESP file, both call it CORE00.ESP, both put it in the
engine's data directory, and the two files share no bytes at all.
| 00 — The predictions | seventy bets, written before a byte of either image was read |
| 01 — What these objects are | two discs, one product, and more than half of each is the other |
| 02 — The spec sheet | every number with the command that makes it again |
| 03 — The frame | Mode 2 on 629,766 sectors, and not one of them uses Mode 2 for anything |
| 04 — The descriptors and the signed holes | a burner that wrote its name into space its own descriptor does not describe |
| 05 — The dates and the nine hours | why last year's proof fails, and the executable that arrived after its own disc |
| 06 — The two discs | 54.4374 %, and the measurement that exists nowhere else in this collection |
| 07 — The port | the texture did not move and the model did, and 169 files of 169 say how |
| 08 — The pictures | five archives that close on the byte, and one compressor that would not open |
| 09 — The film | forty-two minutes of Indeo 5 called .BIN, counted and not decoded |
| 10 — The voice and the music | eight bytes that are a bit mask, and five hours of four-bit speech |
| 11 — The layers and the prerequisites | a codec is not a leftover, and the bill for being a PC went up |
| 12 — The paths and the binaries | zero build machines in 1.48 GB, and two executables of the same size |
| 13 — The crossings and the Saga | twelve files with its own predecessor, 221 with Final Fantasy VIII |
| 14 — The leftovers | 1.2418 %, and the 357,785,576 bytes that are not one |
| 15 — The thesis | 75.3264 % or 68.9810 %, depending on which of six denominators |
| 16 — The tools | 220 in the box, 5 written today, 3 repaired, one retired and un-retired in a day |
| 17 — Corrections | eight errors in the briefing, six of mine, four tools that would have lied |
| 18 — Open questions | what closed, what half-closed, and the four things two discs would not say |
| 19 — The scoring | 36 hit, 29 half, 5 miss, and what the twenty-nine halves have in common |
notes/ holds the raw output of the tools that produced the figures, and
tools/ holds the 220 programs, of which 5 were written for this object and 3
were repaired before they could be believed.
- the discs are never committed, and neither is anything decoded from
them — no audio, no video, no image, no model, no frame of any film and no
illustration from
GALLERY. They are measured, counted and described, and they stay in an ignored working directory; - nothing was executed. Not the game, not either installer, not a driver, and not the Indeo codec — which is why the film is counted from its container and its pixels are never touched. The images were never mounted; they were read as sectors;
- no network was contacted. The nine URLs these discs contain are reported as strings;
- the predictions were written first, with
Write, before any tool was pointed at either object, and they are not edited afterwards; - every figure names its layer, its disc and its denominator. There are six of the last — 657,241,774 · 625,887,450 · 1,283,129,224 · 925,343,648 · 916,460,530 · 1,481,209,632 — and mixing them silently would move the repository's headline figure by twenty-six points;
- where a public definition was used it says so: ISO 9660 and Joliet,
ECMA-130, CD-ROM XA, RIFF/AVI and RIFF/WAVE, Microsoft ADPCM, PE and NE, and
the PlayStation's TIM, TMD and SEQ formats. Everything else was derived from
the bytes or declared underived with the failed attempts written down —
which here means the
.ADTbackground compressor, the interior of the model container, and half of the.RDTroom record; - validate before census, always. The ADPCM decoder was checked block by block against a scalar reference before it was allowed near 5,195 streams, and the raw-sector census verified its own arithmetic on 64 sectors before reading 629,766;
- a tool that finds nothing is not a tool that says zero. Two inherited tools produced confident, well-formatted, entirely meaningless output on this object before they were repaired, and both repairs added an assertion that now kills the tool instead.