The canonical CD-i platform checklist, carried from one documentation pipeline to the next and added to by each.

→ cdi-platform-notes.md

Each Philips CD-i title I document produces two things: a repository about that disc, and whatever it taught me about the format. The second kind of finding does not belong to any one title, and keeping a copy of it in every pipeline is a recipe for three copies that disagree. This repository is the single copy. Pipelines link here rather than fork it.

A working checklist for opening an unfamiliar CD-i disc: what to look at first, what the numbers should be, which traps cost real time, and what is measured versus what is inferred. It is written to be read in order the first time and grepped afterwards.

Findings confirmed on every disc so far are marked [all]; those confirmed on fewer are marked [N of 5] (older marks read [N of 4] and predate the fifth disc). Everything else is named after the disc it came from, and is the kind of thing to test rather than assume.

Every disc below has its own repository, and the full write-up for each one lives in the family index: cdi-gamelist-doc. The table here stays the short form; the index is where the prose is.

The five bracket the format: one disc where the program owns its assets, one where it owns nothing and streams every pixel, one hybrid that streams its level code as well as its pictures, one that owns everything but keeps it inside a single 8 MB container it did not write itself, and one that is not a data disc at all.

Sectors in front of the file system belong to no file, and on all five discs they resolve to real recorded audio. On three of them — different developers, different publishers, different continents, and now two years apart — that audio is byte-identical:

Link     1993   sectors 19-2268   5,229,000 B   md5 a0ed87f2e98b43f91281d16390fb178b
Origami  1993   sectors 18-2267   5,229,000 B   md5 a0ed87f2e98b43f91281d16390fb178b
Merlin   1995   sectors 19-2268   5,229,000 B   md5 a0ed87f2e98b43f91281d16390fb178b

29.6 seconds at 44,100 Hz, two takes, a fundamental drifting around 170 Hz. It is an artefact of the CD-i authoring chain, not title content, and it was still being written onto masters in January 1995. The 1997 disc carries a different recording in the same place, and The Apprentice, in 1994, carries a third (4e61f608e1f1455d9ad5b2a0615dbbd3) — so date does not predict which one you get, and it writes the block twice, head and tail.

No pipeline could have found that alone. Hashing the head region of a new disc against those MD5s is a thirty-second check, it has now answered the question outright four times, and it is exactly the kind of thing that has to live in one place to be useful.

A second check of the same shape has since turned up: the Philips bumper that opens a licensed disc is a finished asset the publisher handed out, and it is pressed verbatim. 714,240 bytes of it — RL7 video, Level B stereo audio and one 384 x 240 DYUV picture — are byte-identical between a 1993 disc made in the USA and a 1994 disc made in the Netherlands. Section 5b has the hashes.

Underneath it, the mechanism: the authoring system wrote sector headers over a CD-DA stream, destroying seven stereo frames — 28 bytes — per sector (sync + header + subheader, plus four bytes zeroed at the EDC position). Section 2 has the measurement.

When a new title turns up something about the format rather than the title:

1. Add it to the relevant section here rather than to the title's repository.
2. Mark it [all] or [2 of N] only if you have actually checked the other discs. Otherwise name the disc it came from.
3. If it contradicts what is already written, correct the text and say so in place — there are several such corrections in sections 2, 5 and 7a, and all of them are more useful with the history attached than without.
4. Update the baseline table in section 10 and the order of work in section 11.

State what is measured and what is inferred, and keep the measurements in the document. Half of what makes a disc interesting is the list of things that are measurably odd and not yet explained.