D050 - Revised starting order after real material arrived

decided · 2026-08-19

Phase 0 shrinks and stops gating phase 1. Its rationale was that no real container could ever live in this repo, so one had to be synthesised. Real containers now exist on disk outside the repo, exactly as intended - so synthetic fixtures are still needed for malformed cases a compiler will never emit (truncation, offsets past EOF, absurd counts), but the parser can be developed against genuine input.

obSCEne is not needed before phase 4, and two findings strengthen that:

• Its main early justification - being our first real container - has evaporated.
• The open toolchain is previous-generation, so it would emit 4f153d1d-wrapped containers: the format we are not parsing first.

Homebrew (third-party corpus, phase 0d) and obSCEne (our own test app) are separate things and it is easy to conflate them. Starting with homebrew does not require obSCEne to exist.

Recommended order, all of it unblocked:

1. 0c - structural seams. Before more code exists; cheap now, expensive later.
2. 0e - observability substrate. So later phases feed the run report rather than retrofitting it.
3. 1 - the container parser, against real material, starting with the 76 KB module rather than a 68 MB executable. Same format, same wrapper, trivially inspectable.
4. 0 alongside, reduced to malformed-case fixtures.

0b any time; 0d whenever homebrew is wanted in the mix.