D034 - Two package entries were derived rather than cited, and there is a command that re-derives them

D033 said the six unknown entries all vary between titles and concluded the block was firm. That was the right measurement and the wrong conclusion drawn from it. "Varies between titles" does not mean "cannot be computed" - it means "is not a constant", and the two are different.

Challenged on it, the honest position turned out to be that this project already derives format meaning from material where the derivation is arithmetic a machine can check: obSCEne's vendor tag assignment came from adjacencies like RELA + RELASZ == HASH holding across every module, and D008 settled a current-generation tag by counting. Principle 5 forbids inventing a meaning nothing tests. It does not forbid stating a falsifiable hypothesis and running it.

Two survived every sample:

• Entry 0x1 is a digest table. One SHA-256 per entry, in entry-table order, sized to exactly 32 bytes per entry - 448 bytes for a fourteen-entry package and 736 for a twenty-three. Its own slot is all zero, because an entry cannot contain its own digest.
• Entry 0x100 is the entry table again. Id at 0x00, offset at 0x10, size at 0x14, big-endian - the same three fields at the same three offsets the outer table uses.

Both computable, which is why they were worth chasing. A writer produces them from data it already holds, and derive::digest_table and derive::entry_table_copy do exactly that.

Four remain unknown - 0x80, 0x400, 0x401, 0x1002 - and are not guessed at. 0x80 is partly digest-shaped (one of its twelve slots is the digest of param.sfo); the other three are dense and high-entropy in every sample.