2026-08-19 - The C runtime, and 395 imports

Done. Scanned public sources for what the first two passes had missed, then acted on the largest finding.

• New 035-libc section: ten positive behavioural checks against the C runtime.
• 105 more censused symbols; 14 groups, 353 names, 395 imports.
• docs/BACKLOG.md written - the gaps found and deliberately not filled.
• ACKNOWLEDGEMENTS.md position hardened into D018.

Verified. 13 sections, 60 checks. Host run: 27 pass, 5 partial, 20 fail, 8 skip; report well-formed. 035-libc scores 10 of 10 against real glibc, and the census reports libc as partial (71 present) - so all three census states are now demonstrated by a real run rather than argued for.

Surprises.

• The biggest gap was the one that needed no research. Every other library needed looking up; the C runtime needed only remembering, and it was the one thing missed. A freestanding probe was written and the guest was then assumed to be freestanding too. A title imports more of libc than of any single vendor subsystem.

• Declaring standard C names as data does not work - clang knows them as builtins. strcpy as a const char is "redefinition as a different kind of symbol". The D014 trick holds for vendor names and collides head-on with anything the compiler has an opinion about. -fno-builtin on the host build fixes it, and is the honest flag rather than a workaround: the host build stands in for a guest platform, so clang genuinely should not assume it knows what these functions do. The target build already had it, which is why the failure appeared only on the host.

• The community SDK for the current-generation console builds its symbol stubs from decrypted vendor libraries. The names are correct and the chain of custody is not one this repository can adopt under rule 1. That rules out the easy route to the current graphics interface, and the honest answer is that the project does not have those names yet rather than that it has guessed them. (D018, backlog item 1)

• The host build finally produces meaningful passes. Until now every green was a negative check agreeing that a stub fails. The libc section passes because real implementations do the right thing - which also proves the positive checks can tell the difference, something no amount of red ever demonstrated.

Not done. The current generation graphics interface is still absent, and deliberately so. Struct-taking functions remain out of reach. Nothing verifies a NID mapping, which is the fault an emulator is most likely to have and this program least able to see. Twelve more libraries unlisted. Never run on hardware or under an emulator.