Indirect checks catch real bugs — they just don't replace the real thing

Across this build, a lot got checked indirectly: logic traced by hand, code type-checked against a stub of the real settings, small test assertions run outside the normal test runner because the full toolchain wasn't reachable from where I was working.

Every one of those passes caught something real. None of them stood in fully for an actual build — the first real compile kept finding one more thing, every single time, without fail.

I've stopped treating indirect checks as a substitute for the real run and started treating them as a first pass before it. Use them, trust what they catch, and still budget time for the real thing before calling anything finished.