test: dead-code census test asserts the tree still holds non-parsing specs -- the population hit 0 #7088

Defect

bootstrap/tests/a_ratio_names_its_denominator.rs::the_dead_code_census_names_what_it_skipped asserts skipped > 0 because "this tree is known to hold specs that do not parse; a 0 means the counter never fires" (#3077 measured 666 walked / 76 did-not-parse).

The spec-fix waves finished their work. On master 96cf7c8da:

Specs walked:     1363
  did not parse:  0
Specs counted:    1363

So the assertion is now factually wrong on master and the test fails (cargo test --release -p t27c --test a_ratio_names_its_denominator, verified on the Railway lab). It did not gate recent PRs only because the queued run had not completed during the Actions outage; the next Bootstrap Test Ratchet run on master goes red.

Fix (the repo's own pattern)

The intent is "the counter is wired", which a live-tree assertion cannot prove once the tree is clean. Give the test its negative control, like damage-negatives: a scratch tree holding one spec that does not parse and one that does; assert did not parse: 1. The scratch pattern already exists in this file (the backlog tests).

Rust test edit, existing file: owner-approved-foreign (standing rule 2026-10-06, self-labeled) + tools/policy/foreign-exceptions.txt entry.

Refs #6655 (found while verifying #7076 on the lab)

🤖 Generated with Claude Code