Function unasserted_codes 

pub fn unasserted_codes(root: &Path) -> Vec<&'static str>

#1662 (track #1648, G2). Registry codes that some test asserts, out of all bynk_syntax::diagnostics::REGISTRY codes, and the count no test asserts.

Static, so it runs inside this harness: the review that set the baseline (2026-10-01, #1647 Part 4) instrumented CompileError::new across a full cargo test --workspace, which this harness cannot afford. Asserted is a subset of produced (a passing test that names a code saw it), so driving the unasserted count to the argued floor meets “every reachable code is produced by a test” a fortiori.

A code counts as asserted when it appears in:

• line 1 of a negative fixture’s expected_error.txt (line 2 is a message substring and may quote other codes);
• any other non-.bynk file under a crate’s tests/ directory (test sources, expected-diagnostics files, JSON goldens). .bynk sources are skipped, because their comments often name the code they provoke, and Rust sources are read with their // comments removed, for the same reason;
• a #[cfg(test)] mod block in a crate’s src/, comments removed;
• a blessed diagnostic transcript, site/src/diagnostics/*.txt.

Gated at the argued floor of 4 (#1662 Decision B). Each of the four is emitted where no compiler test can reach it:

• bynk.deploy.contract_skew: bynk deploy, against a live deployment’s lock;
• bynk.project.read_failed: only when a host’s file overlay omits a discovered file, which neither the CLI nor the LSP does;
• bynk.target.vendor_conflict: needs platform-native capabilities from two platforms, and only Cloudflare ships any today (the decision function is unit-tested);
• bynk.wasm.strip_failed: stripping the compiler’s own emitted TypeScript, which fails only on an emitter bug.

cargo xtask greenfield-status --list-unasserted prints the codes.