Hunt R-C62E44: results

Status: settled. The guard tests/test_docs_numbering.py::test_no_two_docs_share_a_number does not detect a document renamed without its number changing when citations use bare references (such as docs/08), nor does it detect document content drift, title changes, or slug-title divergence. Sibling test test_every_full_filename_reference_to_a_doc_resolves detects a rename only if full-filename references remain in the repository. An exhaustive census across all 37 test files mentioning docs/ confirms that zero tests in the suite inspect document body content, validate H1 headings against filenames, or verify citation semantic validity.

Reproduce: python hunts/r_c62e44/probe.py (~5 s, standard library and existing tree modules only). Data: results.json.

Controls

Both controls ran and held:

All mutations were executed against an isolated temporary directory sandbox, leaving the live repository untouched.

The 20-mutant battery

Summary of findings

1. The primary guard is strictly a collision detector. test_no_two_docs_share_a_number groups filenames matching ^(\d{2})-[\w.-]+\.md$ by their leading two digits and asserts len(filenames) <= 1. It fires on M01 and M09. On any rename where the leading number is preserved (M02, M04, M07), it is silent by construction.
2. The boundary between bare and full-filename references is sharp. test_every_full_filename_reference_to_a_doc_resolves catches a rename only when existing citations use the full filename (M03). Because AGENTS.md explicitly mandates bare citations (docs/08 rather than full names), most repository references are bare. Bare references resolve as long as any file with that number exists, rendering renames transparent to the entire test suite (M02, M04).
3. No test reads document content. Replacing the entire body of a document (M05) or altering its H1 heading (M06) passes all four numbering guards. scripts/make_context.py --check detects line count changes or heading changes only until CONTEXT.md is regenerated: it validates that the index matches the files, not that the content matches the title or that citations are semantically coherent.
4. Format sensitivity. The regex ^(\d{2})-[\w.-]+\.md$ strictly requires exactly two digits and a hyphen. Files named with 3 digits (M10), 1 digit (M11), or underscores (M19) are ignored by _numbered_docs(), which triggers the gap guard test_the_course_has_no_gap_that_hides_a_lost_document and the reference resolution guards.

What I chose and why

1. Direct execution of test functions in sandbox rather than external pytest spawning. Importing the test module directly and parameterizing REPO_ROOT and DOCS allowed running all 20 mutants and controls in ~5 seconds with exact assertion verification.
2. 20 mutants across 8 distinct categories. Testing only the declared miss would confirm a single boolean. The 20-mutant battery maps the complete interaction matrix between the four numbering guards and scripts/make_context.py --check.
3. Exhaustive census of the test suite. Grepping and inspecting all 37 test files referencing docs/ proved that the absence of content validation is structural and tree-wide, not an isolated property of test_docs_numbering.py.

What I could not settle

1. Whether semantic content validation is desirable in CI. Automated checking of whether document contents match their titles or citations requires natural language understanding or structured metadata schemas. Whether to enforce title-slug equality (08-why-it-is-hard.md having # 08: Why it is hard) is an editorial design decision.
2. Historical frequency of unnoticed content drift. I did not reconstruct past git history to count instances where document bodies were rewritten while bare references in other files went unreviewed.

Loose threads

1. Title-slug decoupling is unconstrained. A document named 08-why-it-is-hard.md can have H1 # 08: Topological Field Theory without failing any guard. Why it might matter: Readers and agents navigating via docs/NN rely on title and slug describing the same subject. First step: Add a check in test_docs_numbering.py verifying that words in the filename slug appear in the H1 heading.
2. Subdirectories in docs/ are completely invisible to numbering guards. docs/doors/ and docs/reviews/ contain markdown files that bypass test_docs_numbering.py because _numbered_docs() uses os.listdir(DOCS) without recursion. Why it might matter: Numbered docs placed inside a subdirectory would be ignored rather than validated. First step: Assert in test_docs_numbering.py that no numbered markdown files exist in subdirectories of docs/.
3. Regex pattern ^(\d{2})-[\w.-]+\.md$ allows non-alphanumeric trailing characters. A file named 08-foo..md or 08-foo.bar.md matches \w.-. Why it might matter: Inconsistent file naming across contributors. First step: Constrain the slug regex to ^(\d{2})-[a-z0-9-]+\.md$.