Hunt R-2946DE: results

Question. Does scripts/make_context.py --check detect public functions added under meta/, and what is the exact boundary of this blind region?

Answer. It does not detect them. Across all modules and subpackages in meta/, 100% of symbol additions, in-place renames, new modules, new markdown documents, and line-count changes pass scripts/make_context.py --check completely undetected (0 / 16 curated mutants fired, 0.0% detection rate; 0 / 32 exhaustive public AST symbols detected, 0.0% detection rate). The generated CONTEXT.md is 100% byte-identical (0 diff lines, exit code 0) across every mutant under meta/.

The mechanism is architectural exclusion: scripts/make_context.py defines hard-coded scan roots (PKG = ROOT / "zeta", DISCOVERY = ROOT / "ontology", HARNESS = ROOT / "harness", DOSSIER = ROOT / "dossier", DOCS = ROOT / "docs", TESTS = ROOT / "tests", SCRIPTS = ROOT / "scripts"). Neither build() nor build_flat() references meta/ anywhere in its AST extraction pipelines. Consequently, meta/ forms a total blind region.

Positive controls in scanned packages (zeta/, ontology/, harness/, docs/) confirm the guard and probe apparatus have full detection power on in-scope paths (4 / 4 fired, 100.0% caught, diff >= 2 lines).

Reproduce: python3 hunts/r_2946de/probe.py (~45 s, standard library only). Raw data: hunts/r_2946de/results.json.

Controls

Both baseline and undo controls ran and held cleanly across the entire run:

The curated battery

diff is the number of unified diff lines in regenerated CONTEXT.md. token is whether the mutant's new symbol appears in that diff.

Category 1: Public function additions in meta/

0 / 3 detected (100% blind).

Category 2: In-place public symbol renames in meta/

0 / 3 detected (100% blind).

Category 3: Public classes and constants added to meta/

0 / 3 detected (100% blind).

Category 4: New modules and documentation files in meta/

0 / 4 detected (100% blind).

Category 5: Line counts, docstrings, and formatting in meta/

0 / 3 detected (100% blind).

Category 6: In-scope positive controls in scanned directories

4 / 4 detected (100% caught).

Repository census and exhaustive mutation analysis

A full static AST census of all Python source files under meta/ (meta/ledger.py, meta/evals/asymmetry.py, meta/evals/__init__.py) identifies:

1. 32 public top-level symbols in meta/:
2. 10 classes (Category, Automatability, CaughtBy, Intervention, MetaObservation, OperatorFunction, Resolution, Party, Judgment, Ledger in meta/ledger.py)
3. 16 functions (including load_ledger, validate_ledger, asymmetry_dataset, checker_verdict, detection_rate, false_confidence_rate, asymmetry_e0)
4. 6 module-level uppercase constants (including LEDGER_PATH)
5. Exhaustive symbol mutation results:
6. Every single one of the 32 public symbols was systematically mutated in-place against the sandbox baseline.
7. Result: 0 / 32 fired (0.0% detection rate, 100% blind).
8. Diff lines in CONTEXT.md: 0 lines across all 32 runs.

1. Historical analysis in git history:
2. meta/ was introduced in commit 5a666c0 ("meta: the second laboratory").
3. Across all 14 subsequent commits touching meta/, zero commits triggered or required changes to CONTEXT.md.
4. scripts/make_context.py has never contained references to meta/ at any commit in repository history.

What the numbers say, in order of how much they matter

1. The blind region is total and structural. scripts/make_context.py does not omit meta/ by filter logic or __all__ parsing: meta/ is completely omitted from the directory list make_context.py reads. No symbol, class, function, docstring, file, or line-count change under meta/ will ever trip scripts/make_context.py --check.
2. The line-count tell does not apply to meta/. In scanned packages (zeta/, ontology/, harness/, dossier/), adding a line trips the guard even if the symbol is unexported. In meta/, adding blank lines or modifying file lengths (M14) produces zero diff because CONTEXT.md tracks line counts only for scanned modules.
3. CONTEXT_FLAT.md likewise excludes meta/. In build_flat(), meta/ is not listed among the concatenated source groups (Operating context, Package, Discovery layer, Harness, Dossiers, Scripts, Docs, Tests). Thus, flat context dumps also omit meta/.
4. Architectural reason: meta/ is the second laboratory. Per meta/README.md, meta/ holds evidence about the laboratory and research process, not about the mathematical subject. CONTEXT.md was created to index the public mathematical and computational APIs of the laboratory (zeta/, ontology/, harness/, dossier/, docs/, scripts/, tests/).

What I chose, and why

• Isolated sandbox execution: Mutating a throwaway copy under /tmp prevented any dirty working tree state or interference with concurrent sessions.
• Combined curated battery and exhaustive symbol census: The 16 curated mutants test functional additions, renames, classes, constants, docs, and line counts, while the 32-symbol census exhaustively covers every public definition currently in meta/.
• Positive controls: Evaluating 4 mutants across zeta/, ontology/, harness/, and docs/ proved that the probe and sandbox apparatus reliably detect in-scope modifications.

What I could not settle

• Whether meta/ should remain unindexed or be added to CONTEXT.md: meta/ holds operator tooling and intervention ledgers. Whether CONTEXT.md should expand to include a "Meta-research layer (meta/)" section is an architectural decision for repository maintainers. Leaving it unscanned keeps CONTEXT.md focused on mathematical research; adding it would bring meta/ under the stale-index guard.

Loose threads

• compiler/ is also completely unscanned (known miss B02): Like meta/, compiler/ is a separate package in the repository root that is not referenced in scripts/make_context.py.
• Why it might matter: Changes to public functions in compiler/ (e.g. compiler/semantics.py) are similarly invisible to make_context.py --check.
• First step: Run a sibling hunt measuring compiler/ sensitivity and settle whether compiler/ should be indexed.
• Unscanned test utilities: Test helper files not matching test_*.py (e.g. tests/mutant_helper.py, known miss B04) are ignored by test_counts() which globs test_*.py.
• Why it might matter: Adding shared test infrastructure does not update the test function count in CONTEXT.md.
• First step: Evaluate whether test_counts() should glob *.py in tests/ or maintain the test_*.py pattern.