M2.5: wit logic-report tool for the empty-verdict reality gate

[sepehrsadri]

Progress on #15.

What this PR delivers

Issue #15 asks for the empty-verdict rate across a 30-project / 26 GB real
Logic library. This environment does not have that library — it has exactly
one real .logicx project. Fabricating a "the answer is X%" finding from n=1
would violate this repo's own reproducibility doctrine (AGENTS.md), so this
PR delivers the tool the issue needs, runs it for real on the one project
available, and publishes that honestly-labeled n=1 result — it does not
close #15, because the actual reality-gate finding still requires the real
library.

New: wit logic-report <library-root>

$ wit logic-report "/path/to/YourLibrary"
  scanned 1 project(s), 1 alternative(s), 9 consecutive save pair(s)
  44.4% of save pairs show a structural change Wit can see (4 of 9)
  distribution of change counts per save pair (0 = no visible structural change):
    0 change(s): 5 pair(s)
    1 change(s): 1 pair(s)
    2 change(s): 1 pair(s)
    14 change(s): 1 pair(s)
    17 change(s): 1 pair(s)
  5 pair(s) (55.6%) are byte-different but structurally identical

Walks every discovered Logic/GarageBand alternative's backup chain
(Project File Backups/00–09 oldest-first, then current — reusing M3's
wit_index::discover_logic_projects), compares every consecutive pair at
wit-logic's Structure honesty tier, and reports the three statistics #15
asks for: % of saves with any visible structural change, the distribution of
change counts per save, and how often adjacent saves are byte-different but
structurally identical.

New library pieces

• wit_logic::change_count(a, b) (crates/wit-logic/src/lib.rs) — a new
  diagnostic counting distinct differing signals between two walks (one per
  differing census tag, one per name added/removed across the three
  extracted-name lists, one for a tempo change). Guaranteed 0 exactly when
  semantic_equal reports NoStructuralChange — not part of the Verdict
  itself, just granularity for the "how much changed" question M2.5: Measure the empty-verdict rate on a real Logic library #15 asks.
• wit_index::logic_report / LogicLibraryReport
  (crates/wit-index/src/report.rs) — orchestrates discovery + wit-logic
  walks/comparisons across a whole library, the same way scan.rs already
  orchestrates discovery + the store. Read-only, following discover.rs's
  and dupes.rs's existing discipline.
• wit-cli's LogicReport subcommand follows scan/dupes's privacy
  convention: only project/alternative names in output, never a full path
  (assert_no_home_paths asserted on the rendered report, same as dupes).

Tests

• Synthetic-fixture unit tests in wit-index/src/report.rs (deterministic,
  crafted ProjectData containers, no real material) proving the three
  statistics compute correctly, plus a read-error and an empty-library case.
• wit-logic/src/lib.rs gains unit tests for change_count (zero iff
  NoStructuralChange, symmetric, counts exactly one per differing tag).
• An opt-in real-material test, crates/wit-index/tests/real_fixtures.rs,
  gated on WIT_LOGIC_LIBRARY (mirrors the existing WIT_LOGIC_PROJECT /
  WIT_FIXTURES loud-skip pattern), for a library root rather than a
  single bundle.

Docs

• docs/EXPERIMENTS.md §11 — the n=1 result above, explicitly labeled
  measured and explicitly scoped: this is a data point, not the issue's
  own >50%-of-30-projects decision.
• docs/ROADMAP.md's M2.5 row — tool landed, finding still open, with the
  exact command to run against the real library.

For the repo owner: run this against the real library

WIT_LOGIC_LIBRARY="/path/to/YourLibrary" cargo test -p wit-index --test real_fixtures -- --nocapture --ignored

or directly via the CLI:

cargo run -p wit-cli -- logic-report "/path/to/YourLibrary"

That run is what actually resolves #15's exit criterion (M5 vs. the M2
stretch goal of mapping Logic's volume-fader field) — this PR only gets the
tool in place for it.

Verification

• cargo build --workspace — clean
• cargo test --workspace --locked — clean, all green (opt-in real-material
  tests loudly skip as expected)
• cargo fmt --all --check — clean
• cargo clippy --workspace --all-targets --locked -- -D warnings — clean
• cargo deny check licenses advisories — clean (pre-existing unrelated
  license-not-encountered warning for ISC, not an error)
• python3 .github/workflows/scripts/check_personal_paths.py — clean

🤖 Generated with Claude Code