docs(applied): record the 96cfea7 corpus-sweep discharge; add braid-mode smoke (#300)

[tsondru]

Closes #300.

The 100k-pair published-figure corpus (catgraph-applied/tests/smc_nf_differential_sweep.rs + tests/content_equality_corpus.rs) is guarded day-to-day by a 5k default-mode smoke; the issue's phase-1 probe cited an undated run at 96cfea7 claiming all five --ignored sweeps passed, with no durable record of it in the repo.

Change

• Record. Rather than transcribe the 96cfea7 claim (unverifiable — I never ran the tests at that commit), added a dated module-doc section to smc_nf_differential_sweep.rs recording a fresh, independently-verified pass of the same five sweeps at this branch's base (89241cc, 2026-08-31): published_divergence_figures_reproduce, published_braid_mode_figures_reproduce, published_interleave_mode_figures_reproduce (this file) plus published_corpus_is_closed_by_content, braid_mode_corpus_is_closed_by_content (content_equality_corpus.rs) — 5/5 pass in one combined --ignored run, 15.13s. The 96cfea7 claim is cited as prior context, not asserted as independently checked.
• Optional smoke prefix. Added braid_mode_smoke_prefix_of_the_published_corpus — a non-ignored 5k-pair Mode::Braid prefix, pinning Counts { divergent: 77, in_fragment: 38, marked: 17 } (discovered by running the code, not estimated), mirroring the existing Mode::Default smoke test. Closes the issue's stated gap: the §4.6 Counts divergence-bucket figures for braid mode were previously reachable only via the --ignored 100k tier.

Review (2 rounds, rust-v2:falsifying-reviewer, perturb-in-copy)

• Round 1 (base 89241cc → 1d92c6a, record + smoke test): no blocking/important. One minor: the new smoke test overlaps with a pre-existing, already-CI-visible braid-mode test — canonical_display_corpus.rs's smoke_prefix_braid_corpus (predates this branch, same seed, non-ignored, in ci.yml) — which tracks a different statistic (Gate readback/churn) than this file's Counts divergence bucket. Not a defect (the new test's Counts coverage is genuinely new), but the doc comment's framing needed to disclose the overlap rather than imply this is the first braid-mode CI check. Confirmed the record's own claims check out: re-running the cited command reproduced 5/5 passes (14.96s, consistent with the doc's 15.13s), and both content_equality_corpus.rs test names it cites are real. Confirmed the new smoke test is non-vacuous (perturbing column_pair_is_admissible's has_braid guard reddens it, Counts{66,36,17} vs expected {77,38,17}) and not redundant with the default-mode smoke test (same perturbation leaves that one green — Mode::Default never carries braid atoms). Applied in 99acb53.
• Round 2 (1d92c6a→99acb53 delta): no findings — comment-only diff verified, every clause of the disclosure checked against the code (smoke_prefix_braid_corpus's location, non-ignored status, seed, and Gate's disjoint field set vs Counts) and against git history (smoke_prefix_braid_corpus traced to commit 2268dc0, smc_nf: canonicalize η placement (ι) — the single remaining canonicality freedom on braid-free diagrams #187, well before this branch). Converged.

Every perturbation round ran in a throwaway cp -r copy with its own --target-dir; the worktree stayed clean throughout.

Gates (worktree, --target-dir /tmp/cargo/catgraph-G5-T4)

• cargo nextest run -p catgraph-applied: 669 passed, 0 failed, 12 skipped (default features)
• cargo nextest run -p catgraph-applied --features internal-probes: 686 passed, 0 failed, 20 skipped — includes the new braid_mode_smoke_prefix_of_the_published_corpus
• cargo clippy --all-targets -- -D warnings: clean on default, --no-default-features, --features internal-probes
• RUSTDOCFLAGS="-D warnings" cargo doc -p catgraph-applied --no-deps: clean
• cargo fmt --all --check: clean
• scripts/check_measured_claims.py / scripts/check_audit_counts.py: not applicable — diff carries no MEASURED markers and touches no audit doc

🤖 Generated with Claude Code