2026-10-22 (rs-vec-no-capacity/rs-collect-vec-hot redesign → corpus-ready)

rs-vec-no-capacity (hint): Vec/String::new / vec![] then push in a size-hinted loop without with_capacity.
rs-collect-vec-hot (hint): collect::<Vec…> inside a real loop.
Reverse-opt: #[cfg(test)]/#[test] skip; tests.rs; derive-crate skip for rs-vec-no-capacity; rs-format-rebind-hot brace-wrap skip; rs-string-plus-in-loop mid_string/sc_names skip.
Fixtures: rs-vec-no-capacity_*, rs-collect-vec-hot_*, rs-format-rebind-hot_brace_wrap, rs-vec-no-capacity_cfg_test.
G3 r4 on serde/tokio/clap: 6 clap residuals, 100% TP, FP+DOC+SKIP 0% → corpus-ready: yes.
No G4 escalate for rs-vec-no-capacity/rs-collect-vec-hot.

2026-10-22 (rs-alloc-hot G4 evidence, no severity bump)

rs-alloc-hot G4 microbench on rustc 1.85.1 -C opt-level=3 (evidence/rs-alloc-hot.md + rs-alloc-hot.jsonl; harness examples/bench/evidence_rs001.rs).
- Vec::new each iter vs clear/reuse: ~2.74× (small N ×reps ~9×)
- fixture seed.clone() ± with_capacity: ~1.11×
- growing clone vs clear+push_str: ~2.23×
- to_string vs clear+write!: ~1.75×
Default severity for rs-alloc-hot stayed hint (modest ratios; G3 OK-idiom dominate; no narrow split escalate).
Docs: evidence-policy map, rule-regression, evidence README, kit README.

Split rs-string-plus-in-loop: += / push_str only; default stays hint (amortized ~1×).
Add rs-format-rebind-hot: ident = format!(…) rebind that reuses ident in the format body/args; default error.
- Evidence: evidence/rs-format-rebind-hot.md + rs-format-rebind-hot.jsonl — same measured format! pair as rs-string-plus-in-loop G4 (digits_format_accum ~99× vs write!; aliased_from=rs-string-plus-in-loop; no new timings invented).
Fixtures: bad/rs-format-rebind-hot_format_rebind.rs, good/rs-format-rebind-hot_write.rs, good/rs-format-rebind-hot_let_format.rs.
Docs: evidence-policy, corpus residual note, guide lint map, rule-regression, smell-catalog.

rs-string-plus-in-loop G4 microbench on rustc 1.85.1 -C opt-level=3 (evidence/_removed/rs-string-plus-in-loop.md + evidence/_removed/rs-string-plus-in-loop.jsonl; harness examples/bench/evidence_rs002.rs).
- format! accum vs write!: ~99×
- fixture += / push_str vs with_capacity: ~1.00–1.05×
- small N (×reps for timer): ~2×
Default severity for rs-string-plus-in-loop stayed hint (detector also hit amortized ~1× builders).
Docs: evidence-policy map, rule-regression, guide lint note, evidence README.

Initial rs-alloc-hot–rs-println-hot heuristic checker; G3 reverse-opt (real loop heads, tiny seps, cold CLI paths).