docs: re-validate perf-review I79/I84 against post-rewrite code

[Xof]

Re-validates two 2026-06-01 perf-review items against the current code (both predate the multi-page freemap rewrite and the transaction.rs extraction, so their line refs were stale).

• I84 (freemap allocate_first scans from byte 0, no cursor) → ✅ ADDRESSED. The COW tree's FreeMapTree::allocate_first (freemap_tree.rs:470) now scans from a hint cursor and advances it per allocation — exactly what the finding asked for. Closed.
• I79 (clone→swap at commit) → re-scoped, not the free win it was billed as. The 8 KB freemap-bitmap clone is gone (freemap is a COW tree now); the surviving current_live_slots clone in SlotPacker::commit is read post-commit by sparse_data_pages / data_page_ids_snapshot, so a mem::swap would expose pre-transaction state and needs a read-timing proof + regression test — and it's a microsecond HashMap clone below the fsync floor. Not a no-bench win.

Outcome of the requested re-validation: no cheap confirmed code wins here — one finding was already fixed by the freemap rewrite, the other isn't cheap in the current code. Tracker corrected so neither gets re-triggered.

Docs-only.

[github-actions]

🚦 Bench results: PR vs main

⚠️ 2 regression(s) detected across 2 scenario/mode pair(s)

Scenario	Mode	Δ throughput	Worst Δ
ycsb-b	chisel-mem	-2.9%	p99 +18.4% ⚠️
ycsb-a	redb-strict	-0.0%	p99 +12.5% ⚠️
document-store	chisel-mem	+1.5%	—
document-store	chisel-strict	-0.4%	—
document-store	redb-strict	+2.2%	—
document-store	sqlite-strict	+0.0%	—
mutation-log	chisel-mem	+1.4%	—
mutation-log	chisel-strict	+0.3%	—
mutation-log	redb-strict	-0.0%	—
mutation-log	sqlite-strict	-1.2%	—
ycsb-a	chisel-mem	+14.3%	—
ycsb-a	chisel-strict	-0.1%	—
ycsb-a	sqlite-strict	+4.3%	—
ycsb-b	chisel-strict	+0.7%	—
ycsb-b	redb-strict	-0.2%	—
ycsb-b	sqlite-strict	+0.5%	—

Per-scenario detail (4 metrics × cells)

document-store

Mode	Throughput	p50	p95	p99
chisel-mem	14739 ops/s → 14953 ops/s (+1.5%)	10.6 µs → 10.8 µs (+1.4%)	137.5 µs → 136.3 µs (-0.8%)	790.3 µs → 787.2 µs (-0.4%)
chisel-strict	2923 ops/s → 2913 ops/s (-0.4%)	22.9 µs → 21.4 µs (-6.8%)	1.04 ms → 1.01 ms (-2.3%)	2.79 ms → 2.70 ms (-3.3%)
redb-strict	3877 ops/s → 3964 ops/s (+2.2%)	23.2 µs → 20.0 µs (-13.7%)	692.3 µs → 612.6 µs (-11.5%)	3.16 ms → 2.98 ms (-5.9%)
sqlite-strict	5129 ops/s → 5131 ops/s (+0.0%)	19.0 µs → 19.7 µs (+3.5%)	378.4 µs → 400.0 µs (+5.7%)	1.79 ms → 1.95 ms (+8.6%)

mutation-log

Mode	Throughput	p50	p95	p99
chisel-mem	52638 ops/s → 53358 ops/s (+1.4%)	22.4 µs → 22.4 µs (+0.0%)	28.5 µs → 27.5 µs (-3.7%)	35.1 µs → 33.7 µs (-3.9%)
chisel-strict	1754 ops/s → 1760 ops/s (+0.3%)	321.2 µs → 317.6 µs (-1.1%)	726.8 µs → 734.1 µs (+1.0%)	15.94 ms → 15.85 ms (-0.6%)
redb-strict	1806 ops/s → 1805 ops/s (-0.0%)	161.1 µs → 159.6 µs (-0.9%)	285.2 µs → 289.2 µs (+1.4%)	32.90 ms → 32.19 ms (-2.2%)
sqlite-strict	5127 ops/s → 5068 ops/s (-1.2%)	90.0 µs → 90.1 µs (+0.1%)	263.5 µs → 262.4 µs (-0.4%)	459.6 µs → 436.1 µs (-5.1%)

ycsb-a

Mode	Throughput	p50	p95	p99
chisel-mem	19874 ops/s → 22707 ops/s (+14.3%)	79.8 µs → 67.1 µs (-15.9%)	107.0 µs → 97.1 µs (-9.2%)	183.9 µs → 160.4 µs (-12.8%)
chisel-strict	2228 ops/s → 2227 ops/s (-0.1%)	286.4 µs → 292.7 µs (+2.2%)	843.5 µs → 847.6 µs (+0.5%)	1.54 ms → 1.66 ms (+7.3%)
redb-strict	2683 ops/s → 2682 ops/s (-0.0%)	125.0 µs → 126.5 µs (+1.2%)	225.6 µs → 244.7 µs (+8.5%)	484.5 µs → 545.1 µs (+12.5%) ⚠️
sqlite-strict	196999 ops/s → 205386 ops/s (+4.3%)	4.8 µs → 4.6 µs (-3.1%)	7.4 µs → 6.9 µs (-6.7%)	9.4 µs → 8.8 µs (-6.9%)

ycsb-b

Mode	Throughput	p50	p95	p99
chisel-mem	143468 ops/s → 139266 ops/s (-2.9%)	2.8 µs → 2.6 µs (-6.4%)	80.0 µs → 78.3 µs (-2.2%)	93.3 µs → 110.5 µs (+18.4%) ⚠️
chisel-strict	21193 ops/s → 21347 ops/s (+0.7%)	4.8 µs → 4.8 µs (-0.1%)	307.4 µs → 302.3 µs (-1.6%)	768.0 µs → 766.9 µs (-0.1%)
redb-strict	27202 ops/s → 27149 ops/s (-0.2%)	3.4 µs → 3.3 µs (-2.7%)	126.3 µs → 123.5 µs (-2.2%)	217.3 µs → 199.6 µs (-8.2%)
sqlite-strict	247026 ops/s → 248259 ops/s (+0.5%)	4.0 µs → 4.0 µs (-1.2%)	5.9 µs → 5.8 µs (-1.4%)	7.0 µs → 6.9 µs (-0.8%)

_{Generated by chisel-bench-diff at 2026-06-23T06:28:04Z. Compares PR HEAD against main. Never blocks merge — signal, not gate. Thresholds: throughput 5%, p50 5%, p95 10%, p99 10%.}