Performance log

Each performance review appends a new section below, using the template at the bottom of this file. Do not edit prior sections — they are a historical record.

2026-04-24 — 4.x bench-suite overhaul

Branch / base: 4.x @ c91cc4b
PHP: 8.4-alpine (linux/amd64), run via docker compose -p mobile-detect up runPerfTests
Host: Apple Silicon, OrbStack
PHPBench config: iterations=10, revs=1000, warmup=2, retry_threshold=1 (percent; from /phpbench.json)
Subjects: 13, Assertions: 13, Failures: 0, Errors: 0

Subject	ops/s	rstdev
`benchMatchOnlyBestRegex`	3,421,012	±0.51%
`benchMatchOnlyWorstRegex`	2,338,811	±0.38%
`benchIsMobileCacheWarm`	1,004,507	±0.42%
`benchIsTabletAgainstBestMatch`	303,764	±0.52%
`benchIsIOS`	99,981	±0.36%
`benchIsIpad`	99,263	±0.35%
`benchIsSamsungTablet`	76,596	±0.58%
`benchIsMobileCacheCold`	69,089	±0.44%
`benchIsMobileAgainstBestMatch`	68,982	±0.49%
`benchIsMobileCacheKeyFnCustomAgainstBestMatch`	66,851	±0.52%
`benchIsSamsung`	56,603	±0.53%
`benchIsTabletAgainstWorstMatch`	19,776	±0.16%
`benchIsMobileAgainstWorstMatch`	8,883	±0.30%

Notes

The warm cache hit is ~15× faster than cold (1.00M vs 69k ops/s) — validates the cache-path split was worth adding, and confirms the PSR-16 cache actually short-circuits the regex loop on hit.
Isolated match() runs at 2.3–3.4M ops/s; isMobile() on the same best-match UA runs 50× slower at 69k. So ~98% of an isMobile() call is not the regex — it's constructor + $_SERVER init + cache init/check + loop overhead. Cheap cache-key wins would be the highest-leverage optimization.
The self-audit in setUpMatchOnlyWorst passed — UA_KT107 still matches the last tablet key (currently GenericTablet). If a future release reorders tablet rules, the audit throws and this fixture needs refreshing.

Template (copy this block, do not fill it in-place)

## YYYY-MM-DD — <short change summary>

- Branch / base: `<branch>` @ `<sha>`
- PHP: `<php version/image>` (<arch>), run via `<command>`
- Host: <cpu, runtime>
- PHPBench config: `iterations=N, revs=N, warmup=N, retry_threshold=N` (percent; from `/phpbench.json`)
- Subjects: N, Assertions: N, Failures: N, Errors: N

| Subject | ops/s | rstdev |
|---|---:|---:|
| `benchFoo` | N | ±N% |
| ... | | |

### Notes

- Interpretation bullet 1
- Interpretation bullet 2
- Any self-audit / fixture / regression notes

---