Files
jobtrackingapp/docs/performance/static-encoding/QUALIFICATION.md
T

3.1 KiB

PERF-BASE / PERF-12 static encoding measurement

Recorded 17 November, preserved and qualified 18 November 2026. Frozen source eb2538772e1bfabbeb291e4a5efa177c79074de1, Node24.19.0 Windows, Next16.4.6 static export, explicit API5302 build. This is an offline content-size/encoding baseline, not an application optimisation, current later-source export, deployed compression test or production readiness. The previous performance baseline used a different application source and is not a controlled before/after comparison.

Frozen export group Files Raw bytes Gzip6 bytes Brotli5 bytes
JavaScript 58 3,674,509 1,089,362 1,015,470
CSS 2 3,958 1,189 1,073
SVG 4 28,397 6,971 6,404
HTML 3 32,504 8,339 7,486
Root-declared scripts, deduplicated, including legacy 7 582,094 178,820 169,211

All67 text assets were compressed three times in memory and decompressed with exact byte equality:402 roundtrip assertions passed. Native JSON retains each raw SHA256, encoded sizes and all three elapsed-time samples. Root-declared scripts exclude later dynamic imports; complete export inventory is not a route/network request budget. No source maps were emitted by this frozen build. Headers, TLS, cache state, transfer and on-device decompression are not measured.

The summed per-file median encoding costs were approximately84.9ms gzip6 and180.6ms Brotli5 on this warm-file Windows host. They are exploratory local CPU observations, not a batch-time median or server capacity result. Other development/browser work was active; there was no controlled idle-host/load experiment. These timings justify no production throughput/latency or cost claim. Encoded content sizes are deterministic for the recorded inputs/runtime; no live delivery savings are claimed.

The local desktop-linux Docker endpoint was confirmed to be the local Windows named pipe; a read-only version probe failed because that pipe was absent. No daemon/infrastructure repair, container/image operation or repeated unchanged failure was attempted. Runtime Nginx/static encoding, Accept-Encoding/Vary/cache and private-response boundaries remain unqualified. No gzip/Brotli configuration, CDN, topology, image, API middleware, dependency, file deletion or production change was made.

Next eligible runtime experiment requires an available authorised disposable Nginx/container lane. Compare public immutable assets first; retain HTML revalidation and private API/document isolation, one encoding layer, correct MIME/cache/Vary behavior and exact decode integrity. Brotli module presence and target CPU/storage cost must be proven rather than assumed. Keep PERF-12 PARTIAL and deployed configuration unchanged until that evidence exists. PERF-04 requested-icon and PERF-09 route-chunk concerns retain their own measurements and gates; this census does not qualify either.

Source, producer and native receipts preserve the exact frozen manifest and assert-bearing Node script without adding a runtime/CI runner. No new backend/MariaDB/browser/provider/model/hosted/native-device or production qualification.