Validation · Full 30-source comparison · Original isolated experiment
The dependency-cost bound is now present in live column.rs. The captured executable reports4.3.0. All180 real CLI Native/Hybrid default/fast/thin archives match the freshly rebuilt benchmark worker byte-for-byte. Every archive restores exactly with the captured reader and the4.2.1 reader; the current reader also restores all180 frozen4.2.2 archives (540 restores total). All56 current library tests pass, including the budget boundary test.
The older public_logs_30_4.3.0 source snapshot does not contain the dependency bound. It is preserved; the new measurement has a separate public_logs_30_4.3.0_post_budget identity and complete source/binary hashes.
An initial capture of the CLI preceded concurrent Hybrid arbitration changes. Forty of its180 outputs differed from the newer worker. That capture is retained as cli_validation_superseded.json and is NOT used for final CLI/worker identity claims. The recaptured CLI matches all180 measured PIZD cases. The peer subsequently advanced Cargo metadata to4.3.1; do not relabel this measured4.3.0 snapshot as a4.3.1 result.
30 existing public excerpts; three warmups and five measured rounds;540 configuration rows /8,640 exact restorations /4,320 independent SHA-256 checks. Encoding excludes file I/O; verified decoding includes source integrity verification. These are development excerpts, not new held-out systems.
| Codec | Aggregate archive bytes | Encode sum ms | Verified decode sum ms |
|---|---|---|---|
| PIZD_native_thin_L0 | 820,774 | 366.847 | 34.931 |
| PIZD_hybrid_thin_L9 | 750,015 | 536.604 | 28.244 |
| Zstd_L9 | 835,071 | 294.651 | 13.392 |
| Brotli_Q9 | 815,753 | 990.983 | 25.847 |
These combined-release times are not the isolated patch comparison. The earlier -10.09% Native /-7.83% Hybrid results describe the frozen dependency-budget experiment only. NDJSON, Hybrid routing and other source changes alter the combined release; no new causal version-to-version speed percentage follows from comparing separate runs. All losses versus competing codecs remain in the complete public tables.
| Source | Engine thin | Encoding ms | Archive bytes |
|---|---|---|---|
| 25_nasa_http_jul95.log | PIZD_native_thin_L0 | 12.445 | 36151 |
| 25_nasa_http_jul95.log | PIZD_hybrid_thin_L9 | 15.302 | 35057 |
| 26_nasa_http_aug95.log | PIZD_native_thin_L0 | 11.149 | 37283 |
| 26_nasa_http_aug95.log | PIZD_hybrid_thin_L9 | 14.267 | 35956 |
The original isolated patch preserved all chosen archive bytes. In the combined4.3.0 build,123 of180 actual CLI archives differ from4.2.2 because other modeling/routing changes are present. All of them restore exactly with both readers. Do not attribute every archive-size change to the byte-preserving budget bound.
validation.json links the source/worker/CLI checks; cli_validation.json records every archive hash. The measured source snapshot and all timed trials live in ../public_logs_30_4.3.0_post_budget/. The older4.3.0 report and superseded CLI capture remain intact. Current pending source metadata does not replace the identity of measured binaries.