# PIZD 6.0.0-rc5 acceptance

The release declines a repeat-distance plan when its shifted identifier exceeds the existing24-bit representation. The ordinary distance plan remains available. No wire grammar changes.

- All295 project tests pass, including exact range-boundary regressions under Huffman and opt-in rANS.
- Actual wide matcher failure repaired:16,777,230 source bytes,1,306 token bytes. Old repeat payload634 B was corrupt; corrected954 B payload restores exactly. The smaller corrupt representation is never counted as a size gain. See `generated-after.log` and the preceding rc4 failure evidence.
-162 matched profiles,972 complete restores across30 public excerpts and six4MiB structured controls. All ordinary archive bytes remain identical to frozen rc4. Three interleaved trials are exploratory timings, not a machine-exclusive speed guarantee.
- `./tools/panel_bench.sh` completed for18 enterprise inputs, all four Native/Hybrid fast/density configurations and all comparator restores verified. This is the existing enterprise panel, not a new public30 scoreboard. Full run identities and per-file outcomes are retained in [panel history](../panel_history.json); captured stdout is [panel-bench.log](panel-bench.log).
- `pizd bench` completed three iterations on public HDFS/GitHub with Zstd raw/trained dictionary, gzip and Brotli: [HDFS](comparative-hdfs.md), [GitHub](comparative-bench.md). These are CLI diagnostic configurations, not the thin/template residual research baseline. The built-in dictionary trains on the evaluated input and excludes dictionary bytes/training cost from its reported payload/time, so do not use that row as a standalone or held-out comparison. Verification uses the CLI's source hash; the matched archive gate also checks external source identity.

[Acceptance](acceptance.json), [matched results](results.json), [source identities](fingerprints.json), [source snapshot](runtime_sources.tar.gz), test/build logs alongside. The source snapshot was captured after build and matched verification; there is no rc5 prebuild source manifest. Source fingerprints were held fixed through the matched run.

Research context/MTF frames remain separate and are not integrated into rc5. Historical public benchmark values retain their actual measured version.