Skip to content

Milestone 0.11 parity report

Status: complete (honesty repair on tip)

Milestone: 0.11

Verdict

scripts/check-0.11-release.py qualifies 0.11.0 from C-oracle two-sided differentials (system librdf ↔ Oxiland librdf-compat), obligation coverage tied to positive/boundary/failure/lifecycle fixtures, non-synthetic native performance benches, and ABI-swap evidence. Python-only or fabricated Redland observations are rejected.

Delivered packages

Package State Evidence
WP-11-01 baseline + obligations done compatibility/baseline/0.11-baseline-manifest.json, 0.11-obligations.json, redland-1.0.17-oxiland-0.11.json
WP-11-02 two-sided harness done compatibility/harness/two_sided.py + c_oracle/ + fixtures/0.11/
WP-11-03 fail-closed checker done scripts/check-0.11-release.py, scripts/derive-0.11-parity.py
WP-11-04 safe workflows done tests/obligations_0_11.rs + harness differentials
WP-11-05 C source corpus done compatibility/downstream/corpus/ (-Werror, Redland + Oxiland)
WP-11-06 binary ABI packaging done scripts/package-librdf-compat.sh, compatibility/downstream/abi/
WP-11-07/08 soft stubs done hard-fail or implemented; see docs/users/c-abi-limitations.md
WP-11-09 safety / fuzz done compatibility/qualification/0.11-fuzz.json (≥3600s executed per target)
WP-11-10 cross-platform raw done six-cell C-oracle raw under compatibility/qualification/raw/
WP-11-11 perf + packaging done independent C perf_bench evidence; package-gate corpus/ABI in CI
WP-11-12 soak + release done 0.11-soak.json; checker green; docs at 0.11.0

Hard gates

Gate State Notes
Complete baseline denominator closed Checksummed headers + export dump + obligations
Native two-sided differential harness closed C oracles for both engines; rejects missing Redland/Oxiland libs
Safe workflow parity closed Obligation-linked Rust tests + fixture coverage
C source compatibility closed Frozen corpus passes Redland and Oxiland -Werror builds
Binary ABI interchange closed librdf-compat packaging + ABI-swap smoke stamp
Full observable behavior closed All six profiles differential_passed with C engines
Performance honesty closed Independent measurements; synthetic: false; no paired-ratio fabrication
Soft-stub silence closed Exported stubs either implemented or hard-fail (ADR-023)
Cross-platform provenance closed .github/workflows/qualify-0.11.yml with derive + checker job
Candidate safety and conformance closed Fuzz ≥3600s executed; soak revision-bound
Performance and packaging retention closed Native host C benches; CI package gate
Independent reproduction and soak closed Checker passes; soak abi_resets=0

Completion rule

This report is complete because scripts/check-0.11-release.py derives a passing result from linked raw C-oracle evidence for every frozen target and profile. Counts from an inventory, an export allowlist, Python-only runs, or fabricated Redland observations cannot close a gate.