PostProject 0.9 / ABI 57 qualification

Prepared 0.9.0-alpha.1, Python 0.9.0a1, C ABI 57 / 344 exports / 45 layouts, schema 27, exchange framing 1 and CLI JSON 1. This report records the candidate’s pre-publication qualification. The original ABI-53 report and nineteen-artifact set retain their historical scope. All required integrated checks pass for the source and scope below.

Delivered workflows

Import returns both asset and original-representation identities. Path-based lookup removes conversion wrappers without merging assets. Verified relink selection retains copied evidence and its original coherent base; confirmation stages all locations together, with guarded no-op receipts for known locations. Output registration stages media, actual activity, input roles, parameters and evidence atomically. Real jobs compose guarded completion in that same edit.

Native edits accept attribution at creation. C++ has copied target-wide metadata reads and both Result/exception styles. Python reads and writes unambiguous metadata as scalars and immutable tuples; explicit tags preserve ambiguous meanings and ordered repeated fields. Worker and exchange values have explicit modules. The migration map removes superseded forms across all five surfaces. ADRs 0076–0084 record these decisions; no new standards mapping is introduced.

Provenance inputs do not invent authored dependencies. A separate scene example records actual references. Knowledge state, availability and work state retain their separate meanings; recorded recipe completeness does not establish executable or deterministic reproduction.

Local evidence

Boundary	Completed check
Rust	Six required gates; exact Rust 1.85 tests and doctests
Native	Nine installed contracts, 344 exports and 45 Rust/C/ctypes layouts; both C++ styles
Memory/ownership	Nine ASan/UBSan contracts; relocated installed consumers with Cargo prohibited
Python	Final licensed wheel: 157 tests on both 3.11 and 3.14, installed CLI enabled; fifteen wrong-ID negatives and positive typing
Examples	96 extracted installed checks; strict generated documentation and coverage
Packages	Installed Manager 0.6.0 and demo 0.4.0; Manager nine tests on both supported OpenAssetIO dependency sets
Real hosts	Blender 5.2.2: 24 passed/one future-host skip; all eight OBS 32.2.2 scenarios; Natron 2.5.0 normal/negative/plugin-free renderer
Native pilots	Ardour 9.8 installed resolver; complete Kdenlive 26.08.2 build and two test suites
Shared production	Both OBS → Blender and Kdenlive → Blender → Manager pass identity, relocation, output, conflict and staleness checks

The five-task comparison preserves runnable qualified-0.9 before/current after programs. The measurements retain schema-27 baselines, new samples, hashes and explicit allocation/timing limits.

Strict selection rejects sampled-only evidence. The hosts retain explicit policies: Kdenlive also checks its clip hash, Blender and Ardour can reuse an already recorded sampled file address, and sequence pilots require complete frames before accepting sampled collection evidence. These policies do not establish full-content verification. The eight-second audio regression also rejects new sampled discovery and mismatched recorded content. No helper guarantees filesystem atomicity.

Source and CI scope

The consumer source record names all thirteen maintained repositories and scoped checks. Runtime payloads come from SDK 9bb89993; subsequent SDK changes record evidence, format tools, initialize a C example receipt, strengthen version checking and improve test assertion diagnostics. They do not change native or Python runtime bytes.

Pipeline and source	Individual conclusions
SDK 38628074732, 4c6c0563	All 16 jobs pass, including Linux/macOS/Windows, installed packages, glibc-2.28 wheel, MSRV, sanitizers and strict docs/links
Downstream 38628076081, 4c6c0563	All 11 jobs pass, including full Ardour with/without SDK and both Manager dependency sets
Kdenlive 38628074380, d5f6a81	Full with/without-SDK builds, tests and handoff pass
Blender 38628064357, 35f77c0	Stable 5.2.2 and lint pass; optional 5.3 beta fails
Ardour 38637387778, c80b549	Current full with/without-SDK builds and macOS sampled-audio regression pass

The individual job record retains exact inputs and conclusions. Earlier formatting, current-Clippy and MSVC example failures were corrected; failed runs remain in the record. Final report/checksum edits reuse these checks with an explicit source comparison; strict documentation and deployment are checked for the final records.

Distribution and limits

The prepared set contains native shared/static libraries, CLI, headers, relocatable CMake/pkg-config metadata, typed platform wheel, Blender extension, Manager/demo wheels and thirteen source archives. Native packages include dependency notices. These local Linux artifacts are workstation builds; the separate glibc-2.28 CI wheel has its own platform qualification. The artifact record identifies source snapshots, payload hashes and all nineteen prepared files. Its manifest and checksums remain outside the archives to avoid self-reference.

Deferred location confirmation uses locator.confirm.v1; older experimental decoders reject that command explicitly. Schema, existing record encodings and protocol framing are unchanged. Bindings and native library must match ABI 57.

The pinned OTIO linker retains its existing deprecation warning. Optional Blender 5.3 beta 346d005c9994 fails project-variable relocation and completed render fixtures. Licensed Nuke execution and a second independent real worker remain unexecuted. Stable host support does not imply other-host compatibility, timeline conversion, media transfer, editable replicas or upstream endorsement.

At this qualification checkpoint, the deployed landing source was f823e7f; preview and deployment evidence retain their recorded revisions. Prepared review archives are separate from tag-built release assets. The release checklist describes tagging, GitHub publication and versioned documentation; registry publication is not part of that workflow.