Adaptive execution 0.53 qualification

scripts/check_adaptive_0_53.py executes the adaptive planner, both protected Sol review suites and the real-backend physical qualification campaign. Required tests must execute and pass: skipped tests cannot qualify support. The campaign covers five-family empty/nullable differential execution, exact branch and directional dual-port patterns, partial scopes, finite collection, local checkpoints/reuse, definite publication failure, atomic file overwrite and qualified executor lifecycle/port ownership.

Run uv sync --locked --group dataframes, then install the exact qualified backends with uv pip install "polars==1.42.1" "pandas==2.3.3" "pyarrow==25.0.0". Core and first-party plugin distributions are 0.53.0 for this release candidate. Installation alone grants no qualification.

Use --write to record observed results (including failures); it does not grant release approval. The default verifier runs the campaign afresh without writing and requires matching source fingerprints, executed test identities/statuses, and digests of the committed sanitized JUnit and aggregate process observations. Evidence excludes captured logs, exception payloads, source rows and credentials. A failed/conflicted record deliberately fails the verifier until Sol adjudicates the protected verification contract and the complete campaign passes.

The packaged etlantic.runtime/adaptive_support.json is independent of submitted plan topology and generated run fingerprints. It restricts matching to exact logical/port shapes, contiguous target assignments and qualified dependency versions. Compute units are currently unfused; a fused realization requires its own passing signature and is rejected by these rows.

CI executes qualification on Linux, macOS and Windows with Python 3.11, 3.12 and 3.13, retaining separate environment proof artifacts. Those jobs must actually pass before any environment qualification claim; a single local environment proof does not stand in for the remaining CI environments. All functionality remains Experimental pending the independent 0.54 qualification decision.