1 oracle.pipelines.renderer_experiment_report

oracle.pipelines.renderer_experiment_report

H16 validation projection; no acquisition, independent metric implementation or corpus admission.

CLI (only after canonical integration review): python scripts/verify_renderer_h16_readback.py –plan PLAN –bindings PREFLIGHT –output NEW.json

PREFLIGHT is independently frozen BEFORE acquisition: plan_hash, revision, units (mapping unit_id -> scene, seed, mode, source_index, shard_id, capture_root, expected_identity). revision is the full GenerationRevision JSON dictionary, not a Git SHA. Current reader source is recorded separately; portable readback does not require checking out historical acquisition code. capture_root and endpoint.geometry_path may be confined plan-root-relative payload locators; original_capture_root and endpoint parent paths remain sealed original identity. Runtime capture namespaces are discovered ONLY under each exact registered capture_root/shard_id.attempt-* path. Multiple completed captures remain ambiguous; no latest-wins selection or silent retry selection.

Endpoint validation uses the shared numerical owner plus promoted store seals and actual selected-label joins. Missing endpoint geometry remains explicitly nonqualifying. Canonical recipe readback owns portable writer/store resolution. This projection preserves sealed original paths; complete cold qualification still needs an actual hydrated relocation verification, rather than these interface tests alone.

1.1 Functions

Name	Description
read_renderer_experiment_report	Verify the frozen twenty-cell experiment and project every intended outcome.

1.1.1 read_renderer_experiment_report

oracle.pipelines.renderer_experiment_report.read_renderer_experiment_report(
    plan_path,
    binding_path,
)

Verify the frozen twenty-cell experiment and project every intended outcome.

The independently frozen binding_path names source identities, exact attempt namespaces and prepared endpoint geometry. Bundle-relative payload locators resolve inside plan_path.parent; original absolute identities remain sealed provenance. Current canonical readers verify acquisition evidence without checking out its historical source revision.

Canonical trace comparison owns candidate/history equivalence, shared metric owners recompute endpoint quantities, and promoted store readers verify content and selected-label joins. Missing or partial attempts stay explicit. Numeric endpoint agreement and exact raw-cloud identity are distinct fields; frozen qualification also requires full H16 and authoritative parent timing. Invalid frozen bindings raise ValueError rather than selecting a retry.

# oracle.pipelines.renderer_experiment_report { #aria_nbv.oracle.pipelines.renderer_experiment_report } `oracle.pipelines.renderer_experiment_report` H16 validation projection; no acquisition, independent metric implementation or corpus admission. CLI (only after canonical integration review): python scripts/verify_renderer_h16_readback.py --plan PLAN --bindings PREFLIGHT --output NEW.json PREFLIGHT is independently frozen BEFORE acquisition: plan_hash, revision, units (mapping unit_id -> scene, seed, mode, source_index, shard_id, capture_root, expected_identity). revision is the full GenerationRevision JSON dictionary, not a Git SHA. Current reader source is recorded separately; portable readback does not require checking out historical acquisition code. capture_root and endpoint.geometry_path may be confined plan-root-relative payload locators; original_capture_root and endpoint parent paths remain sealed original identity. Runtime capture namespaces are discovered ONLY under each exact registered capture_root/shard_id.attempt-* path. Multiple completed captures remain ambiguous; no latest-wins selection or silent retry selection. Endpoint validation uses the shared numerical owner plus promoted store seals and actual selected-label joins. Missing endpoint geometry remains explicitly nonqualifying. Canonical recipe readback owns portable writer/store resolution. This projection preserves sealed original paths; complete cold qualification still needs an actual hydrated relocation verification, rather than these interface tests alone. ## Functions | Name | Description | | --- | --- | | [read_renderer_experiment_report](#aria_nbv.oracle.pipelines.renderer_experiment_report.read_renderer_experiment_report) | Verify the frozen twenty-cell experiment and project every intended outcome. | ### read_renderer_experiment_report { #aria_nbv.oracle.pipelines.renderer_experiment_report.read_renderer_experiment_report } ```python oracle.pipelines.renderer_experiment_report.read_renderer_experiment_report( plan_path, binding_path, ) ``` Verify the frozen twenty-cell experiment and project every intended outcome. The independently frozen ``binding_path`` names source identities, exact attempt namespaces and prepared endpoint geometry. Bundle-relative payload locators resolve inside ``plan_path.parent``; original absolute identities remain sealed provenance. Current canonical readers verify acquisition evidence without checking out its historical source revision. Canonical trace comparison owns candidate/history equivalence, shared metric owners recompute endpoint quantities, and promoted store readers verify content and selected-label joins. Missing or partial attempts stay explicit. Numeric endpoint agreement and exact raw-cloud identity are distinct fields; frozen qualification also requires full H16 and authoritative parent timing. Invalid frozen bindings raise ``ValueError`` rather than selecting a retry.