Runner Schema Overview¶

Status: orientation index. This page does not define a new schema. It lists the current Runner-adjacent contracts, their stability scope, and whether they are archive evidence, projection output, or experiment-only measurement output.

For top-level namespace governance across assay.runner.*, assay.experiment.*, and assay.observability.*, see ../schemas-overview.md.

Namespace Rules¶

assay.runner.* is for Runner archive, evidence, projection, or report contracts that may be consumed by Runner tooling.
assay.experiment.* is for time-limited experiment evidence. These shapes may change between experiment slices and do not become stable Runner contracts unless a later reference page explicitly promotes them.
Content-addressed evidence digests use sha256:<hex> style values. Git commit anchors use bare Git object IDs, full or abbreviated.

Active Contracts¶

Schema	Scope	Status	Sidecar
`assay.runner.archive_manifest.v0`	Runner archive manifest metadata	archive contract	described in `artifacts-v0.md`
`assay.runner.capability_surface.v0`	Capability surface snapshot inside archives	archive contract	described in `artifacts-v0.md`
`assay.runner.sdk_event.v0`	SDK-layer NDJSON events	archive contract	described in `artifacts-v0.md`
`assay.runner.kernel_event.v0`	Kernel-layer NDJSON events	archive contract	`kernel-event-v0.schema.json`
`assay.runner.runtime_drift.v0.2`	Cross-runtime drift report emitted by the current comparator	projection/report contract	`runtime-drift-v0.2.schema.json`
`assay.runner.runtime_drift.v0`	Historical cross-runtime drift report shape	retained for older reports	`runtime-drift-v0.schema.json`
`assay.runner.path_projection.v0`	Additive path projection block embedded in runtime-drift reports	embedded projection vocabulary	described in `runtime-drift-v0.md`
`assay.runner.network_projection.v0`	Additive network projection block embedded in runtime-drift reports	embedded projection vocabulary	described in `runtime-drift-v0.md`
`assay.runner.runtime_noise_taxonomy.v0`	Shared vocabulary for runtime/provider/task/noise classes	vocabulary-only	described in `runtime-drift-v0.md`
`assay.runner.drift_report_provenance.v0`	Render and capture provenance embedded in runtime-drift reports	embedded report metadata	described in `runtime-drift-v0.md`
`assay.runner.fidelity_verdict.v0`	Derived claim gate from one `observation_health.v0` record	internal helper contract; no archive member or sidecar yet	described in `fidelity-verdict-v0.md`
`assay.runner.coverage_descriptor.v0`	Per-dimension capture method, blind spot, and claim-kind gate descriptor	internal helper contract; no archive member or sidecar yet	described in `coverage-descriptor-v0.md`
`assay.runner.cross_runtime_diff.v0.clean`	Earlier clean-output cross-runtime diff shape	reference schema	`cross-runtime-diff-v0-clean.schema.json`

Experiment Schemas¶

Schema	Scope	Status	Sidecar
`assay.experiment.overhead_sample.v0`	One overhead measurement sample for runner-vs-OTel	experiment-scoped; sidecar active; phase timing diagnostics included	`overhead-sample-v0.schema.json`
`assay.experiment.overhead_summary.v0`	Aggregated overhead summary for runner-vs-OTel	experiment-scoped; sidecar active; phase timing diagnostics included	`overhead-summary-v0.schema.json`
`assay.experiment.runner_phase_timing.v0`	Phase-timing side-log emitted by `assay runner-spike run --phase-timing-log`	experiment-scoped; sidecar active	`runner-phase-timing-v0.schema.json`
`assay.experiment.event_rate_sweep.v0`	Event-rate/workload-intensity cell metadata embedded in overhead samples and summaries	experiment-scoped; sidecar active; Slice 10 smoke-verified	`event-rate-sweep-v0.schema.json`
`assay.experiment.event_rate_sweep.v0.1`	Event-rate/workload-intensity cell metadata with Slice 12 extended targets	experiment-scoped; sidecar active; Slice 12 measured	`event-rate-sweep-v0.1.schema.json`
`assay.experiment.agent_observability_fidelity.calibration.v0`	Requested-vs-observed fidelity calibration embedded in overhead samples and summaries	experiment-scoped; sidecar active; fidelity guardrail harness-ready	`fidelity-calibration-v0.schema.json`
`assay.experiment.agent_observability_fidelity.evidence_pack.v0`	Portable synthetic evidence-pack manifest for one agent-observability scenario	experiment-scoped; prototype-ready; not a product API	`evidence-pack-v0.schema.json`
`assay.experiment.agent_observability_fidelity.redaction_manifest.v0`	Explicit redaction record carried by every agent-observability evidence pack	experiment-scoped; prototype-ready	`redaction-manifest-v0.schema.json`
`assay.experiment.agent_observability_fidelity.semantic_gap_verdict.v0`	Bounded verdict for all six Slice 4 semantic-gap synthetic scenarios	experiment-scoped; synthetic matrix-ready; not a delegated finding	`semantic-gap-verdict-v0.schema.json`
`assay.experiment.agent_observability_fidelity.synthetic_trace.v0`	Synthetic trace fixture used by the local semantic-gap harness	experiment-scoped; schema-string only; not delegated evidence	none
`assay.experiment.agent_observability_fidelity.synthetic_runner_archive.v0`	Synthetic Runner-archive fixture used by the local semantic-gap harness	experiment-scoped; schema-string only; not a Runner archive contract	none
`assay.experiment.agent_observability_fidelity.interop_coverage_cell.v0`	OTel GenAI / OpenInference / Runner coverage, joinability, and claim-strength matrix row	experiment-scoped; harness-ready; synthetic only; not a product API	`interop-coverage-cell-v0.schema.json`

These experiment schemas remain experiment-scoped. They are validated by local harness tests against synthetic samples, summaries, phase side-logs, event-rate sweep cells, fidelity calibration sidecars, and agent-observability evidence packs/verdicts, but they are not Runner archive contracts and are not promoted to stable product surface.

Version Notes¶

assay.runner.runtime_drift.v0.2 is a compatible tightening of the runtime drift report shape, not a replacement for the raw archive contracts. Consumers should parse dotted minor suffixes semver-like: v0.2 means major 0, minor 2. The current comparator emits v0.2 only; v0 is retained for historical reports already committed under experiment runs.