Assay Artifact Families Inventory¶
Status: orientation inventory. This page classifies current and proposed artifact families as canonical, reference, experiment-scoped, or proposed. It does not promote any artifact and does not define a new schema.
Why This Exists¶
Assay now has several evidence-bearing artifact families: Trust Card, Trust Basis, Runner archives, receipt families, observability join rows, experiment sidecars, and planned fidelity/evidence-pack outputs. This inventory keeps those families legible so new experiments do not accidentally present proposed artifacts as canonical product surfaces.
Status Classes¶
| Status | Meaning |
|---|---|
canonical | Product or release-line artifact family with existing user-facing meaning. |
reference | Stable research/reference vocabulary or schema family used to interpret evidence. |
experiment-scoped | Local measurement or comparison artifact for one experiment line. |
proposed | Planned or working-term artifact family; not yet a stable contract. |
historical | Kept for traceability, not a recommended new surface. |
Current Families¶
| Family | Status | Namespace / docs | Role |
|---|---|---|---|
| Trust Card | canonical | CLI/reference docs | User-facing claim summary surface. |
| Trust Basis | canonical | CLI/reference docs | Lower-level evidence basis for trust claims. |
| Runner archive | canonical | assay.runner.* | Measured-run evidence captured by Runner. |
| Runner projection/report schemas | reference | docs/reference/runner/ | Runner-adjacent reports, diffs, and projections. |
| Receipt families | reference | receipt-families.md | Bounded imported evidence receipts. |
| Observability claim classes | reference | observability/claim-classes-v0.md | Vocabulary for what traces, archives, and joined artifacts can honestly claim. |
| Observability join rows | reference | observability/join-contract-v0.md | Join-grade rows for trace/archive/receipt comparisons. |
| Overhead experiment sidecars | experiment-scoped | assay.experiment.* under runner-vs-otel-overhead-2026-05/ | Samples, summaries, phase timings, paired sequences, and event-rate sweep cells. |
| Cross-runtime drift outputs | experiment-scoped | cross-runtime-drift-2026-05/ | Runtime capability-surface drift comparisons. |
| Fidelity calibration | experiment-scoped | assay.experiment.agent_observability_fidelity.calibration.v0 | Requested-vs-observed fidelity verdicts and per-layer count methods embedded by the overhead harness. |
| Evidence pack | experiment-scoped | assay.experiment.agent_observability_fidelity.evidence_pack.v0 | Prototype portable bundle carrier for one run or scenario, with manifest, summary, health, archive/trace references, and explicit redaction manifest. |
| Binding evidence / join receipts | proposed | undecided | Working term for tool-call input/output/effect binding evidence. Not a product line yet. |
| MCP tool evidence binding | experiment-scoped | assay.experiment.mcp_tool_evidence_binding.binding_cell.v0 under mcp-tool-evidence-binding-harness-2026-05/README.md | Synthetic description/call/effect binding rows for MCP tool context. Includes plural visible-description sets and one tunnel-context fixture, without detecting poisoning, ranking MCP implementations, deploying live MCP servers, or promoting a receipt family. |
| Semantic-gap finding | experiment-scoped | assay.experiment.agent_observability_fidelity.semantic_gap_verdict.v0 under semantic-gap-scenario-plan.md, delegated baseline planning in delegated-baseline-plan.md, the Slice 7 smoke record in runs/slice7-delegated-baseline/summary.md, and the delegated hidden_write sidecar in runs/delegated-hidden-write/summary.md | Synthetic matrix verdicts for reported-intent vs measured-effect divergence, a delegated positive-baseline smoke gate, and one delegated hidden_write sidecar finding. Not a product API or broad delegated gap campaign. |
| Interop mapping | experiment-scoped | assay.experiment.agent_observability_fidelity.interop_coverage_cell.v0 under interop-matrix-plan.md | Synthetic Slice 6 coverage, joinability, and claim-strength rows between OTel GenAI, OpenInference, Runner measured effects, and Assay claim vocabulary. Not a product ranking, translator, delegated finding, or product API. |
Promotion Rule¶
Proposed or experiment-scoped artifacts should not be described as canonical until a promotion PR names:
- the consumer that needs the artifact;
- the namespace and stability promise;
- the validation fixtures or golden examples;
- the migration path from the experiment artifact, if any;
- the non-claims the promoted artifact still carries.
See experiments/namespace-governance.md for naming and promotion details.
Non-Claims¶
- This inventory does not create new artifact families by itself.
- This inventory does not require current experiment artifacts to be renamed.
proposedmeans "useful working term," not "committed product surface."