Use-case framework
Good enough for what?
Trace quality is not universal. A debugging trace may be useful without preserving the principal or externally verified state, while an incident or delegated-authority review may depend on both.
Direct answer
Start with the decision the evidence must support.
Name the reconstruction use case, list the evidence it requires, and then assess a trace against that target. Trace Map documentation identifies represented evidence; compatibility exercises show what survives selected mappings; Inspect evaluates one supported local execution; and Gate compares baseline and candidate outcomes. None of those layers claims that an arbitrary whole system satisfies the use case.
Use case 01
Basic debugging
Reconstruct the recorded model/tool sequence, arguments, responses, timing, and errors around a failed execution.
Required evidence
- Session ID — Identity: Session, conversation, thread, run, or trace grouping.
- Model identifier — Models: Named model.
- Tool call ID — Tools: Request/result correlation identifier.
- Tool arguments — Tools: Arguments supplied to the tool.
- Tool result — Tools: Tool-reported response.
- Tool error — Tools: Tool failure or exception.
- Timestamps — Integrity: Event or span timing.
Intended conclusion: Enough correlated runtime evidence to debug execution without claiming authority or external settlement.
Related directional mappings
- OpenTelemetry GenAI development snapshot → OTLP 1.11: established
- MLflow native 3.15.2 → OTLP 1.11: established
- MLflow 3.15.2 native → GenAI-transformed OTLP: established
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 02
Multi-agent reconstruction
Follow work across parent agents, subagents, sessions, handoffs, and tool calls.
Required evidence
- Agent instance ID — Identity: Stable executing agent instance identifier.
- Parent/subagent relationship — Identity: Correlation between agent parents and children.
- Session ID — Identity: Session, conversation, thread, run, or trace grouping.
- Delegation chain — Principal: Ordered authority delegation evidence.
- Tool call ID — Tools: Request/result correlation identifier.
- Timestamps — Integrity: Event or span timing.
Intended conclusion: A stable hierarchy and correlated event sequence across agents.
Related directional mappings
- OpenTelemetry GenAI development snapshot → OTLP 1.11: established
- OpenInference 0.1.33 → OTLP 1.11: partial
- MLflow native 3.15.2 → OTLP 1.11: partial
- MLflow 3.15.2 native → GenAI-transformed OTLP: partial
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 03
Tool-action attribution
Attribute a specific tool request and result to an agent execution and initiating principal.
Required evidence
- Initiating principal — Principal: Human or service that initiated activity.
- Agent instance ID — Identity: Stable executing agent instance identifier.
- Tool identity — Tools: Invoked tool name or definition.
- Tool call ID — Tools: Request/result correlation identifier.
- Tool arguments — Tools: Arguments supplied to the tool.
- Tool result — Tools: Tool-reported response.
- Downstream correlation — Tools: Link to provider or destination evidence.
Intended conclusion: Join principal, agent, and tool evidence without treating the result as destination proof.
Related directional mappings
- OpenTelemetry GenAI development snapshot → OTLP 1.11: established
- OpenInference 0.1.33 → OTLP 1.11: not established
- MLflow native 3.15.2 → OTLP 1.11: established
- MLflow 3.15.2 native → GenAI-transformed OTLP: established
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 04
Human-delegation reconstruction
Establish which human or service delegated what authority, including approvals and elevation.
Required evidence
- Initiating principal — Principal: Human or service that initiated activity.
- Authenticated principal — Principal: Identity established by authentication.
- Delegated identity — Principal: Identity acting through delegation.
- Delegation chain — Principal: Ordered authority delegation evidence.
- Credential scope — Principal: Authority scope of a credential.
- Credential expiry — Principal: Credential validity boundary.
- Authority elevation — Principal: Elevation or privilege-change evidence.
- Human approval — Human control: Recorded approval.
Intended conclusion: Evidence the human/service authority chain separately from the agent hierarchy.
Related directional mappings
- OpenTelemetry GenAI development snapshot → OTLP 1.11: not established
- OpenInference 0.1.33 → OTLP 1.11: not established
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 05
External-effect verification
Distinguish requested operations and tool responses from provider acceptance and resulting destination state.
Required evidence
- Requested effect — Effects: Operation requested at a tool/provider boundary.
- Provider acceptance — Effects: Evidence that a provider accepted the operation.
- External effect state — Effects: Observed destination state.
- External state delta — Effects: Before/after change in external state.
- Downstream correlation — Tools: Link to provider or destination evidence.
- Externally verified success — Outcome: Success established from destination evidence.
Intended conclusion: Correlate the requested effect to independently observed external state.
Related directional mappings
- OpenTelemetry GenAI development snapshot → OTLP 1.11: external evidence required
- MLflow native 3.15.2 → OTLP 1.11: external evidence required
- MLflow 3.15.2 native → GenAI-transformed OTLP: external evidence required
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 06
Incident reconstruction
Build an evidence timeline across agent, identity, authorization, tool, and destination systems.
Required evidence
- Initiating principal — Principal: Human or service that initiated activity.
- Agent instance ID — Identity: Stable executing agent instance identifier.
- Model version — Models: Version or build identity.
- Tool arguments — Tools: Arguments supplied to the tool.
- Tool result — Tools: Tool-reported response.
- Credential scope — Principal: Authority scope of a credential.
- Human approval — Human control: Recorded approval.
- External effect state — Effects: Observed destination state.
- Timestamps — Integrity: Event or span timing.
- Tamper evidence — Integrity: Evidence that modification can be detected.
Intended conclusion: A time-correlated account of observable actions, authority, effects, and uncertainty.
Related directional mappings
- OpenInference 0.1.33 → OTLP 1.11: partial
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 07
Audit support
Preserve attributable, reviewable evidence with approvals, lifecycle controls, and integrity context.
Required evidence
- Authenticated principal — Principal: Identity established by authentication.
- Delegation chain — Principal: Ordered authority delegation evidence.
- Human approval — Human control: Recorded approval.
- Timestamps — Integrity: Event or span timing.
- Clock provenance — Integrity: Source and reliability of timestamps.
- Tamper evidence — Integrity: Evidence that modification can be detected.
- Retention duration — Privacy: Data lifecycle duration.
- Deletion control — Privacy: Removal mechanism.
Intended conclusion: Support an audit inquiry without overstating record authenticity or completeness.
Related directional mappings
- OpenTelemetry GenAI development snapshot → OTLP 1.11: not established
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 08
Cross-tool portability
Identify which trace semantics can move to another system and where meaning may be lost.
Required evidence
- Native export — Portability: Native JSON, JSONL, or vendor export.
- Schema version — Portability: Version marker for the trace schema.
- OpenTelemetry / OTLP — Portability: OpenTelemetry representation or transport.
- W3C Trace Context — Portability: W3C trace propagation.
- Import destination — Portability: Documented consumer or backend.
- Documented conversion — Portability: Documented mapping between formats.
- Semantic loss — Portability: Explicitly described mapping loss.
Intended conclusion: Describe export/import semantics and explicit loss; v0.1 does not test conversions.
Related directional mappings
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Use case 09
Privacy-controlled archival
Retain useful evidence while controlling sensitive capture, redaction, retention, and deletion.
Required evidence
- Customer-controlled capture — Privacy: Customer selection of captured material.
- Redaction — Privacy: Removal or replacement of sensitive values.
- Filtering — Privacy: Selective capture control.
- Secrets exposure — Privacy: Risk or handling of credentials/secrets.
- PII exposure — Privacy: Personal data capture.
- Retention duration — Privacy: Data lifecycle duration.
- Deletion control — Privacy: Removal mechanism.
Intended conclusion: Make sensitive-data presence and lifecycle explicit without requiring remote upload.
Related directional mappings
Results are scoped to individual evidence assertions in the named mapping, not the use case as a whole.
Machine-readable target
Carry the requirement into local reconstruction work.
Frozen Trace Reconstruction Requirements v0.1 and additive v0.2 provide JSON Schemas and worked examples for expressing the evidence a team needs. The local Inspector evaluates supported trace inputs without uploading them; Gate compares baseline and candidate reconstruction outcomes under the same current evaluator.