Traces And Inspectability

Boundline traces make delivery explainable. They preserve what the runtime decided, what it used as evidence, what it ran, what it skipped, what failed, and what should happen next.

Where Traces Live

Workspace traces live under:

.boundline/traces/

The active session state lives at:

.boundline/session.json

When CLI output reports trace_location, treat it as the authoritative trace reference for that run or inspection.

What Traces Contain

Depending on the command and lifecycle phase, traces can include:

• command name and exit status
• rendered operator output
• session status
• context summary and credibility
• context primary inputs and provenance
• plan state and planning rationale
• verification strategy
• route owner and route config projection
• selected guidance and guardian sources
• loaded and skipped packs
• catalog validation findings
• guardian timeline and findings
• changed files and validation status
• checkpoint refs
• Canon packet or project-memory refs when governed delivery is active
• next command, corrected command, or stop reason

Inspect Decisions

Use:

boundline inspect --workspace .

Use JSON when an assistant or script needs structured fields:

boundline inspect --workspace . --json

Ask these questions while reading inspect output:

• Why did Boundline select this context?
• Was context credible enough to continue?
• Which guidance shaped the plan?
• Which guardians ran, skipped, degraded, or blocked?
• Which route owned the current step?
• What evidence supports the next action?

Inspect Guidance And Guardians

Look for:

• guidance_resolution_summary
• loaded_guidance_sources
• loaded_guardian_sources
• loaded_packs
• skipped_packs
• catalog_validation_findings
• guardian_timeline
• guardian_findings_summary
• guardian_blocking_outcome

These fields explain capability selection and validation behavior.

Inspect Context

Look for:

• context_summary
• context_credibility
• context_primary_inputs
• context_provenance
• context_staleness_reason
• retrieval_index_state
• semantic_capability_state
• semantic_fallback_reason
• retrieval_recovery_guidance

If context is weak, repair it by narrowing the goal, adding a brief, producing validation evidence, or pointing at relevant files.

Inspect Recovery

After a failed or blocked run:

boundline status --workspace .
boundline next --workspace .
boundline inspect --workspace .

Preserve reported checkpoint fields:

• latest_checkpoint_id
• latest_checkpoint_scope
• latest_checkpoint_restore_command

Use the restore command only when intentionally rewinding the bounded workspace slice.

Operator Rule

Do not infer continuation from chat history. Continue from status, next, inspect, and trace-backed evidence.