Trajectory labels

Trajectory labels and step names¶

Each projected trajectory carries a sorted labels list in memory_trajectories.labels_json. Labels are deterministic tags derived from audit event cores — not free-form agent text.

Label	When set
`change_control_workflow`	Any change-controller event (`intent.`, `patch_contract.`, …)
`verified_finish`	`patch_contract.verified` with accepted outcome
`scope_clean`	`intent.checked` with status `clean` or `expanded`
`scope_expanded`	`intent.expanded` present
`queue_used`	`intent.queued` present
`patch_trail_recorded`	`patch_trail.computed` present
`receipt_issued`	`review_receipt.created` present
`claim_validated`	`claim_validation.completed` present
`analysis_observed`	Standalone `analysis.completed` workflow (no change-control events)
`memory_used`	`manage_engineering_memory` tool use in the stream
`recovered`	`intent.promoted` (queue recovery)
`foreign_conflict_seen`	Workspace conflict
`hook_blocked`	Hook surface warn/error
`claim_guard_failed`	Claim validation violated
`baseline_abuse_detected`	Baseline abuse
`external_changes_accepted`	Finish accepted with external changes

Routine successful edit cycles should carry change_control_workflow and verified_finish at minimum. Empty labels indicates a projection bug or a legacy row that needs memory trajectory rebuild.

Each step in MCP trajectory_get includes step_label — a human-readable name from codeclone/memory/trajectory/step_labels.py (event catalog + status). CLI memory trajectory show prints labels and step labels.