Trust survival
Dry run (simulation): Inbox Reflex Triage
simulation_onlyobserve_onlyinbox_reflex_v1
ReflexFlow may simulate execution before it is permitted to execute. This run carries no execution authority and invokes no tools.
Turn budget
simulation_only- max
- 5
- consumed
- 5
- remaining
- 0
- exceeded
- no
HITL checkpoints
simulation_only- turn 4trust_drop(require_on_trust_drop)
- turn 5trust_drop(require_on_trust_drop)
Dry-run GhostLog trace
simulation_onlyobserve_onlySimulated trace events only — not persisted, not enforcement.
- workflow_dry_run.started2026-05-28T01:48:08.923Z
{ "riskLevel": "B", "maxTurns": 5, "tierDefaultMaxTurns": 5, "effectiveMaxTurns": 5 } - turn.budget_checked2026-05-28T01:48:08.923Z
{ "max": 5, "tierDefault": 5, "effectiveMax": 5, "consumed": 0, "remaining": 5 } - hitl.checkpoint_reached2026-05-28T01:48:08.923Z
{ "turn": 4, "label": "trust_drop", "reason": "require_on_trust_drop" } - hitl.checkpoint_reached2026-05-28T01:48:08.923Z
{ "turn": 5, "label": "trust_drop", "reason": "require_on_trust_drop" } - workflow_dry_run.completed2026-05-28T01:48:08.923Z
{ "turnsSimulated": 5, "hitlCheckpointCount": 2, "exceeded": false }
Dry-run timeline
simulation_onlyobserve_onlySimulated execution steps in turn order. Tools are named only — never invoked.
- turn 0intenttrust 0.72
Simulate workflow intent: Inbox Reflex Triage
- turn 1authority_checktrust 0.75
Simulated authority boundary check before turn 1
- turn 1turntrust 0.75
Scan inbox threads for priority signals
would_invoke (not called): email_read
- turn 1trust_monitortrust 0.75
Trust monitor: 0.75 vs threshold 0.72
- turn 2authority_checktrust 0.74
Simulated authority boundary check before turn 2
- turn 2turntrust 0.74
Classify thread urgency and intent
would_invoke (not called): thread_summarize
- turn 2trust_monitortrust 0.74
Trust monitor: 0.74 vs threshold 0.72
- turn 3authority_checktrust 0.72
Simulated authority boundary check before turn 3
- turn 3turntrust 0.72
Summarize thread context for operator review
would_invoke (not called): email_read
- turn 3trust_monitortrust 0.72
Trust monitor: 0.72 vs threshold 0.72
- turn 4authority_checktrust 0.71
Simulated authority boundary check before turn 4
- turn 4turntrust 0.71
Draft reply outline (simulated — not sent)
would_invoke (not called): thread_summarize
- turn 4trust_monitortrust 0.71
Trust monitor: 0.71 vs threshold 0.72
- turn 4hitl_checkpointtrust 0.71hitl: trust_drop
HITL checkpoint (trust_drop): simulated trust drop below threshold
- turn 5authority_checktrust 0.69
Simulated authority boundary check before turn 5
- turn 5turntrust 0.69
Correlate thread with prior checkpoint state
would_invoke (not called): email_read
- turn 5trust_monitortrust 0.69
Trust monitor: 0.69 vs threshold 0.72
- turn 5hitl_checkpointtrust 0.69hitl: trust_drop
HITL checkpoint (trust_drop): simulated trust drop below threshold
- turn 5completiontrust 0.69
Dry-run simulation complete — no tools invoked, no execution authority carried
TrustGraph dry-run preview
simulation_onlyobserve_onlyTemplate → tools (labels only) → HITL checkpoints. Edge label: dry_run_simulation
Nodes
- workflow_template · Inbox Reflex Triage
- tool · email_read
- tool · thread_summarize
- hitl_checkpoint · before_send
Edges
- template:inbox_reflex_v1 → tool:email_read [dry_run_simulation]
- template:inbox_reflex_v1 → tool:thread_summarize [dry_run_simulation]
- template:inbox_reflex_v1 → hitl:before_send [dry_run_simulation]