Interactive replay prototype

Install -> Replay -> Inspect

A replay-safe, client-side click-through for the centerpiece demo. Walk the success path, then toggle into a failure inspection where the evidence chain surfaces the mismatch.

Start hubFirst replayDocs
Illustrative guided demo only. All events, receipts, trust deltas, and summaries are simulated client-side fixtures, not live customer telemetry. Starter posture is inspect-only where referenced.

Developer replay templates

Jump into preset, simulated replay postures. Templates only preselect scenario copy and starting step.

?template=
Failed tool chainPolicy mismatchRetry stormMulti-agent divergenceTimeoutRecovery chain
Slice AB · inspect a failure

Policy mismatch surfaced

A canned failure path shows how a mismatch, continuity break, and negative trust delta become inspectable evidence.

Failed tool chain: Template loaded: inspect the first tool-selection break before the chain compounds.

ScenarioEscalation review
Trust delta
-14
Replay posture
Escalation review
StableWarningDrift detectedReplay availableRecovery suggested
Evidence continuity
Partial
Lineage completeness
Gaps detected
Replay confidence
Medium
T+00:00 · Step 1

Agent requests vendor.update outside the approved mission scope.

Mismatch suspected

The simulated event swaps from a read-only lookup to a write-capable vendor update, triggering a policy check.

Expected vs observed

Simulated replay diff for this step. Values are illustrative fixtures only.

trust -3
Intended mission path
support-read -> vendor.lookup -> policy P-104
Observed tool chain
runtime.event -> vendor.update -> policy.mismatch
Drift note: Divergence begins at tool selection: observed write action exceeds read scope.

Why trust broke

  • Divergence began when the runtime selected vendor.update instead of the vendor-read lookup path.
  • Continuity broke at the policy edge P-104 because the observed tool chain crossed from read scope into write authority.
  • Replay confidence shifted negative after GhostLog GL-3091 and drift scoring agreed on the same authority boundary break.
Evidence chain
  • tool: vendor.update
  • scope: support-read
  • policy: P-104 mismatch
Trust Agent forensic summary
  • ops> runtime.event requested write-capable tool
  • ops> declared mission scope allowed read-only inspection
  • ops> policy mismatch marked for replay review
GhostLog explorer
T+00:00.090runtime.eventlineage:LN-failure-01trust -1checkpoint:CP-21
T+00:00.280tool.intentlineage:vendor.updatetrust -2checkpoint:CP-21
T+00:00.640policy.mismatchlineage:P-104trust -3checkpoint:CP-21
Same mission, different runtime
OpenAI+6 guided / -12 failure

Fast policy echo, review still required

Same canned mission, simulated as a managed runtime handoff.

Ollama+4 guided / -9 failure

Local receipt chain, slower stabilization

Same canned mission, simulated as a local runtime replay.

Shareable replay snapshot

Exports a static JSON bundle with steps, posture chips, truncated GhostLog events, diff summary, and timeline.

Simulated click-through. No live telemetry, no customer data, no autonomous action.

Starter framing: inspect-only. This prototype previews the replay and failure-inspection workflow while full Runtime Event Explorer acceptance criteria remain deferred.

Start hubFirst replayDocs