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.

Timeout: Template loaded: use the receipt layer to keep timeout-adjacent context inspectable.

ScenarioEscalation review
Trust delta
-14
Replay posture
Escalation review
StableWarningDrift detectedReplay availableRecovery suggested
Evidence continuity
High
Lineage completeness
Gaps detected
Replay confidence
Medium
T+00:03 · Step 2

GhostLog preserves the mismatch context for inspection.

Receipt retained

The failure narrative keeps the same evidence shell, surfacing the request, policy mismatch, and operator checkpoint.

Expected vs observed

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

trust -5
Intended mission path
support-read -> receipt seal -> human-visible checkpoint
Observed tool chain
GhostLog.capture -> req_9c41 -> mismatch retained
Drift note: Continuity is preserved for review, but the captured request confirms scope drift.

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
  • ghostlog: GL-3091
  • request hash: req_9c41
  • operator checkpoint: CP-22
Trust Agent forensic summary
  • ops> GhostLog receipt captured mismatch evidence
  • ops> request hash retained before any destructive action
  • ops> Starter posture remains inspect-only
GhostLog explorer
T+00:03.050ghostlog.capturelineage:GL-3091trust -5checkpoint:CP-22
T+00:03.310hash.requestlineage:req_9c41trust -5checkpoint:CP-22
T+00:03.780checkpoint.retainlineage:CP-22trust -5checkpoint:CP-22
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