Provenance timeline
Records.
Every run, on the record.
One immutable timeline of every Reaction and action. Filter to what you care about, then drill into a single run to see exactly which atoms fired, which model served them, which filters matched, which policies hit, and how the risk score added up. Reproducible from provenance. Every time.
The timeline (left) and a single run drilled to its atom trace, model usage, gates, and findings (right).
What's on the record
Enough to reproduce any answer.
Every atom, in order
The full step trace with status, latency, and token + cost telemetry per atom.
Which model served it
The resolved LLM profile and scope tier for every step. Tied to the spend dashboard.
Prompts + outputs
The actual prompt sent and the model's response, captured for inspection.
Filter + trust events
Every redaction, block, and injection detection that fired on the run.
Policies + findings
Which governance policies evaluated and which findings were raised.
Risk breakdown
The composite score and the per-factor contributions that produced it.
Why it matters
One timeline for ops, audit, and trust.
Operations
Spot stuck and failed runs, retry or force-fail, and see live in-flight Reactions. No log spelunking.
Compliance & audit
Hand an auditor a single, immutable, hash-chained trail instead of stitching screenshots together.
Trust & explainability
Every answer is defensible because its provenance is one click away. Who, what, which model, which rules.