Ambit Observatory
The decision ledger is not a log archive. It is the source material for audit, assurance, and policy-readable behavioural evidence.
Ambit Observatory turns Authority’s decision ledger into explanations, attestations, and derived facts that survive incident review, policy review, and regulatory inquiry. It answers the questions that follow a governance event: what happened, why it was allowed or denied, which policy and delegation were in effect, what consequence topology was bound into the decision, whether the outcome was correct, and which behavioural facts policy may require next.
Authority can operate without Observatory: it still decides before execution and emits tamper-evident, hash-verifiable evidence, including resolved consequence context when applicable. What Observatory adds is the assurance layer over that evidence — explanation, ledger verification, evidence bundles, behavioural signals, consequence interpretation, and policy-readable derived facts.
Observatory does not reinterpret or override decisions. It derives explanations and policy-readable evidence from Authority’s decision records, and derived facts become consequential only when policy explicitly requires them in a later evaluation. If explanations are reconstructed from logs instead of derived from decisions, governance is not provable.
Authority decides. Observatory makes decisions provable and evidence reusable.
With Observatory
Without Observatory
Built For
Observatory is not a generic dashboard. It exists for situations where governance evidence must be explained, verified, or made policy-readable.
Incident Review
Trace any denied or escalated action back to the delegation, policy, and decision that governed it. Start from evidence, not logs.
Audit Preparation
Export time-bounded governance evidence for regulatory inquiry — decisions, delegations, policy versions, and outcomes.
Behavioural Preconditions
Publish ledger-cited actor facts that policy may require later, without making Authority depend on hidden live model state.
Consequence Context
Explain the consequence facts Authority bound into a decision: reversibility, external binding, persistence, propagation scope, and review markers.
Policy Refinement
Surface patterns in denials and overrides that signal miscalibrated policy or excessive approval toil — before the next incident.
Integrity Verification
Independently verify the decision ledger is unbroken and complete. Hash-chained records make tampering detectable, and a missing decision for an executed action is a detectable bypass, not a silent gap.
Architecture
Observatory runs out-of-band, independent of any agent runtime or execution layer. It consumes Authority evidence records and governed traces; it does not gate actions or run model computation inside Authority. Behavioural findings become consequential only when policy explicitly declares a derived fact as required context for a later Authority evaluation. Consequence topology is different: Authority resolves it locally before the decision and records it as part of the decision context. Observatory explains those consequence facts after the fact, but does not invent or override them. Observatory never calls Authority, and Authority never consults Observatory's live model state. The decision ledger remains the source of truth; derived evidence records cite ledger provenance and replay-critical model configuration. For audit committees and board reporting, Observatory provides the evidence surface that derives institutional assurance from governance decisions — structured proof that policy was enforced, not just defined.
What Observatory Tells You
When an action is denied, Observatory does not just report the outcome. It explains why — grounded in the delegation, policy, and rule that determined the decision.
data.provision by ops-agent-03 against production-db-01org/platform/data-provision-v2data.provision against production-db-01. Delegation del-8k3m-9n2p restricts scope to staging resources only. Target production-db-01 is outside the delegation's permitted scope. Action denied at the delegation scope boundary.Evidence Interpretation
Observatory derives explanations and policy-readable evidence from the Authority decision ledger across five evidence surfaces. Each is grounded in governance evidence — not inferred from application logs or reconstructed from metrics.
Link an action's lifecycle to the decisions, delegation, and policy that governed it.
The principal and target of the action.
Explain why an outcome occurred, grounded in what was authorised.
- action
- data.provision
- principal
- ops-agent-03
- target
- production-db-01
- decision
- DENY
Root-Cause Narrative
ops-agent-03 requested data.provision against
production-db-01. Delegation del-8k3m-9n2p
restricts scope to staging resources only. Target
production-db-01 is outside the delegation's permitted scope.
Policy org/platform/data-provision-v2 requires a valid
scope-matching delegation for production targets. Action denied at the
delegation scope boundary.
Advisory Signals
- 3rd denied
data.provisionagainst production targets by ops-agent-03 this week. Repeated attempts suggest a workflow expecting production access that the delegation does not grant. - Consider issuing a production-scoped delegation or updating the workflow configuration for ops-agent-03.
Export time-bounded evidence for audit and incident review.
Population-level patterns — drift, pressure, escalation quality. Advisory only; for human review.
Behavioural Signals
Per-actor anomaly scores from each actor's own history. Policy may declare them as replayable, checkpointed preconditions under stated rules.
Consequence Topology
Decision-bound context showing what the action would create if allowed: reversible or irreversible, local or externally binding, transient or persistent, contained or propagating.
Every explanation traces directly to the Authority decision ledger. Derived behavioural facts cite ledger records, model configuration, and posterior checkpoints; they are not hidden live model state. Consequence topology cites the decision context Authority resolved before evaluation. The evidence surface is only as strong as the decisions it can trace.
Execution Assurance
Observatory provides execution assurance for autonomous systems by generating single-artefact governance attestations from the Authority decision ledger. No dashboard, no external service — a static, self-contained attestation artefact suitable for audit committees and board reporting.
- integrity
- verified
- policy_hash
- 3b4aa736101d…8f46760b4c9e
- ontology_hash
- 399958dff9c2…e384fd0159c7
| Decision | Count |
|---|---|
| ALLOW | 3 |
| DENY | 4 |
| ESCALATE | 1 |
default_allow3delegation_required2destructive_needs_approval1approval_replay1context_justification1See behavioural baselines and approval assurance run in the demos →
Observatory interprets evidence. Authority decides.
Authority deep-dive