AI Decision State & Evidence Monitor

AI Decision State & Evidence Monitor is a deterministic, stateless comparison Actor for workflows that need to know whether a previously justified structured decision still holds after evidence, assumptions, or blockers change.

It is designed for agentic workflows that need explicit decision-state deltas and reproducible checkpoints — not for website/page monitoring.

How a monitor sequence works

On the first run, provide only request. Decision Monitor creates revision 1 of the checkpoint and an idempotency record. This is a baseline, so outcome is BASELINE_CREATED, there is no comparison report, and no pay-per-event charge is emitted.

On a later run, pass a new request plus the previous checkpoint. Decision Monitor compares the prior and current structured decision snapshots and evidence inventories, returns outcome: COMPARED, writes the next checkpoint, and emits the decision-monitor-comparison billing event.

A comparison is billable even when the legitimate result is no change. The paid work unit is the deterministic comparison, not the existence of a transition.

For an exact retry, pass the same request together with its previously returned prior_record. When the request fingerprint matches, the Actor returns outcome: REPLAYED and does not charge again. Reusing the same idempotency key with a different payload is rejected.

What it detects

A comparison report can identify:

• decision transitions;
• changed assumptions;
• added or removed blockers;
• added, removed, or changed evidence IDs;
• unchanged evidence count;
• changed source IDs and URLs;
• deterministic reasons for the observed delta.

What it does not do

Decision Monitor does not crawl websites, watch page selectors, fetch new evidence, schedule itself, infer a decision policy, or maintain hidden cross-customer persistence. It only compares caller-supplied structured state.

The caller or orchestrator owns persistence and passes the prior checkpoint explicitly. The beta contract assumes a single active writer per monitor_id; expected_checkpoint_id can be used as a stale-write guard.

Input

The top-level input contains request plus an optional previous checkpoint and optional prior_record for exact replay. The request contains a monitor_id, idempotency_key, optional expected checkpoint ID, current decision snapshot, and current evidence inventory.

Output

One dataset item contains:

• outcome: BASELINE_CREATED, COMPARED, or REPLAYED;
• comparison report when applicable;
• the next checkpoint;
• idempotency_record;
• billing_event only for a newly performed comparison.

Pricing

The beta price is USD 0.01 per new decision-monitor-comparison. Baseline creation is free. An exact idempotent replay is free. A new comparison that confirms no change is still charged once because the comparison itself is the paid unit.

Typical uses

Use Decision Monitor after a grant/tender qualification, research conclusion, operational assessment, or other structured agent decision when later evidence or assumptions may invalidate the original justification.

Limitations

The Actor's causal reasons describe deterministic changes in the supplied snapshots and evidence. They are not a claim that an external real-world event caused the transition. The Actor does not arbitrate concurrent writers or authorize downstream side effects.

Suez Evidence & Decision Infrastructure for AI Agents

Decision Monitor is the change-detection layer of the Suez evidence-and-decision tool family. Agent Readiness audits machine-facing publication, Research Verification checks evidence sufficiency, Task Feasibility preflights execution, Opportunity Decision provides a bounded qualification decision, and Decision Monitor determines whether that decision still holds as structured state changes.