What is a Dry Run?
A Dry Run tests how your rules evaluate a single set of input facts against a specific policy version — without any side effects. No execution logs are written, and dry runs are not billed or counted toward your plan’s usage. They have a separate per-minute limit. Think of it as a unit test for your rules.Dry Run vs Impact Simulation vs Decision Replay: A Dry Run tests one input you type for quick validation. Impact Simulation tests a version against hundreds or thousands of historical inputs and compares the results against a baseline version. Decision Replay re-evaluates one real past decision against a candidate version and diffs the outcome. Use Dry Runs while iterating, Simulation for regression testing before deployment, and Replay to answer “what would this exact production decision have been?”
When to Use Dry Run
- After creating or modifying rules — verify they match expected inputs
- Before publishing a DRAFT version — your safety net
- When debugging unexpected execution results — reproduce the scenario
- When comparing two versions with the same input (Dry Run Compare)
Running a Dry Run
Console
Navigate to a policy version → Dry Run tab → enter input facts → click Run.CLI
API
403 (A-002). See Scope.
Request metadata goes in contexts, as in an execution call. A dry run takes only facts, contexts, and includeDebugInfo, and a comparison takes only facts, contexts, versionIdA, and versionIdB. Any other field fails with 400 (C-001), even when it is empty. That includes the old name context.
Response
versionNo is the number of the version that ran, and latencyMs is how long the evaluation took, in milliseconds. inputContexts is the contexts you sent, or {} if you sent none. A dry run is not recorded, so the response has no traceId or executedAt.
State Layers
Thedata object exposes three layers of state explicitly — never merged into a single blob:
inputFacts— facts as received from the request, normalized.mutatedFacts— only the facts whose values were changed by rule actions.generatedVariables— variables created by rule actions that did not exist in the input. This includes per-action change keys:{targetVar}__deltacarries the signed change applied byMUTATE_FACT.
{ ...inputFacts, ...mutatedFacts, ...generatedVariables }. Keeping them separate makes audits, replays, and debugging deterministic — you always know which value originated where.
Execution Traces
Each rule that the engine evaluated produces one execution trace. The trace records the exact match expression that was checked and the actions it generated:policyGroupId, policyVersionId, ruleId) make every trace self-locating: you never need an external lookup to know which group, version, and rule produced it. The rule saw the facts in inputFacts, which the response gives once rather than repeating in each trace.
Decision Traces
While execution traces record what was evaluated, decision traces record what was selected. Each rule appears indecisionTraces with its final status and the reason code that produced it:
status is the top-level classification of the outcome; reasonCode is the specific cause within it. reasonDetail is nullable — it carries a human-readable message when the engine has additional context to surface (e.g., the specific limit value that was hit).
See Decision Trace in the Policy Rules reference for the full list of
status and reasonCode values and their mapping.Checking Requirements
Before running a Dry Run, check which facts the version expects:Dry Run Compare
Compare how two versions evaluate the same input facts, side by side.diff of their outputs:
diff lists only the keys whose values differ between the two versions. before is version A’s value and after is version B’s.
Limits
Each organization can run up to 60 evaluations per minute and up to 1,000 evaluations per day, shared by the calls below and by every console user and API key. These calls don’t use your plan’s TPS or monthly limits.
A call over the per-minute limit gets
429 (C-007). Wait a minute before trying again. A call over the daily limit gets 429 (AN-041). Try again after 00:00 UTC. To test many inputs at once, use Impact Simulation.
Values in the facts and contexts you send to a dry run have the same value limits as values in an execution call. A value over a limit fails with 400 (P-064).
A dry run request also has the same request body limit as an execution call. A larger body fails with 413 (C-011), and one nested deeper than the limit fails with 400 (C-012).
Next Steps
Impact Simulation
Test against hundreds of historical inputs at once.
Deployments
Deploy your validated version to production.

