The claim, the rule, the trace.
Eight expense claims, six rules, two exceptions, one human override. Check a claim and watch the engine name the rule that fired — or flag the case it cannot decide.
The compliance checker lab
8 claims · illustrative policy · deterministic
EXP-1042
Client dinner · Cafe LumenRule trace
Policy switches
Decision
The trace will show every rule, the decisive rule, and any exception.
Manual override
Check a claim first. A human override is allowed after a rejection or a review, and it is logged with a reason.
Audit log
- No events yet. Checks, overrides, and resets appear here in order.
Precedence
- R3 prohibited category
- R5 duplicate claim
- R6 currency conversion
- R1 receipt threshold
- R4 travel pre-approval
- R2 meal cap
- R0 default accept
The first failing rule in this order is decisive, and a hard failure outranks a review. The order is a policy choice, not a logical consequence. E1 and E2 can waive exactly one failed rule each; they never create a new fact.
A verdict is a small proof — or a flag.
The rule that fired
EXP-1042 fails the meal cap at $62.00 per attendee against a $60.00 cap. With E1 on, the decision is accept with exception; with E1 off, it is reject R2. The trace shows both the failure and the exception, so a reader can see that the claim was not simply “fine.”
Review is a verdict, not a failure
EXP-1045 is a EUR 540.00 flight with no FX rate. The engine cannot compute the USD value, so it returns review rather than guessing. Strict mode converts that review into a rejection. The difference is a policy about missing data, not a difference in arithmetic.
The human act is logged
EXP-1046 is rejected by R4 for travel without pre-approval. Choose a reason and override it: the decision becomes accept · override, the audit log records the original rule and the reason, and the rule trace is unchanged. An override does not rewrite the policy; it records a person taking responsibility for an exception.
The six rules, the two exceptions, the $60 / $75 / $100 thresholds, and the eight claims are fixtures invented to make the mechanism visible. They are not a real expense policy, not legal or accounting advice, and not a model of any organization's controls. The lab has no learned extraction, no OCR, no currency service, no fairness metric, and no real audit trail storage. It isolates one mechanism: a decision that names the rule that produced it, keeps its exceptions visible, and records the human who overrode it.