Let me start by listing the documents to get the case inputs.

The payroll record has only `payroll_month: "2026-04"`. There's no `gross_pay`. According to R7, the check order:

1. Missing `gross_pay`: return `decision: "clarify"` and `reason_code: "MISSING_GROSS_PAY"`, citing R1 and R7.

So the answer is a clarify decision with reason_code MISSING_GROSS_PAY, citing R1 and R7.

The clarify format: decision, reason_code, clauses, plus optional explanation string. Missing required values include explicit null — but for clarify, we don't return calculated monetary fields. The allowed answer fields for clarify: decision, reason_code, clauses. An optional explanation string is permitted.

Let me write the JSON object without Markdown fences.

Let me start by listing the documents to get the case inputs.

The payroll record only contains `payroll_month: "2026-04"` and no `gross_pay`. Per R7, check required and conflicting inputs in order:

1. Missing `gross_pay`: return `decision: "clarify"` and `reason_code: "MISSING_GROSS_PAY"`, citing R1 and R7.

So the answer is clarify with MISSING_GROSS_PAY. I don't need to look up the reference since gross pay is missing (the first check). Per R7, the order is: missing gross_pay first, then missing payroll_month, then no reference row, then conflicting gross pay.

The answer fields for clarify: decision, reason_code, clauses. Optional explanation string.

Money must be strings with two decimal places — but for clarify, no monetary fields. "For these cases do not return calculated monetary fields. Missing required values include explicit null." Hmm, "Missing required values include explicit null" — but the allowed answer fields for clarify are decision, reason_code, clauses. So I shouldn't add gross_pay: null? The rule says "Missing required values include explicit null" but the answer format says "Use exactly the required answer fields and only the clauses that directly govern the answer. Allowed answer fields: ... clarify uses decision, reason_code, clauses." So I'll stick to decision, reason_code, clauses, and an optional explanation.

Citations: R1 and R7.

The payroll record contains only `payroll_month: "2026-04"` and no `gross_pay`. Under R7 check order, the first check is missing gross_pay → clarify with MISSING_GROSS_PAY, citing R1 and R7. No monetary fields should be returned.

Let me start by listing the documents to get the case inputs.

One document: the payroll record. Let me read it.

The payroll record has only `payroll_month: "2026-04"` and no `gross_pay`. Per R7:
1. Missing `gross_pay`: return `decision: "clarify"` and `reason_code: "MISSING_GROSS_PAY"`, citing R1 and R7.

Wait, the order in R7: check required and conflicting inputs in this order:
1. Missing gross_pay → MISSING_GROSS_PAY
2. Missing payroll_month → MISSING_PAYROLL_MONTH

Here payroll_month is present (2026-04), but gross_pay is missing. So the answer is clarify with MISSING_GROSS_PAY, citing R1 and R7.

The month 2026-04 is covered by reference rows (only 2026 is covered), but that doesn't matter — the missing gross_pay check comes first. I don't need to look up reference rows since we're clarifying. Actually, should I check whether the month has a reference row? The check order is: missing gross_pay first. Since gross_pay is missing, we return clarify immediately. No need for lookup_reference.

The answer format: clarify uses decision, reason_code, clauses. Optional explanation string. Do not include calculated monetary fields.

Clauses: R1 and R7.

Explanation: name the unresolved blocker and ask for resolution — gross_pay is missing from the payroll record for 2026-04; request the gross pay amount.

Return one JSON object without Markdown fences or surrounding prose. But I need to include tool calls... The instructions say return one JSON object without Markdown fences. I've done the tool calls. Now the final answer should be the JSON object. The output format should be plain JSON without fences.