Procureops Erp Answers

ProcureOps ERP answer builder

You are turning a task prompt plus a local memo/payload into one JSON object that matches a provided answer_template.json, using the live ProcureOps ERP API as the system of record. The five task families you will see are nomination readiness, receiving/AP reconciliation, AP close, change-control decision, and AP release/chargeback netting — but the mechanics below transfer to any unseen task in this domain. Lead with the rules, not with memorized answers.

Golden rules (true for every task)

1. The API is the source of truth; the local memo is a pointer, not data. Memos and exports name the anchors (program, POs, invoices, receipts) and give business controls (tax rate, cutoff date, "exclude cancelled POs", opening balance). Whenever a quantity, price, status, or amount differs between a stale local file and the API, the API wins. Pull every figure you put in the answer from the API record, not from the memo's narrative.
2. Match the template exactly. Output only the JSON object, with the keys, nesting, and enum spellings the template specifies. Copy the literal task_id / required_value the template names. Do not add commentary keys.
3. Money is rounded to cents (2 decimals) unless a field says otherwise (e.g. receipt_completion_ratio precision 4, quantity_variance_pct 1 decimal). Use the API's own total/subtotal/tax/freight fields rather than recomputing them — they are authoritative and already rounded.
4. List fields are sets (order-insensitive, deduped) unless the template says "sorted ascending" / "ordering: ... ascending" / "alphabetical". When it says sorted, sort ascending as strings. When it says "set; evaluator sorts", you may still emit sorted for readability.
5. Everything is scoped to an as-of / cutoff date. Read it from the prompt ("as of YYYY-MM-DD"), the memo, or the template's date field. Exclude records dated after it. "Open or monitoring as of date" means status filtering AND event_date <= as_of.
6. Scope to the named targets only. If the memo lists specific invoices / POs / SKUs, answer for exactly those, in the template's order. Do not pull in the supplier's other invoices unless the task asks for a vendor-wide balance.

Step-by-step SOP

1. Read the prompt, the memo/payload, and answer_template.json in input/payloads/. The template is your contract: list its required keys, note every enum's allowed values, note each field's rounding/ordering, and note the literal task_id. Identify which task family this is from the keys (see references/task_families.md).
2. Extract anchors and controls from the memo: program_id, target POs / invoices / receipts / SKUs, the as-of/cutoff date, the tax rate, the opening balance, and any explicit rule ("exclude cancelled POs", "payment scheduled through DATE reduces balance", "approved actions = [approved]").
3. Pull the live records. For each anchor, GET it by id; for set-based needs (all POs on a contract, all receipts on a PO, all payments for a supplier, all risk events for a supplier), use list endpoints with query filters. See the API cheatsheet below.
4. Compute the derived values using the business rules in references/business_rules.md. The recurring computations — budget headroom, contract ceiling headroom, qty reconciliation, variance, chargeback netting, open-risk filtering — are defined there with exact formulas.
5. Decide the enums (decision / disposition / hold_decision / reason & blocker codes) from the rules. Decisions are deterministic functions of the computed checks; do not editorialize.
6. Assemble and validate with scripts/validate_answer.py (checks the JSON parses, top-level keys/enums look right, money is 2dp, sorted lists are sorted). Fix anything it flags. Emit only the JSON.

ProcureOps API cheatsheet

Base URL comes from the prompt (commonly http://127.0.0.1:8056, mirror at :8006; both serve identical data). Read-only GET, JSON.

• Fetch one record: /<collection>/<id> returns the bare object.
• List with filters: /<collection>?field=value returns {"count": n, "results": [ ... ]}. Filters match a field exactly, case-insensitive, including fields nested inside list values.
• Date range: start= and end= filter the collection's primary date field (receipts.receipt_date, ap_invoices.invoice_date, payments.scheduled_date, vendor_risk_events.event_date, etc.).

Collections: /programs, /suppliers, /items (by sku), /contracts, /purchase_requisitions, /purchase_orders, /receipts, /ap/invoices, /ap/payments, /approval_events, /budget_snapshots, /vendor_risk_events. /manifest gives record counts and anchor ids; /health confirms it is up.

Filter gotcha: approval events are keyed by object_id + object_type, NOT requisition_id. Use /approval_events?object_id=REQ-... (filtering by requisition_id returns 0). When a filter unexpectedly returns count: 0, re-check the actual field name on a sample record before concluding "no data".

Record shapes and the full field list are in references/api_fields.md.

Derived-value quick reference (full detail in references/business_rules.md)

• Budget headroom / remaining budget = budget_cap − committed_amount (from the program or its budget_snapshots record; they agree). NOT minus pending_invoice_amount.
• Contract ceiling headroom = ceiling_amount − Σ(subtotal of non-cancelled POs on that contract). Exclude status == "cancelled" POs. A requested line fits if headroom_after_change = headroom_before − requested_subtotal >= 0.
• Quantity reconciliation for a PO line: received_qty = sum of quantity_received across that PO's receipt lines (in scope per the task); short_qty_vs_po = ordered − received; unreceived_billed_qty = max(0, billed − received); quantity_variance = billed − received; quantity_variance_pct = round(variance / ordered_qty * 100, 1).
• No receipt → quantity_received = 0, variance = billed, pct = 100; the invoice holds with reason NO_RECEIPT / no_receipt_on_po.
• Open / monitoring risk as of date = vendor_risk_events for the supplier with status in {open, monitoring} and event_date <= as_of. "Severe" = severity == "high". Closed events are excluded.
• Scheduled payment reduces balance only if its scheduled_date <= the memo's cutoff (commonly month-end). Sum matching payment amounts.
• Chargeback netting: net_release = invoice_total − approved_chargeback where approved chargeback = basis_quantity × unit_cost summed over register rows with status == approved. Pending (e.g. pending_quality_review) chargebacks do NOT net — they force a hold.
• Duplicate / out-of-scope receipts: if a PO has more than one receipt but the task scopes one, list the others under the "excluded same-PO receipt ids" field and hold them for their own invoice — do not net them here.

Common misjudgments to avoid

• Using the memo's numbers when the API's differ (the API wins).
• Recomputing invoice/PO totals by hand and getting penny drift — use the record's total/tax/freight fields directly.
• Forgetting freight: invoice total includes freight; PO total may not.
• Subtracting pending_invoice_amount from the budget cap — headroom is just cap − committed.
• Counting cancelled POs in contract usage, or counting receipts/risk events dated after the as-of date.
• Treating a submitted/escalated/returned approval as approved — only the latest event with action == approved clears the approval gate.
• Letting a pending (not yet approved) chargeback net the invoice down — pending means HOLD, only approved chargebacks reduce the release amount.
• Pulling a supplier's unrelated invoices into a scoped, named-targets answer.
• Emitting prose outside the JSON, or dropping/renaming a required key.

When you need the precise decision tables, enum-to-condition mappings, and the per-task field walkthroughs, read the two reference files. They are the heart of this skill — open them before composing the answer.