Reflect 3 Attempt 01

ProcureOps Task Group 006 — Reusable Solve Skill

Environment & Sources

1. API base URL: Use the environment-provided GDPEVO_ENV_BASE_URL. Never use localhost unless that is the actual remote target.
2. API is source of truth: Task memos/payloads provide anchors (IDs, scope) and business rules. Operational data (quantities, prices, statuses, dates, risk events) comes from the API only.
3. Endpoints used: /programs, /suppliers, /items, /contracts, /purchase_requisitions, /purchase_orders, /receipts, /ap/invoices, /ap/payments, /approval_events, /budget_snapshots, /vendor_risk_events.
4. API data is flat: All endpoints return full result sets — use client-side filtering, not query params.

Field Conventions

5. Sort ascending for: SKU lists, ID lists (PO, receipt, invoice, supplier, risk event), nomination lines (by SKU), invoice_decisions (by invoice_id), vendor_balances (by supplier_id), program_summary (by program_id), payment hold/release queues.
6. List fields are sets: Unless sorting is specified in the template, the evaluator sorts. Still produce deterministic output.
7. USD to cents: All currency amounts rounded to 2 decimal places with standard rounding. Include .00 suffix for whole-dollar values.
8. Precision by field: Ratios (receipt_completion_ratio) to 4 decimal places; percentages (quantity_variance_pct) to 1 decimal.
9. task_id format: Use the exact value from the answer template's required_value or task_id field. Match the template convention.

Date & Time Filtering

10. as_of_date is a hard cutoff: Exclude receipts, invoices, risk events, and approval events dated after the as_of_date.
11. Receipts dated after as_of_date are invisible: They don't count toward receipt_evidence_ids or quantity_received calculations.
12. Scheduled payments through the cutoff: Only payments with scheduled_date on or before the stated cutoff date (e.g., 2026-06-30) count.

Arithmetic & Business Rules

Budget & Contract

13. Budget headroom = budget_cap − committed_amount (from program endpoint or budget snapshot — they agree).
14. Contract noncancelled_subtotal: Sum subtotal of all POs under the contract with any status EXCEPT cancelled.
15. Contract ceiling exposure: Before tax and freight. Only the line subtotal counts against the contract ceiling.
16. Program budget exposure: Line subtotal plus estimated tax. Freight only if the change memo/payload explicitly provides it.
17. Tax rate: From the task memo's tax_rate_percent or business_controls (typically 7.25%). Apply to subtotal, round to cents.
18. requested_total = requested_subtotal + requested_tax (no freight unless provided).
19. budget_after_change = remaining_budget − requested_total.
20. max_quantity_with_current_budget = floor(remaining_budget / (unit_price × (1 + tax_rate))).

Receipt & Invoice Reconciliation

21. Received vs ordered: short_qty_vs_po = ordered_qty − received_qty (positive = shortage).
22. Billed vs received: unreceived_billed_qty = billed_qty − received_qty.
23. Receipt completion ratio = received_qty / ordered_qty (to 4 decimal places).
24. Three-way match: PO unit_price vs contract unit_price vs invoice unit_price. contract_price_match = true when contract and invoice prices agree.
25. Invoice exception codes: Derive from discrepancies — INVOICE_QTY_EXCEEDS_RECEIPT when billed > received; PARTIAL_RECEIPT when PO status signals partial; SUPPLIER_WATCH_RISK when supplier has watch rating with open risk; PRICE_MISMATCH when prices differ; DAMAGE_REJECTION when rejected_qty > 0.

Payment & Close

26. opening_balance for close slices: Use the memo-stated value (often 0.00 for target-invoice-only slices).
27. scheduled_payments: Sum of payment amounts for TARGET invoices only, scheduled through the cutoff date. Do NOT include payments for non-target invoices even if the same supplier.
28. vendor close_balance = opening_balance + invoice_total − scheduled_payments.
29. net_balance_impact (per invoice) = invoice_total − scheduled_payment_amount.
30. total_close_balance = sum of vendor close_balances across the slice.
31. held_invoice_total: Sum invoice_total for invoices with hold_decision = HOLD.
32. releasable_invoice_total: Sum invoice_total for invoices with hold_decision = RELEASE.
33. balance_status: FULLY_SCHEDULED when close_balance = 0; OPEN_HELD when held_invoice_total > 0; OPEN_APPROVED when releasable but not yet scheduled.

Chargebacks & Release

34. Chargeback amount = basis_quantity × unit_cost (from the chargeback register, not the API).
35. net_release_amount = invoice_total − approved_chargeback_amount (when releasing net of chargeback).
36. approved_chargeback_total = sum of approved_chargeback_amount across all invoices.
37. pending_chargeback_total = sum of pending_chargeback_amount across all invoices.

Supplier Risk

38. Risk events are supplier-scoped: Include all open or monitoring events for the supplier, regardless of which PO/SKU they relate to.
39. Severe events: Only events with severity: "high" count as severe. medium and low do not.
40. Supplier risk_rating contexts: watch → supplier_watch blocker. high → supplier risk blocker. Both are advisory unless a severe open event exists.

Approval

41. Check only good_actions: From business_controls. Normally only "approved". "submitted", "returned", "escalated" do NOT count as approved.
42. Latest event only: Sort by event_date descending; the most recent event's action determines approval_ok.

Nomination & Package

43. Package line SKUs: From the memo anchors, sorted ascending.
44. package_po_ids: Only the anchor POs named in the memo, not all program POs for the SKU.
45. commercial_basis_id: The contract_id from the PO, or null if the PO has no contract.
46. Blockers for nomination lines: Check each of: missing_contract (no active contract), supplier_watch (risk_rating=watch), open_supplier_risk (any open risk event), ap_hold (invoice status=on_hold with hold_code), pending_receipt (received < ordered), late_due_date (PO due_date < as_of_date).
47. overall_readiness: not_ready if any line is not_ready; at_risk if any line is at_risk but none not_ready; ready only if all lines are ready.

Change Control

48. Decision precedence: Check contract → budget → approval → supplier risk. Combine blockers if multiple fail.
49. Contract check uses the contract's own POs: Noncancelled_subtotal and included_po_ids are scoped to the specific contract_id.
50. Program budget check: Uses the program's budget_snapshot.

Receiving Exceptions

51. exception_codes: Underage Quantity (received < ordered), AP Quantity Variance (billed ≠ received/ordered), Inspection Hold (receipt status = inspection_hold), Severe Unmatched Quantity (large billed vs received gap).
52. chargeback_status: From the local chargeback register: approved, pending_quality_review, or not_applicable.
53. resolution_status: net_release_ready when approved chargeback covers the gap; hold_for_quality_review when pending; accepted_no_receiving_exception when no exceptions; missing_receipt when no receipt exists.

Output Schema Pitfalls

54. Do not fabricate IDs: Every ID in the answer must come from the API responses or task payloads.
55. Empty arrays not null: Use [] for empty ID lists, never null.
56. Booleans: Use JSON true/false, not strings.
57. Null vs empty: null for missing references (e.g., no contract, no receipt, no hold_code). [] for empty lists.
58. Evidence section: Include the actual API record IDs you queried (POs, receipts, invoices, contracts, risk events). Also list the task payload files reviewed.
59. reason_codes: Only include codes that genuinely apply. Use alphabetical sort.

Task-Specific Patterns

60. Receiving closeout (train_002): One batch, one PO, one receipt, one invoice. Focus on the quantity variance and supplier risk context. Exception codes include the PO-level status flags.
61. AP close desk (train_003): Slice is limited to named invoices. Opening balances are 0.00 for the slice. Scheduled payments only count for target invoices through the cutoff window.
62. Change control (train_004): Separate contract ceiling check (before tax) from program budget check (after tax). Contract noncancelled_subtotal is across all non-cancelled POs under that contract.
63. AP release (train_005): Chargeback amounts from the local register. Exclude receipts for the same PO that aren't linked to the invoice (excluded_same_po_receipt_ids). Followup actions should include hold_luma_duplicate_receipt_for_separate_invoice when a receipt exists for the PO but the invoice doesn't reference it.