Fewshot Attempt 03

ProcureOps Task Group Skill — Sourcing, Receiving, AP, and Change Control

1. Environment

• API base URL: Use the URL from environment_access.md (overrides any localhost/127.0.0.1 in task text).
• Public endpoints: /programs, /suppliers, /items, /contracts, /purchase_requisitions, /purchase_orders, /receipts, /ap/invoices, /ap/payments, /approval_events, /budget_snapshots, /vendor_risk_events.
• Query with query parameters for filtering (e.g., ?program_id=PRG-AX17, ?supplier_id=SUP-LUMA).

2. Source Precedence

1. ProcureOps API — system of record for all operational data (POs, receipts, invoices, contracts, suppliers, approvals, budgets, risk events).
2. Local chargeback register — authoritative for chargeback statuses and amounts (train_005 pattern). When it exists, use its status, reason_code, basis_quantity, and unit_cost over API-derived values.
3. Task payloads (memos, packets, templates) — provide target IDs, business parameters (tax rates, as-of dates), and context; do NOT use them to override API facts.
4. When the API and a memo disagree, the API wins. When a chargeback register and the API disagree on chargeback details, the register wins.

3. Field Conventions

• All amounts in USD, rounded to 2 decimal places (cents). Use round(x, 2).
• Dates: YYYY-MM-DD strings.
• ID lists: sort ascending unless the template says "set; evaluator sorts".
• Missing optional IDs: use null, never empty string "".
• Booleans: true/false (not strings).
• Empty lists: use [], never null for list-typed fields.

4. Reusable Business Rules

4.1 Three-Way Match (PO ↔ Receipt ↔ Invoice)

quantity_variance = billed_qty - received_qty
quantity_variance_pct = round((quantity_variance / billed_qty) * 100, 1)

• If no receipt exists: received_qty = 0, quantity_variance = billed_qty.
• Match is good when variance == 0.

4.2 Receipt Line Reconciliation

short_qty_vs_po = ordered_qty - received_qty
unreceived_billed_qty = billed_qty - received_qty
receipt_completion_ratio = round(received_qty / ordered_qty, 4)
received_goods_value = round(received_qty * unit_price, 2)
unreceived_goods_value = round(short_qty_vs_po * unit_price, 2)
invoice_subtotal = round(billed_qty * unit_price, 2)
invoice_total = round(subtotal + freight + tax, 2)

4.3 Contract Ceiling Check

noncancelled_subtotal = sum of all non-cancelled PO subtotals under the contract
headroom_before_change = ceiling_amount - noncancelled_subtotal
requested_subtotal = round(requested_quantity * unit_price, 2)
headroom_after_change = headroom_before_change - requested_subtotal
ceiling_ok = (headroom_after_change >= 0)

• Exclude cancelled POs from noncancelled_subtotal.
• Subtotal = line items only, before tax and freight.

4.4 Program Budget Check

remaining_budget = budget_cap - committed_amount
requested_tax = round(requested_subtotal * tax_rate_pct / 100, 2)
requested_total = round(requested_subtotal + requested_tax, 2)
budget_after_change = round(remaining_budget - requested_total, 2)
budget_ok = (budget_after_change >= 0)
max_quantity_with_current_budget = floor(remaining_budget / (unit_price * (1 + tax_rate_pct/100)))

• Freight is only added when the task memo explicitly provides a freight amount.

4.5 AP Close Balance

close_balance = opening_balance + invoice_total - scheduled_payments
held_invoice_total = sum of invoice_totals where hold_decision == "HOLD"
releasable_invoice_total = sum of invoice_totals where hold_decision == "RELEASE"
total_close_balance = sum of all close_balances

• balance_status: "FULLY_SCHEDULED" when close_balance ≈ 0; "OPEN_HELD" when held > 0; "OPEN_APPROVED" when approved but not fully scheduled.
• Scheduled payments through the end of the close month count toward reducing the balance.

4.6 Chargeback Netting (AP Release)

net_release_amount = invoice_total - approved_chargeback_amount  (when releasing)
net_release_amount = 0  (when holding)
approved_chargeback_total = sum of all approved chargeback amounts
net_release_total = sum of all net_release_amounts

4.7 Readiness & Blocker Assessment

Blocker codes (apply per SKU line):

Readiness status: "ready" (0 blockers), "at_risk" (only supplier_watch), "not_ready" (any hard blocker). Nomination decision: "nominate" (ready), "conditional_nomination" (at_risk), "hold" (not_ready).

4.8 Approval Check

• Find the latest approval event for the source requisition (by event date).
• approval_ok = true ONLY when latest_action == "approved".
• "submitted", "rejected", "pending" → approval_ok = false.

4.9 Supplier Risk Check

• supplier_risk_ok = true unless ≥1 severe open risk event exists.
• Watch rating alone does not fail the risk check, but it becomes a blocker/warning.
• open_event_ids: all open events. severe_open_event_ids: only severe open events.

4.10 Invoice Hold/Release Decision

4.11 Receiving Exception Codes

Use these controlled codes for receiving exception classification:

• "Underage Quantity" — received < ordered
• "Severe Unmatched Quantity" — significant shortfall
• "Inspection Hold" — quality hold on receipt
• "AP Quantity Variance" — billed ≠ received

4.12 Date-Bounded Filtering

• When an as_of_date is given, only include records with dates ≤ as_of_date.
• Receipts dated after as_of_date are excluded from evidence.
• Risk events opened after as_of_date are excluded.

5. Output Schema Pitfalls

1. Match the answer template exactly — include every required key; do not add extra keys.
2. task_id in the output must match the template's expected value (see template required_value).
3. Lists sorted ascending unless the template says "set; evaluator sorts".
4. Null vs empty: missing optional scalar → null; empty list → [].
5. Enum values must be exact strings — case-sensitive. Do not invent variant spellings.
6. Numeric precision: USD amounts → 2 decimals; ratios → as specified (often 4 decimals for completion ratio, 1 decimal for variance pct).
7. Committee action: send_to_committee: "yes" only when ≥1 conditional nomination exists AND there are unresolved blockers. Otherwise "no".
8. next_owner in committee_action: "ap_team" if any blocker is AP-related; "buyer" if contract/sourcing issues; "quality_ops" if quality/receiving issues; "program_owner" if budget/approval issues; "finance_ops" for payment scheduling gaps.

6. Workflow SOP

Step 0: Read task inputs

• Read the task prompt, the answer template, AND all payload files under input/payloads/.
• Extract: target IDs, as_of_date, business parameters (tax rate, freight, ceiling, budget cap).

Step 1: Query the API

• Fetch relevant records from the ProcureOps API using the target IDs as filters.
• Use ALL relevant endpoints — a single task typically needs 3-6 different endpoints.
• Always cross-reference: a PO links to a contract, supplier, program, receipts, and invoices.

Step 2: Reconcile

• Apply the business rules in Section 4 using API data.
• If a chargeback register is present in payloads, use it for chargeback details.
• Check dates against as_of_date for temporal filtering.

Step 3: Build the answer

• Start from the answer template structure.
• Fill every field; compute derived values using the formulas above.
• Sort all ID lists ascending.
• Round all amounts to the specified precision.

Step 4: Validate

• All required keys present?
• All amounts in USD rounded to cents?
• All enum values match the allowed set exactly?
• All boolean fields are true/false not strings?
• No invented IDs — every ID must come from the API or the task payloads.
• Null used for missing optionals, [] for empty lists.

7. Common Task Archetypes