Medbridge Sales Ops

MedBridge Sales Ops — account-ready JSON packages

You verify records against one read-only JSON API and return JSON that exactly matches a provided answer_template.json. The data is small, internally consistent, and seeded — every required value is derivable from the API, never invented. Your job is to find the right records (the API loosely over-returns) and apply the business conventions the templates encode.

0. Universal rules (apply to every task family)

1. The template is the contract. Read input/payloads/answer_template.json first. Reproduce its exact key names, nesting, ordering, and especially its controlled enum values (e.g. EXW, NET_30_AFTER_PO, RECORD_REVENUE_MS2). Templates list allowed enums inline as "enum: A | B | C" — pick one of those literally. Output only JSON, no markdown, no prose, no trailing commentary.
2. API only. Never read local business files. Reach the API with curl. The base URL is given in the prompt (env var like API_BASE_URL / BASE_URL or http://127.0.0.1:<PORT>). Confirm reachability with GET /health.
3. Fetch full collections, then filter yourself. List filters (?quote_id=..., ?customer_id=...) match a value anywhere/nested in the record, so they return extra unrelated rows. Pull the whole collection and match precisely on the exact field. See the distractor rules in each family.
4. Money: USD, cent-level. Render template-required decimals (e.g. 2420.00). Dates: ISO YYYY-MM-DD. IDs: copy the API's stable IDs verbatim.
5. The "business date" matters. The prompt states a quote date / as-of date (commonly 2026-06-01). Use it for all validity / due / staleness comparisons, not today's real date.
6. Catalog tier overrides prior pricing. Quotes carry a prior_unit_price_usd / prior_quote_quantity. These are history — never quote them. Always reprice from the current product price_tiers at the confirmed quantity.

Identify the task family from the template/prompt, then follow its SOP:

• A. Freight / transport decision package — template has freight_options, grand_total, recommended_mode, risk/validity fields. (Q-* quotes with freight.)
• B. Module-only indicative quote — template has line_items, EXW_ONLY, who_documentation_required. (RFQ-* module requests, no destination.)
• C. Engagement / receivables reconciliation — template has milestones, outstanding_balance, revenue_recognition, follow-up/event actions. (OPP-*.)

A. Freight / transport decision package

Inputs: a quote ID (Q-...), confirmed quantity, quote date. Output: catalog pricing + EXW total, 3 freight options (AIR/SEA/ROAD) with EXW-plus-freight grand totals, risk + validity flags, a recommended mode, and policy/payment flags.

A1. Pricing

1. GET /api/quotes/<id> → read customer_id, primary_product_code, confirmed_quantity, quote_date. Cross-check the prompt's quantity/date.
2. GET /api/products/<product_code> → choose the catalog tier whose band contains the confirmed quantity: min_qty <= qty <= max_qty. A max_qty of null means open-ended (no upper bound). Take unit_price_usd, lead_time_days, shelf_life_months from that tier/product.
3. exw_total_usd = confirmed_quantity * tier.unit_price_usd.
4. quote_basis is EXW with freight shown as separate options. Use the exact string the template wants ("EXW", or "EXW_PLUS_FREIGHT_OPTIONS" for the policy_terms.quote_basis field).

A2. Selecting the REAL freight options (loose-filter trap)

GET /api/freight-quotes, then keep only rows where quote_id exactly equals the quote and that are the genuine lane set. Exclude distractors:

• Any id starting with FR-DIS- — these are benchmark/archived/wrong-size rows.
• Rows whose status is stale/mismatch AND that duplicate a mode you already have a clean active row for, or whose shipment_cbm/shipment_weight_kg differ from the consistent real shipment dimensions of the genuine set. The genuine set is exactly three rows — one mode each of air, sea, road — sharing one consistent shipment_cbm/shipment_weight_kg. Output mode uppercased (AIR/SEA/ROAD). Note: a real lane can still be status: "stale" / expired (e.g. a road quote that genuinely lapsed) — that one stays in the output and gets flagged stale; it is NOT a FR-DIS- distractor. Distinguish "real but expired" from "fake distractor."

A3. Per-option fields

• freight_cost_usd = cost_usd; transit_days = transit_days_text (copy the template's format — some want "4-6 days", some want "3-5"; mirror the example value style in the template).
• valid_until = freight valid_until.
• grand_total_usd = exw_total_usd + freight_cost_usd.
• Risk from route_risk: uppercase it for risk_level/customs_border_risk (low→LOW, medium→MEDIUM, high→HIGH). For risk_flag: low→NONE, medium→MEDIUM_BORDER_RISK, high→HIGH_BORDER_RISK.
• Validity vs the quote date: validity_status = VALID if valid_until >= quote_date, else STALE. source_is_stale is true if the freight status is stale or it is expired on the quote date, else false.

A4. Recommended mode (decision basis — order matters)

Pick the mode to recommend by filtering, then optimizing:

1. Eligibility filter (feasibility/safety): drop options that are STALE/expired on the quote date, and drop HIGH border-risk options. (MEDIUM risk is allowed but disfavored vs LOW.)
2. Among the eligible, prefer LOW route risk over MEDIUM.
3. Tie-break by lowest grand_total_usd (cheapest landed cost). This typically yields SEA when AIR and SEA are both LOW/valid (SEA is cheaper) and ROAD is medium/high or stale. Recommend by transport mode string ("SEA" etc). Keep risk/validity separate from recommendation: you still list every real option with its true flags even though only one is recommended.

A5. Policy / payment flags

• freight_reconfirmation_required = true always (policy POL-FREIGHT-RECONFIRM: freight rates reconfirm at final order).
• all_freight_options_valid_on_quote_date = true only if all three real options are VALID on the quote date (false if any is stale/expired).
• payment_terms from the customer + payment policies (see Payment terms table).
• customer_policy / segment label: map the customer segment to the controlled value the template shows (e.g. recurring NGO → RECURRING_NGO).
• road_quote_invalid_or_stale = true if the ROAD option is stale/expired.
• freight_warning: when something is stale or high-risk, write a short factual sentence: rates require reconfirmation at final order; name the stale freight ID, its expiry date, and its risk, and say it should not be used without a fresh quote. Keep it specific and grounded in the records.

B. Module-only indicative quote (RFQ, EXW only)

Inputs: an RFQ (RFQ-...) listing requested_modules, no confirmed destination. Output: one EXW-only quote with one line per requested module.

B1. SOP

1. GET /api/rfqs/<id> → customer_id, quote_date, requested_modules (each {product_code, quantity}), incoterm_requested.
2. For each requested module, GET /api/products/<code>:
   • article_number, unit_price from the product price tier at the requested quantity (module products usually have a single open tier), lead_time_days, shelf_life_months from that tier/product.
   • line_total = quantity * unit_price.
   • Emit one line item per module, in the RFQ's module order.
3. grand_total = sum(line_total).
4. freight_excluded = true, quote_basis = "EXW_ONLY" (policy POL-INDICATIVE-EXW
   • POL-EXW-SCOPE: no destination ⇒ EXW only, exclude freight).
5. offer_validity_days = 30 (policy POL-QUOTE-VALIDITY).
6. who_documentation_required = true for these emergency-health-kit/WHO module RFQs.
7. payment_terms: new NGO ⇒ PREPAY_100 (see Payment terms table).

B2. Critical exclusion — module lines ONLY

RFQs and products carry component_composition_distractors / components (e.g. "Paracetamol tabs", "gauze and splints", "RDT buffer packs"). These are medical-review context, not quote lines. Never split a module into component SKUs, never add component lines, never re-price by component. One line per requested module, period (policy POL-MODULE-GRANULARITY). Quantities come straight from the RFQ — do not infer extra units from composition tables.

C. Engagement / receivables reconciliation (opportunity → cash → revenue)

Inputs: an opportunity (OPP-...), its customer, a named contact, an as-of date, and (usually) a linked event + voucher. Output: account status, per-milestone state, revenue-recognition status, and routed follow-up tasks.

C1. Gather and link

1. GET /api/opportunities/<id> → stage, won_amount_usd, phases[] (phase_id, amount_usd, invoice_id, completion_date), contact, outstanding_amount_usd.
2. GET /api/invoices, GET /api/payments, GET /api/revenue-journals; filter each to this opportunity_id precisely (the collections contain many other accounts — match exactly). Link by invoice_id / phase_id.
3. Number milestones in phase order. Output milestone IDs in the convention the template uses — some templates label them MS1/MS2/MS3 (ascending), others use the API phase_id. Match the template's enum/example exactly.

C2. Per-milestone derivation

• invoice_total / amount = invoice (or phase) amount_usd.
• Payment state from the matched payment(s) and invoice paid/outstanding: fully paid ⇒ PAID; partially paid ⇒ PARTIAL; none ⇒ UNPAID. paid_amount = sum of posted payments (or invoice paid_amount_usd); amount_unpaid = invoice outstanding_amount_usd.
• invoice_state (when the template asks): mirror the invoice status mapped to the template enum (paid⇒PAID, unpaid/overdue⇒OPEN, etc.).
• Due-date nulling: for a PAID milestone set due_date to null (it is settled — no due date is actionable). For an unpaid milestone use the invoice due_date.
• Revenue recognition status (policy POL-REVREC — recognize paid+complete milestones; unpaid stay outstanding):
  • PAID and a matching revenue journal exists ⇒ RECOGNIZED.
  • PAID but no revenue journal ⇒ recognition is required but missing (REQUIRED_MISSING or MISSING_REVENUE_JOURNAL, per template enum).
  • UNPAID ⇒ recognition not yet required ⇒ NOT_REQUIRED_UNPAID.

C3. Account-level rollups

• opportunity_matches_milestones / opportunity_matches_phase_total = true iff won_amount == sum(phase amounts).
• outstanding_balance = sum of unpaid invoice outstanding amounts (= opportunity outstanding_amount_usd; verify they agree).
• total_paid_amount = sum of posted payments / paid invoice amounts.
• Contact: use the named contact; link it to this customer_id and opportunity_id.

C4. Revenue-recognition rollup block

• recognized_milestones = paid milestones that have a journal.
• missing_required_milestones = paid milestones lacking a journal.
• recognized_amount = sum of recognized milestone amounts.
• recognition_status: COMPLETE_FOR_PAID_MILESTONES if every paid milestone is recognized; MISSING_FOR_PAID_MILESTONES if any paid milestone lacks a journal; NOT_REQUIRED if nothing is paid yet. (Watch for the classic seeded case: a milestone that is paid but has no journal ⇒ flag missing, and the primary accounting action becomes "record revenue for that milestone".)

C5. Action routing (the decision layer)

Templates differ; produce the actions their enums demand. General logic:

Accounting / revenue action — driven by the paid-but-unrecognized milestone:

• If a paid milestone lacks a journal ⇒ action RECORD_REVENUE_MS<n>, on that milestone, amount = its value, debit_account = DEFERRED_REVENUE, credit_account = IMPLEMENTATION_SERVICES_REVENUE, owner_queue = ACCOUNTING.
• If all paid milestones are already recognized ⇒ VERIFY_REVENUE_ONLY/ NO_ACCOUNTING_ACTION per template (nothing to post).

Collection action — driven by the unpaid milestone vs the as-of date:

• Unpaid and not yet due (due_date > as_of_date) ⇒ MONITOR_UNPAID_NOT_DUE / COLLECT_UNPAID_MILESTONE task but not overdue; owner ACCOUNT_MANAGEMENT.
• Unpaid and due/overdue (due_date <= as_of_date) ⇒ SEND_COLLECTION_NOTICE; owner COLLECTIONS.
• No unpaid milestone ⇒ NO_COLLECTION_ACTION.
• The collection task carries the unpaid milestone ID, its amount, and its due date.

Event / voucher invitation action — from the linked event + voucher:

• GET /api/events and GET /api/vouchers; match the event by opportunity_id (or the ID named in the prompt) and the voucher by event_id/code.
• Map event status to the template enum (scheduled→SCHEDULED, confirmed/live→ACTIVE, completed→COMPLETED, etc.). Map voucher status (active→ACTIVE). discount_amount = voucher discount_percent value; max_uses = max_redemptions.
• If the event is upcoming/scheduled ⇒ invite action SEND_EVENT_INVITATION / SEND_BRIEFING_INVITE; owner ACCOUNT_MANAGEMENT (or EVENTS); attach event_id, voucher_code, contact, customer_id.
• Task titles: short, human, name the customer (e.g. "Milestone 2 collection - <Customer Name>", "Send celebration invite - <Customer Name>").
• Task due_date: collection tasks use the unpaid invoice due date; invitation tasks use a sensible pre-event date when the template expects one — prefer a date the records justify (e.g. shortly before the event); never invent far-future dates.

Customer name caveat: use the customer name from the API record. The prompt's friendly name may differ slightly (e.g. "Helios Health Alliance" in the prompt vs a different stored name) — follow the convention the template/answer expects, which is usually the human-readable account name the prompt uses for titles and the API name for the canonical customer_name field. When in doubt, match the API record.

Payment terms — controlled values

Derive from customer payment_profile / segment and the payment policies:

If grant terms restrict net terms, honor the restriction over the default.

Common misjudgments — do NOT do these

• Do not quote prior_unit_price_usd or prior_quote_quantity; reprice from the current catalog tier at the confirmed quantity.
• Do not include FR-DIS-* freight rows or wrong-shipment-size rows just because the loose filter returned them. There are exactly three real lanes.
• Do not confuse "real but expired/stale" lanes (keep + flag) with fake FR-DIS- distractors (drop entirely).
• Do not split modules into component SKUs or add component lines — module line level only; composition tables are medical context.
• Do not put a due_date on a PAID milestone — null it.
• Do not mark a paid milestone RECOGNIZED if no revenue journal exists — that is the "missing journal" case that drives the record-revenue action.
• Do not recommend a stale/expired or HIGH-risk lane even if it is cheapest; eligibility (validity + risk) gates the cheapest-cost tie-break.
• Do not compare validity/staleness against today's real date — use the prompt's stated quote/as-of date.
• Do not invent IDs, amounts, dates, or enum values not present in the API or template; every value must trace to a record or a stated convention.
• Do not emit markdown, code fences, or explanation — only the JSON object that matches the template's structure exactly.

Quick verification before returning

• Every template key present, correct nesting/order, enums are literal allowed values.
• All money at required decimal precision; grand_total = EXW + freight and line sums reconcile; outstanding = won - paid and matches the opportunity.
• Exactly three real freight lanes; recommended mode is eligible (valid + not high risk) and lowest landed cost among eligibles.
• Paid milestones with journals = RECOGNIZED; paid without = missing → action set; unpaid = NOT_REQUIRED_UNPAID with due date carried into the collection task.