eCommerce: Load Context

This is the canonical extract of Steps 1 + 3 from today's recommend-ecommerce-strategy.md, scoped to the fields eCommerce categories actually need.

Step 1 — Resolve siteId

If a siteId is not already known, call ListWixSites:

• If the merchant referenced a site by name, match it.
• If exactly one site exists, auto-select it.
• Otherwise, ask the merchant which site to use.

Do not proceed without a siteId.

Step 2 — Load business profile (the canonical endpoint)

Same call today's recommend-ecommerce-strategy.md Step 3 makes. The field list below is the eCommerce subset — orchestrator-needed metrics (visitors / orders / GPV → AOV) plus locale/currency/industry used for dispatch tags and recommendations.

Extract each field into conversation context as siteData:

Missing fields ⇒ no data for that field (don't fabricate). If merchant_business_country, suggested_main_industry, or online_gpv_last_30_days are missing or null, surface to the merchant: "Cannot resolve required site data: ." and stop.

Derived value: siteData.aov = parseInt(gpv30d) / parseInt(orders30d) — in siteData.currency units.

Step 2b — Detect catalog presence and load analytics

Call GetCatalogAnalytics once here so all downstream pricing/promotions flows can reference siteData.catalogAnalytics without re-fetching.

Save the full categoryGroups array as siteData.catalogAnalytics. From the "All Products" group extract:

• siteData.catalogProductCount = count() value (0 if missing or call fails)
• siteData.hasCatalog = siteData.catalogProductCount > 0

If the call fails, set siteData.hasCatalog = true (assume catalog exists; let downstream fail naturally).

This data is used by all pricing & promotions discount flows. Flows that require a product catalog (Bundle & Save, Upsell Boost, Stock Mover, Seasonal) MUST check siteData.hasCatalog before proceeding — if false, stop with: "This site has no products. Set up your product catalog first before running promotions."

Step 3 — Derive region (used by dispatch context tags)

From siteData.country, set siteData.region:

• BR | AR | MX | CL | CO | PE → LATAM
• EU member states (DE | FR | IT | ES | NL | BE | PL | SE | IE | AT | PT | FI | DK | CZ | HU | GR | RO | BG | HR | SK | SI | LT | LV | EE | CY | MT | LU) → EU
• JP | CN | KR | SG | AU | NZ | IN | TH | ID | MY | PH | VN | HK | TW → APAC
• otherwise → null

(UK is not in region:EU. AU/NZ are in region:APAC but Tax inclusive-pricing rules differ — handled inside Tax promotions.)

Step 4 — Return

Return immediately. Subsequent eCommerce category dispatches read siteData.* from conversation context without re-fetching. If the agent crosses into a different L1 (e.g. Stores) within the same session, that L1's loader will see siteData.country already loaded, skip its Steps 1-3, and only fire its own general derivations.

Architectural rule — general site data only

This file must contain only general / cross-cutting site data — fields that every category in this L1 needs. It must not include L3-category-specific runtime detection (e.g. Tax calculator, Catalog V1/V3, payment-provider state, shipping-coverage state).

Reasoning:

• Per-category detect calls accumulate cost on every session entry, even when the agent never visits that category.
• L3-specific state can use APIs that aren't TPA-public (e.g. the Wix Tax FQDNs are exposure: INTERNAL); if the L1 loader depends on them, the entire L1 fails to load. Keeping the loader general-only contains the blast radius.
• Per-category state can change independently of the merchant; loading it eagerly invites stale-data bugs.

Where category-specific runtime data is detected instead: inside the category's default.md (Step before dispatch) or inside the specific promotion that needs it. The L1 loader does not prime per-category fields; the category does its own detection lazily, only when its own intent dispatch fires.

Concretely for the categories we know about:

• Tax — calculator detection (Manual vs Avalara) belongs in ecom-tax.md (the merged category-doc + dispatcher) or in each Tax promotion when it runs. Not here.
• Catalog V1/V3 version detection — belongs in the Stores category loader when Stores migrates. getCatalogAnalytics (product count/margin/price) is already loaded here as a cross-cutting concern; V1/V3 detection is a separate, Stores-specific signal.
• Payments & finance (when Get-paid migrates) — payment-provider state belongs in finance-and-payments/get-paid-finance-default.md.

What this file does NOT do

• Does not detect per-category runtime state (see architectural rule above).
• Does not load tracking history. But: if any recipe in this session generates a recommendation to present to the merchant — regardless of which eCommerce category — it MUST load API: Recommendation Tracking and follow the lifecycle: load history before generating (Query), persist as PROPOSED via BatchCreate before presenting, and track execution via MarkExecuting / MarkDone / MarkFailed. This obligation applies to ALL eCommerce domains.
• Does not load non-eCommerce L1 fields (catalog V1/V3 version detection, payment provider, …). Those belong inside the owning L1's category-level files (per the architectural rule). Note: getCatalogAnalytics (product count, margin, price distribution) IS loaded here as a cross-cutting concern; Catalog V1/V3 version detection is separate and belongs in the Stores category's own loader.
• Does not enforce dispatch. The category's default.md is what scores tags and picks a promotion — this file only fills general site context.

Pattern for future L1 loaders

When a new L1 domain is migrated to the routing tree, author a <l1>-load-context.md at its root:

• references/stores/stores-load-context.md — Stores L1 loader. Steps 1-3 same boilerplate, Step 4 runtime-detects Catalog version (V1/V3) and any other Stores-specific fields.
• references/get-paid/get-paid-load-context.md — Get-paid L1 loader. Step 4 inspects payment provider state, enabled methods.
• references/contacts/contacts-load-context.md — Contacts L1 loader. Possibly minimal — most CRM ops don't need runtime detection.

Each lives sibling to its L1's category-docs. Each is referenced from every category's default.md within its L1. Skip-if-loaded check across L1s makes cross-L1 sessions cost-free.