template-apparel-erp: AI demand forecasting on production-order screen (stretch)

[willgriffin]

Parent

Epic: #1245

What to build

Stretch goal — completes the AI showcase by adding a fourth demonstrated capability beyond brand voice / chat / translation.

When creating a ProductionOrder for a Style, an inline agent suggests how to break the total production qty across colorways and sizes based on historical sales velocity. Surfaces as a "Suggest breakdown" button on the production-order form.

Behavior

• Input: (styleId, totalQty, lookbackDays) — defaults: 200 units, 90 days
• The agent calls a forecast.styleSizeMix(styleId, days) tool that queries past Orders + WholesaleOrders for that Style's SKUs, computes a colorway-and-size mix percentage with confidence intervals, and returns a recommended breakdown
• UI shows the recommendation as an editable grid the planner can override
• A second tool, forecast.seasonalAdjustment(styleId, targetMonth), applies seasonality from prior-year same-month data when the target production date is in a future season

Tier gate

commerce.aiForecasting feature: off for Starter and Growth, on for Enterprise only. Most of the demo runs without it.

Acceptance criteria

• Two tools registered (forecast.styleSizeMix, forecast.seasonalAdjustment)
• "Suggest breakdown" button on the production-order form
• Agent uses real seeded historical data and returns a plausible breakdown
• Tier gate hides the button when the feature is off
• Manual verification: with seeded historical data, click "Suggest" on a 200-unit production order, confirm the grid populates with reasonable percentages

Blocked by

• #1258 — seed data (forecasting needs historical orders to compute against)

Notes

This is explicitly stretch. If timing is tight, defer to the post-v1 backlog. The AI showcase is already strong with brand-voice + chat + translation.