pallet-alignment

name	pallet-alignment
description	Detects broken-layer (SD-PLT-001) and partial-pallet (SD-PLT-002) violations and proposes a round-down plan aligned to full layers / full pallets. Triggered on PALLET_CONFIG_VIOLATION events.
metadata	{"version":"1.0.0","author":"CPG Expert-Systems-Architect","required_tools":["mcp-sap-connector","mcp-logistics-calc"],"recipes":["PalletAlignmentRecipe.py"],"constrained_generation":["Guidance","Outlines"]}

Skill: Pallet Configuration Alignment

1. Overview

Pallet violations arise when ordered quantities don't align to full layers / full pallets, producing freight waste and labour overhead. The skill identifies broken-layer and partial-pallet conditions and proposes a round-down plan. Threshold fractions are policy-injected into PalletAlignmentRecipe.py.

2. Reasoning Loop

1. Confirm event_type is PALLET_CONFIG_VIOLATION.
2. Read metadata.pallet_lines — each entry carries sku, layer_qty, pallet_qty, ordered_qty.
3. Classify intent as PALLET_CONFIG.
4. Select PalletAlignmentRecipe.py.
5. Compliance Shadow always returns YELLOW when pallet data is present — operator sign-off is required because the round-down reduces ordered quantity.

3. Constrained Generation Policy

• intent must be PALLET_CONFIG
• recipe must be PalletAlignmentRecipe.py
• shadow verdict GREEN | YELLOW | RED
• recipe classification constrained to NO_VIOLATION | BROKEN_LAYER | PARTIAL_PALLET | MIXED_VIOLATION

4. Recipe-to-Intent Mapping

• PALLET_CONFIG → PalletAlignmentRecipe.py
• All other intents on this event type → FAIL_TO_HUMAN

5. Execution Protocol

Pallet-config violations never hard-block (no RED from policy hits alone). They always enter the YELLOW review queue so an operator can confirm the round-down and communicate reduced quantities to the buyer.

6. Output Requirements

• total_ordered_cases, loose_cases_total, order_line_count
• classification
• lines[] with complete_layers, loose_qty, full_pallets, pallet_fill_pct, violation_type
• suggested_plan[] with current, suggested, delta, reason
• recommended_action (ROUND_DOWN_TO_LAYERS | ACCEPT_AS_IS | REVIEW)