UCP Variant Data: The #1 Reason Agent Checkouts Fail

interface SelectionAxis {
  axis_id: string;
  display_name: string;
  allowed_values: string[];
}

interface ProductDefinition {
  product_uid: string;
  title: string;
  selectable_axes: SelectionAxis[];
}

interface AxisSelection {
  axis_ref: string;
  value_label: string;
}

interface VariantNode {
  variant_uid: string;
  selections: AxisSelection[];
  pricing: { base_amount: number; currency: string };
  inventory: { status: InventoryStatus; sku: string };
}

es, or the axis declaration must be pruned.

function validateVariantCardinality(product: ProductDefinition, variants: VariantNode[]): boolean {
  const expectedCombinations = product.selectable_axes.reduce(
    (acc, axis) => acc * axis.allowed_values.length, 1
  );
  return variants.length === expectedCombinations;
}

Rationale: Phantom axes cause agents to request combinations that do not exist. Lenient agents guess; strict agents abort. Synchronizing cardinality eliminates both failure modes.

Step 4: Attach Deterministic Availability Signals

Inventory state must be embedded directly in the variant payload. Agents use this to filter out-of-stock items before cart submission, preventing checkout-layer rejections.

type InventoryStatus = 'in_stock' | 'backorder' | 'preorder' | 'out_of_stock' | 'discontinued';

interface VariantNode {
  // ... previous fields
  inventory: {
    status: InventoryStatus;
    restock_date?: string;
    sku: string;
  };
}

Rationale: Checkout endpoints reject unavailable SKUs. By surfacing status at the variant level, agents can apply substitution logic or surface accurate availability messages to users, reducing retry cycles and cart_created stalls.

Pitfall Guide

1. Opaque Identifiers Without Axis Mapping

Explanation: Variants are exposed as bare IDs (var_8821) with no option metadata. Agents cannot map user intent to a specific SKU and default to the first entry or random selection. Fix: Every variant must include a selections array that explicitly declares its position across all product axes. Never rely on implicit ordering.

2. Compound Attribute Strings

Explanation: Multiple dimensions are merged into a single label (e.g., "Medium / Regular Fit"). Agents split the string inconsistently across models, causing divergent variant picks for identical prompts. Fix: Decompose compound strings into separate axes. "Medium" belongs to Size, "Regular" belongs to Fit. Each axis gets its own AxisSelection entry.

3. Lexical Inconsistency Across Siblings

Explanation: Variants on the same product use different labels for the same value ("M", "Medium", "med", "Medium "). Agents perform exact string matching; lexical drift breaks resolution. Fix: Enforce a canonical label registry at the product level. Validate all variant payloads against this registry during ingestion. Strip trailing whitespace and normalize case.

4. Silent Stock Omissions

Explanation: Variants lack availability flags. Agents add sold-out items to the cart, triggering checkout rejection and silent session termination. Fix: Always populate the inventory.status field. Use standardized enum values. If stock data is unavailable, default to out_of_stock rather than omitting the field.

5. Phantom Axis Declarations

Explanation: selectable_axes declares dimensions that no variant actually implements. Agents attempt to satisfy selections that have no corresponding SKU. Fix: Run a cardinality check before publishing. If an axis has no variant coverage, remove it from selectable_axes. Do not leave dangling dimensions.

6. Locale-Dependent Label Drift

Explanation: Variant labels are translated dynamically without preserving a stable canonical key. Agents caching responses in one locale fail when switching regions. Fix: Separate value_label (display) from value_key (stable identifier). Agents should match against value_key, while UI layers render value_label based on locale.

7. Stale Variant Caching

Explanation: Product catalogs cache variant payloads for performance, but inventory or pricing updates are not propagated. Agents operate on outdated data, causing cart rejections. Fix: Implement cache invalidation hooks tied to inventory management systems. Use ETags or versioned variant payloads (variant_version: 14) to force agent refreshes on state changes.

Production Bundle

Action Checklist

• Audit existing variant payloads for missing selections arrays and replace opaque IDs with explicit axis mappings
• Decompose all compound attribute strings into discrete axes with dedicated AxisSelection entries
• Establish a canonical label registry and enforce lexical consistency across all product variants
• Embed inventory.status in every variant payload using standardized enum values
• Run cardinality validation to ensure selectable_axes combinations match actual variant counts
• Separate display labels from stable value keys to prevent locale-induced resolution failures
• Implement cache versioning or ETag headers to force agent refreshes on inventory/pricing updates
• Add a CI/CD validation step that rejects variant payloads failing axis alignment or cardinality checks

Decision Matrix

Scenario	Recommended Approach	Why	Cost Impact
Simple product with single variant	Omit selectable_axes; use metadata.attributes for descriptive properties	Prevents agents from rendering unnecessary pickers; reduces payload size	Low (schema simplification)
Multi-variant product with stable inventory	Canonical selections array + explicit inventory.status	Guarantees deterministic agent matching; eliminates checkout rejections	Medium (data modeling effort)
Dynamic inventory with frequent stockouts	Variant-level status field + cache versioning	Enables agents to filter unavailable SKUs before cart submission; reduces retry latency	Medium (inventory sync integration)
Global marketplace with multiple locales	Stable value_key + locale-aware value_label	Prevents resolution failures across regions; maintains agent consistency	High (localization pipeline)
Legacy catalog with compound strings	Automated decomposition script + canonical registry	Fixes agent divergence without rewriting frontend UI; aligns with spec requirements	Medium (data migration)

Configuration Template

// product-catalog.types.ts
export interface AxisDeclaration {
  axis_id: string;
  display_name: string;
  canonical_values: string[];
}

export interface VariantSelection {
  axis_ref: string;
  value_key: string;
  display_label: string;
}

export type StockStatus = 'in_stock' | 'backorder' | 'preorder' | 'out_of_stock' | 'discontinued';

export interface VariantPayload {
  variant_uid: string;
  selections: VariantSelection[];
  pricing: {
    base_amount: number;
    currency: string;
    discount_applied?: number;
  };
  inventory: {
    status: StockStatus;
    sku: string;
    restock_estimate?: string;
  };
  metadata?: Record<string, string | number | boolean>;
}

export interface ProductCatalogEntry {
  product_uid: string;
  title: string;
  description: string;
  selectable_axes: AxisDeclaration[];
  variants: VariantPayload[];
  schema_version: '2026-04-08';
}

Quick Start Guide

1. Extract current variant payloads from your catalog API and run a lexical consistency check against a canonical label registry. Flag any compound strings or mismatched labels.
2. Refactor variant objects to include a selections array. Map each variant to its corresponding axes using explicit axis_ref and value_key pairs. Remove implicit ordering.
3. Attach inventory status to every variant. Use standardized StockStatus enums. Ensure out-of-stock items are explicitly marked rather than omitted.
4. Validate cardinality by comparing selectable_axes combinations against actual variant counts. Prune unused axes or generate missing variants to maintain synchronization.
5. Deploy with cache versioning. Add a variant_version field or ETag header to force agent refreshes. Monitor cart_created session rates; a drop below 5% indicates successful alignment.