v2.1.0-unstable.1

Dev Release

• Environment: dev
• Source branch: main
• Commit: 1a71907
• Successful deploy run: https://github.com/Azure-Samples/holiday-peak-hub/actions/runs/23122486292
• Deploy mode: UI-only (token path)

Known Follow-up

• Full infra deploy run (23121723677) failed in provision due Azure name-conflict and should be remediated separately.

v2.1.0-unstable.0

Unstable pre-release for upcoming 2.1.0 workstream.\n\n- Status: pre-release (not stable)\n- Base: main@60e858b\n- Purpose: validation and integration testing before stable cut

v2.0.0

Breaking release\nThis release removes backward compatibility in product-management-consistency-validation and enforces a single operation model for completeness workflows.\n\n### What changed\n- Removed legacy validator and product-events compatibility path\n- Removed legacy MCP tools: /product/consistency/check, /product/consistency/product\n- Canonical operation path is now /product/completeness/evaluate and completeness-jobs event processing\n- Updated docs and changelog for migration\n\n### Migration\n- Update callers to use /product/completeness/evaluate\n- Route async completeness jobs to completeness-jobs\n

v1.1.0 — Enterprise Connectors, Product Truth Layer, HITL Review System

Highlights

This release introduces enterprise system connectors, the Product Truth Layer foundation, Human-in-the-Loop (HITL) review capabilities, and enterprise hardening patterns for production workloads.

Enterprise Connectors

Production-ready integrations for major retail enterprise systems:

• Oracle Fusion Cloud SCM — Inventory, Purchase Orders, Shipments (OAuth 2.0 + JWKS)
• Salesforce CRM & Marketing Cloud — Contacts, Accounts, Leads, Campaigns (OAuth 2.0)
• SAP S/4HANA — Material Master, Inventory Positions, Purchase Orders (OData v4)
• Dynamics 365 Customer Engagement — Contact, Account, Opportunity, Case (Dataverse API)
• Generic REST DAM — Configurable asset management connector

Product Truth Layer (Phase 1-2)

Foundation for unified product data management:

• Pydantic v2 data models (\TruthAttribute, \ProposedAttribute, \GapReport, \AuditEvent)
• Truth Ingestion service with Cosmos DB integration
• Sample data, category schemas, and seeding scripts

HITL Staff Review UI

Human-in-the-loop interface for AI-proposed attribute validation:

• Review queue page with filtering and pagination
• Entity detail review with side-by-side comparison
• UI components: \ReviewQueueTable, \ProposalCard, \ConfidenceBadge, \CompletenessBar, \AuditTimeline\

Enterprise Hardening

Resilience patterns for production deployments:

• Circuit Breaker — Failure isolation with half-open recovery
• Bulkhead Pattern — Semaphore-based resource isolation
• Rate Limiter — Token bucket algorithm with async support
• Telemetry — OpenTelemetry spans and metrics integration
• Health Probes — Enhanced Kubernetes liveness/readiness

PIM Writeback Module

Bi-directional sync with Product Information Management systems:

• Opt-in tenant configuration
• Circuit breaker protection
• Conflict detection with version comparison
• Full audit trail

Frontend Enhancements

• Stripe checkout integration (real payments)
• Dashboard and profile with live API hooks
• Server-side route protection (Next.js middleware)
• Admin UI for Truth Layer management

Testing

• 508 tests passing (↑343 from v1.0.0)
• Connector test suites: Oracle (48), Salesforce (48)
• Enterprise hardening: 45 tests
• PIM writeback: 25 tests

Breaking Changes

None — this is a backward-compatible feature release.

Upgrade Notes

No migration required. New features are opt-in via configuration.

Full Changelog

See CHANGELOG.md for complete details.

---

Contributors: @Copilot, @github-actions[bot]

v1.0.0 — First Stable Release

Holiday Peak Hub v1.0.0 — First Stable Release

Agent-driven retail accelerator with cloud-native architecture, complete frontend, and comprehensive backend.

Framework (lib/) — First Stable Version

The holiday_peak_lib micro-framework is now production-ready:

Agents

• BaseRetailAgent: Abstract base for domain agents with SLM/LLM routing
• AgentBuilder: Composable builder pattern for agent configuration (memory, tools, model targets)
• ServiceAgent: Production wrapper with health, lifespan, and MCP surface
• FastAPIMCPServer: Expose agent capabilities as MCP tools for agent-to-agent communication
• Foundry integration: Azure AI Foundry V2 SDK (�zure-ai-projects>=2.0.0b4) with PromptAgentDefinition, auto-ensure on startup, strict enforcement mode

Memory (Three-Tier Architecture)

• HotMemory — Redis with connection pooling for real-time context
• WarmMemory — Cosmos DB for medium-term state (with SDK diagnostics and retry-after handling)
• ColdMemory — Azure Blob Storage for long-term archival
• MemoryBuilder — Cascading read/write across tiers

Adapters / Connectors

• BaseAdapter with canonical connector interfaces
• Domain adapters: CRM, Funnel, Inventory, Logistics, Pricing, Product
• Mock adapters for local development and testing

App Factory

• build_service_app() — Wires FastAPI with standard middleware, logging, health endpoints, settings, agent registration

Schemas (Pydantic v2)

• Core, CRM, Funnel, Inventory, Logistics, Pricing, Product schemas
• Validated request/response models across all agent domains

Utilities

• EventHubPublisher — Async event publishing with error handling
• Structured logging with correlation IDs and Application Insights
• Retry helpers with exponential backoff

Configuration

• ServiceSettings / MemorySettings / PostgresSettings — Pydantic Settings with env var loading
• FoundryAgentConfig — SLM-first routing with LLM upgrade paths

---

Agent Services (21)

• E-commerce (5): catalog-search, product-detail-enrichment, cart-intelligence, checkout-support, order-status
• Product Management (4): normalization-classification, acp-transformation, consistency-validation, assortment-optimization
• CRM (4): profile-aggregation, segmentation-personalization, campaign-intelligence, support-assistance
• Inventory (4): health-check, jit-replenishment, reservation-validation, alerts-triggers
• Logistics (4): eta-computation, carrier-selection, returns-support, route-issue-detection

CRUD Service

• 31 REST endpoints across 15 route modules (FastAPI + PostgreSQL/asyncpg)
• JWKS-based JWT auth (Microsoft Entra ID), RBAC with 4 role tiers
• APIM-routed agent client with circuit breaker and tenacity retry
• Event publishing to 5 Azure Event Hubs topics

Frontend

• Next.js 16.2.0-canary, React 19, TypeScript 5.7
• 13 pages, 52 components (atomic design), MSAL auth
• 6 TypeScript API services, 5 React Query hooks

Infrastructure

• Bicep modules (14 AVM-based) for AKS, ACR, APIM, Redis, Event Hubs, Key Vault, Cosmos DB, PostgreSQL, SWA
• azd-first deployment with 8-job CI/CD pipeline
• Helm chart scaffolding with KEDA autoscaling

Testing

• 165 tests passing (lib + agent + CRUD)
• Coverage: 73% (lib core)
• CI gates: lint (black, isort, pylint), format, test with coverage floor

Documentation

• 21 Architecture Decision Records (ADRs)
• Full components catalog, playbooks, test plans
• CRUD service implementation guide, frontend integration guide