fix: dry-run patches — guard enrollment.updateMany on replay + contextual TT CTA after validated replay

feat: Time-Travel complet via Redis — effectiveDay override, coupe-circuits SUITE/INSCRIPTION, skip COMPLETED en mode replay

feat: 4 UX/backend fixes — anti-ghost image, time-travel replay, UX message order, contextual AI remediation

fix: imageUrl || audioUrl fallback in image flow (R2 upload #2 resilience)

fix: re-validate COMPLETED→PENDING for text exercise responses + guard businessProfile.upsert at Day 10+

fix: pass correct imageUrl (R2 stored URL) to vision pipeline instead of empty audioUrl

fix: restore image forwarding and result propagation in internal routes

fix: comprehensive vision hardening (loosen wordcount bypass + 60s timeout)

fix: resolve Module 8 image response failure (Vision GPT-4o + Caption fix)

chore: trigger fresh build to bake in environment variables

fix: remove hardcoded WhatsApp fallback and use environment variables

fix: restore missing loops and fix syntax in whatsapp route

fix: update production WhatsApp number in web app

fix(scripts): move reset script to apps/api to fix IDE error

fix: restore whatsapp routes and add diagnostic logs

feat: upgrade Vision to GPT-4o with J11 bypass and enhanced logging

fix: restore image interception and vision pipeline with [IMAGE-FLOW] logs

feat: Global Vision Activation and Day 8 Visual Proof Analysis

fix: Anti-False Validation for Card Interactions

fix: Correct Gemini model name and increase OpenAI timeout

fix: AI Provider Resilience and Worker Lock Recovery

feat: Implement Non-Blocking Card UX and Prompt Hook Hook

fix: Unlock Day 7 Button Handlers and Optimize AI Latency

fix: Implement 5-minute re-validation window and Flow Sync Logging

fix: Harden Pipeline with Day 11 Guards and Null Safety

fix: Resolve Production Crash & Add UX Guidance for Completed Exercises

chore: Add observability log for feedback enqueuing

fix: Resolve Production Crash (Gemini 404, OpenAI Zod, Job Payload mismatch)

feat: Async-First Webhook Architecture & Idempotency Hardening

fix(progression): atomic logic inversion and command unlocking

fix(worker): reorder db updates before whatsapp and add safety checks

fix(whatsapp): idempotency lock and progression blockers

feat(ai): team building visual experience and persistence

chore(qa): final QA adjustments for multimodal and search

feat(ai): implement multimodal vision and dynamic data for business context

feat(ai): coaching refinements - temperature support, memory filter, and flexible guardrails for Day 6/7

feat: stabilize AI engine, harden Zod schemas, and implement UX guardrails

feat: Implement fuzzy matching for commands and input guardrails in whatsapp.ts

fix: Correct audio handler brace structure - FR users now get confirmation and feedback after voice message

fix: Update Gemini model names and harden transcription body reader

fix: Decouple A/V providers in AIService and refactor routes for robustness

feat: Implement Multi-AI Architecture with Google Gemini provider and failover logic

fix(whatsapp): Sprint 42 - Sécurisation routage Deep Dive et immuabilité secteur

fix(ai): Sprint 41 - Détanchéité linguistique Wolof stricte et réinjection boucle Deep Dive

feat(Sprint 40): Boucle interactive Deep Dive & persistences des réponses

feat(ai): refactor coach feedback to 3-pillar format with mandatory browsing (Sprint 39)

chore: execute Sprint 38 technical debt resolution (Type Safety, Zod validation, Vitest, Mock LLM extracted)

feat: Genspark-Standard upgrade, MLOps audit fixes, and XAMLÉ branding

feat(ai): add bank-level pitch deck and business plan generator script

test(e2e): add full journey simulation script that generates final AI pitch deck