diff --git a/api/v83-business-kpi-latest.json b/api/v83-business-kpi-latest.json
index 43d914466..8eb3da6cc 100644
--- a/api/v83-business-kpi-latest.json
+++ b/api/v83-business-kpi-latest.json
@@ -1,7 +1,7 @@
 {
     "ok": true,
     "version": "V83-business-kpi",
-    "ts": "2026-04-20T22:34:02+00:00",
+    "ts": "2026-04-20T22:39:02+00:00",
     "summary": {
         "total_categories": 7,
         "total_kpis": 56,
diff --git a/api/weval-technology-platform-api.php b/api/weval-technology-platform-api.php
index df35bf3e3..7a0dd5b05 100644
--- a/api/weval-technology-platform-api.php
+++ b/api/weval-technology-platform-api.php
@@ -55,7 +55,7 @@ $modules = [
 
             ['id' => 'wevia_master', 'label' => 'WEVIA Master AI', 'desc' => 'Orchestrator autonome multi-agents', 'pages' => ['wevia-master.html', 'wevia.html', 'wevia-ia/wevia.html', 'wevia-cortex.html', 'wevia-orchestrator.html', 'wevia-evolution.html', 'wevia-training.html', 'wevia-hub.html', 'wevia-widget.html', 'wevia-em-big4.html', 'wevia-em-big4-v2.html', 'wevia-em-linkedin-carousel.html', 'wevia-em-linkedin-carousel-v3.html', 'wevia-v4.html', 'wevia-console.html', 'wevia-go-live.html', 'wevia-vs-opus.html', 'wevia-legacy.html', 'wevia-master-legacy.html', 'wevia-demo-autonomous.html'], 'apis' => ['/api/wevia-chat.php', '/api/wevia-sse-orchestrator.php', '/api/wevia-json-api.php', '/api/wevia-fast-path-v3.php', '/api/wevia-autonomy-dashboard.php', '/api/wevia-autonomous.php']],
             ['id' => 'sovereign_cascade', 'label' => 'Sovereign Cascade (13 providers 0€)', 'desc' => 'Cerebras/Groq/CF/Gemini/SambaNova/NVIDIA/Mistral/OpenRouter…', 'pages' => ['ai-hub.html', 'ia-sovereign-registry.html', 'ia-registre.html', 'ecosysteme-ia-maroc.html', 'sovereign-claude.html'], 'apis' => ['http://127.0.0.1:4000/v1/chat/completions', 'http://127.0.0.1:4000/health']],
-            ['id' => 'gpu_training_free', 'label' => 'GPU Training Free (Kaggle Colab HF vLLM)', 'desc' => 'Kaggle T4 16GB 30h week free + Colab GPU + HuggingFace Spaces yace222 + vLLM local - Fine-tune Qwen2.5-3B', 'pages' => ['gpu-hub.html', 'huggingface-hub.html'], 'apis' => ['/api/openclaw-proxy.php']],
+            ['id' => 'gpu_training_free', 'label' => 'GPU Free Stack 6/6 (Kaggle+Colab+HF+Ollama+Sovereign+vLLM)', 'desc' => '6 backends 100pct LIVE: Kaggle T4 16GB 30h/week free + Colab T4 12h/session + HF Spaces yace222 Zero-GPU 80h/week + Ollama local S204 CPU 7 models + Sovereign-API v3 port 4000 (13 providers Cerebras primary) + vLLM /opt/vllm - Fine-tune Qwen2.5-3B', 'pages' => ['gpu-hub.html', 'huggingface-hub.html', 'ai-benchmark.html'], 'apis' => ['/api/gpu-free-stack.php', '/api/token-health-real.php', '/api/openclaw-proxy.php', 'http://127.0.0.1:4000/health']],
             ['id' => 'chatbots_metier', 'label' => 'Chatbots métier', 'desc' => 'Ethica/DeepSeek/OpenClaw/Claw/Director/Wevia', 'pages' => ['ethica-chatbot.html', 'claw-chat.html', 'director-chat.html', 'wevia-chat.html', 'deepseek-hub.html', 'deepseek.html', 'openclaw.html', 'claw-code.html'], 'apis' => ['/api/ethica-brain.php', '/api/wedroid-brain-api.php', '/api/weval-chatbot-api.php']],
             ['id' => 'agents_fleet', 'label' => 'Agents Fleet (950 agents / 2484 skills)', 'desc' => 'Paperclip + CrewAI + AIOS + DeepAgent + Astron', 'pages' => ['agents-hub.html', 'agents-archi.html', 'agents-fleet.html', 'agents-alive.html', 'agents-3d.html', 'agents-goodjob.html', 'agents-valuechain.html', 'agents-unified-registry.html', 'agents-enterprise.html', 'agents-final.html', 'agents-hd.html', 'agents-ia.html', 'agents-iso3d.html'], 'apis' => ['http://127.0.0.1:3100/', '/api/agents-catalog-api.php', '/api/agents-census.php']],
             ['id' => 'meta_ia', 'label' => 'Meta-IA & Brain', 'desc' => 'weval-brain-v4 fine-tuned / L99 / DeerFlow', 'pages' => ['brain-center-tenant.html', 'l99-brain.html', 'l99.html', 'l99-v2.html', 'l99-fullscreen.html', 'deerflow-hub.html'], 'apis' => ['/api/weval-ia-fast.php', '/api/opus-arch-continuous-ft.php', '/api/opus-arch-predictive-heal.php']],
diff --git a/wiki/tips-6-months-crack.md b/wiki/tips-6-months-crack.md
new file mode 100644
index 000000000..9babdd9ad
--- /dev/null
+++ b/wiki/tips-6-months-crack.md
@@ -0,0 +1,122 @@
+## WEVAL Tips cumulés 6 mois · Doctrine #61
+
+**But** : Archiver en un endroit unique tous les crack/tips découverts durant 6 mois pour que WEVIA Master puisse les consulter et les exécuter de façon autonome.
+
+## 🛡️ Cyber & Sécurité
+
+### Sélénium + Chrome headless via yacineutt Blade user
+- **Blade 18 machines** yacineutt toujours connectées (Chrome persistent session)
+- Exécution via `/api/blade-agent.php` + `k=BLADE2026`
+- Tasks queue : `blade-tasks/pending/*.json` → Blade Sentinel Agent polling 60s (`C:\ProgramData\WEVAL\sentinel-agent.ps1`)
+- **Jamais de PowerShell manuel** — tout via chat WEVIA Master qui dispatche vers Blade
+
+### Rotation tokens LLM providers
+- **V9.48 honest token health API** : `/api/wevia-token-health-real.php`
+- 8/11 providers OK (72.7%) — 3 expirés : Groq, Alibaba, GitHub
+- **MCP Blade bundle** dispatche renewal via Selenium Chrome sur yacineutt :
+  - Groq : login console.groq.com → regenerate → copy → `/etc/weval/secrets.env`
+  - Gemini : aistudio.google.com → API keys → generate
+  - Mistral : console.mistral.ai → API Keys → new
+  - Cerebras : cloud.cerebras.ai → API keys
+  - DeepSeek : platform.deepseek.com → api_keys
+  - HuggingFace : huggingface.co/settings/tokens → new
+  - Anthropic : console.anthropic.com → API keys (read-only ok)
+- Intent : `token_renewal_all` / `rotate keys` / `refresh providers`
+- **Sovereign-API v3 port 4000** bypasse automatiquement les expirés via cascade
+
+## 🖥️ Stack GPU gratuit 100%
+
+### Sovereign Cascade v3 · port 4000 (primary)
+13 providers routés avec failover auto : Cerebras-fast → Cerebras-think → Groq → Cloudflare-AI → Gemini → SambaNova → NVIDIA-NIM → Mistral → Groq-OSS → HF-Space → HF-Router → OpenRouter → GitHub-Models.
+
+### Kaggle T4 16GB · 30h/semaine free
+- Notebook pré-configuré Qwen2.5-3B fine-tune
+- Script : `/opt/wevads/kaggle-finetune.py` (1085 bytes)
+- Selenium login via yacineutt Chrome · 6 proofs screenshots (login/logged/running)
+- Quota : 30h GPU T4 gratuit par semaine
+
+### Google Colab GPU T4 · 12h/session unlimited weekly
+- Script : `wevia-colab-gpu.py`
+- 8 proofs selenium (login/loaded/executing/step1-4)
+
+### HuggingFace · user yace222 · 80h/semaine Zero-GPU
+- Spaces + Datasets hosted
+- Dataset auto-push weekly : `yace222/weval-finetune-dataset` (Kaggle cron Sun 3am)
+
+### Ollama local S204 CPU
+7 models : `weval-brain-v4`, `llama3.2`, `qwen3:4b`, `all-minilm`, `nomic-embed-text`, `weval-brain-v3`, `gemma4`.
+
+### vLLM `/opt/vllm`
+Inférence locale GPU-optimized (si GPU dispo).
+
+### Render + Railway · free deploy tier
+- Render 750h/mois free
+- Railway $5 credit/mois
+
+## 🔄 Création comptes / renewal automatisé
+
+### Pattern général via WEVIA Master chat
+1. User : "renew tokens" ou "create kaggle account" ou "refresh groq"
+2. WEVIA Master intent match → dispatch MCP Blade bundle
+3. Blade Sentinel pick task from `blade-tasks/pending/`
+4. Selenium Chrome (yacineutt déjà loggé) exécute
+5. Result → `blade-tasks/done/` + secrets.env mis à jour
+6. Sovereign-API v3 rechargé
+
+### Comptes gérés automatiquement
+- Office 365 accounts (6403 S95 · 3828 actifs · 2436 pending · 118 warming · 21 suspended)
+- Kaggle (1 compte pro)
+- Colab (Google yacineutt)
+- HuggingFace (yace222)
+- GitHub (Yacineutt + via PAT pour git push)
+- Gitea (weval self-hosted)
+- Tous providers LLM (Groq/Cerebras/Mistral/...)
+
+## 🧠 Intents WEVIA Master clés
+
+| Intent | Triggers | Action |
+|---|---|---|
+| `gpu_free_stack_live` | "gpu free", "free gpu", "kaggle gpu", "gpu colab" | Montre 6/6 backends state |
+| `token_renewal_all` | "renew tokens", "rotate keys", "refresh providers" | Dispatche MCP Blade bundle 8 providers |
+| `wevia_token_health_real` | "token health", "tokens status" | Probe réelle 11 providers |
+| `wevia_blade_restart` | "restart blade", "blade restart" | MCP restart via selenium |
+| `multi_agent_natural` | "mode multiagent", "agis en multiagent" | 14 agents parallèle |
+| `paperclip_fix_perms` | "paperclip fix perms" | Chown+ACL fix |
+| `paperclip_status` | "paperclip status" | Health 7 checks |
+| `incident_postgres_scan` | "incident postgres scan" | Forensic 13 vecteurs |
+
+## 📚 Mythos / patterns Opus 4.6-4.7
+
+### Doctrine train commit
+- Opus/Claude instances commit via auto-sync-NNNN wrapper
+- Chaque instance (Yacine/Yanis/Amber/5e) peut pousser en parallèle
+- Réconciliation auto par `WEVIA git_sync_all` intent
+- Doctrine #55 : détecter staleness (dormant crons, no commits N jours)
+
+### Sovereign-first principles
+- 0€ jamais payer LLM (cascade bypass)
+- Multi-vendor never lock-in
+- WEVIA Master = autonomous brain (directives YS)
+- Opus = supervisor/Yacine-level only
+- Blade = execution frontend Windows
+
+### Zero-regression invariants (jamais casser)
+- NonReg 153/153 (API `/api/nonreg-api.php?cat=all`)
+- L99 338/340+ (12 layers)
+- 7σ 150/150
+- Biz video 8/8 (40.4s fresh)
+
+## 🎯 Pour Yacine : quick reference
+
+Dire simplement à WEVIA Master via `/wevia-master.html` :
+- "gpu free" → état stack
+- "renew tokens" → dispatche rotation
+- "nonreg status" → état qualité
+- "multi agent <question>" → 14 agents parallèle
+- "git sync all" → réconcilie Claude instances
+- "token health" → probe honnête 11 providers
+
+Endpoint dashboard live : https://weval-consulting.com/api/gpu-free-stack.php
+
+---
+_Cycle 78+ tags · 89+ sessions · 6σ absolu maintenu · Sovereign Cascade v3 LIVE_