backup: snapshot post-démo GHT 2026-05-19

Backup état complet après enregistrement vidéo démo de bout en bout.
À utiliser comme point de référence pour la consolidation post-démo.

Changements majeurs de la session 18-19 mai :
- AIVA-URGENCE : page autonome avec preset URL + auto-focus chain
- Workflow Demo_urgence_3_db : merge linux_db + steps AIVA + pause humaine NoMachine
- Bypass LLM (static_result / static_text) dans replay_engine
  pour démos déterministes sans appel Ollama
- Fix api_stream:3013 — replay_paused au premier polling /next
- dag_execute : lift duration_ms vers top-level pour wait runtime
- NPM bypass auth /aiva-urgence/ via location ^~ (proxy_host/10.conf hors git)
- scripts/cancel-replays.sh — workaround Stop VWB qui ne purge pas la queue

Anchors visuels (468) forcés dans le commit pour garantir restorabilité.
DB workflows actuelle + ~12 .bak DB de la journée incluses.

Sujets identifiés pour consolidation post-démo (TODO) :
1. Bug VWB recapture anchor ne régénère pas le PNG
2. Léa client accumule état mémoire (restart périodique requis)
3. Stop VWB ne purge pas la queue serveur (lien manquant vers /replay/cancel)
4. Bug coord client mss tronqué 2560x60 → mapping Y cassé
5. delay_before/delay_after ignorés au runtime (fix partiel duration_ms)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tools/probe_qwen3vl_processor.py

@@ -0,0 +1,80 @@
+"""
+Dump runtime des attributs et comportement effectif du processor
+Qwen3-VL-8B-Instruct. Script jetable, à supprimer après usage.
+
+Usage : python tools/probe_qwen3vl_processor.py
+"""
+from transformers import AutoProcessor
+from PIL import Image
+import torch
+
+MODEL_ID = "Qwen/Qwen3-VL-8B-Instruct"
+FIXTURE = "data/training/live_sessions/bg_DESKTOP-58D5CAC_windows/shots/heartbeat_1773792436.png"
+
+print("=" * 70)
+print("DUMP PROCESSOR :", MODEL_ID)
+print("=" * 70)
+
+proc = AutoProcessor.from_pretrained(MODEL_ID)
+ip = proc.image_processor
+
+# Section 1 — Attributs bruts
+print("\n--- ATTRIBUTS BRUTS ---")
+print("class:", type(ip).__name__)
+print("size:", ip.size)
+print("patch_size:", ip.patch_size)
+print("merge_size:", ip.merge_size)
+for attr in ['min_pixels', 'max_pixels', 'temporal_patch_size',
+             'image_mean', 'image_std', 'do_resize', 'do_rescale',
+             'rescale_factor', 'do_normalize', 'do_convert_rgb']:
+    print(f"{attr}:", getattr(ip, attr, '<absent>'))
+
+# Section 2 — Comportement effectif sur fixture
+print("\n--- COMPORTEMENT EFFECTIF SUR FIXTURE ---")
+img = Image.open(FIXTURE)
+print(f"Image source : {img.size} (W×H)")
+
+out = ip(images=img, return_tensors='pt')
+print(f"Keys retournées : {list(out.keys())}")
+print(f"pixel_values shape : {out['pixel_values'].shape}")
+print(f"image_grid_thw : {out.get('image_grid_thw')}")
+
+# Section 3 — Reconstruction des dimensions resize
+print("\n--- RECONSTRUCTION DIMS RESIZE ---")
+grid = out.get('image_grid_thw')
+if grid is not None:
+    grid = grid[0].tolist()  # [t, h, w]
+    factor = ip.patch_size * ip.merge_size
+    H_resized = grid[1] * factor
+    W_resized = grid[2] * factor
+    print(f"grid_thw : t={grid[0]}, h={grid[1]}, w={grid[2]}")
+    print(f"factor calculé (patch_size × merge_size) : {factor}")
+    print(f"Dims resize reconstruites : {W_resized}×{H_resized} (W×H)")
+    print(f"Dims source : {img.size}")
+    print(f"Ratio resize : {W_resized / img.size[0]:.4f} (W), "
+          f"{H_resized / img.size[1]:.4f} (H)")
+
+# Section 4 — Test borne haute pour comprendre min/max_pixels
+print("\n--- TEST BORNE HAUTE (image grande) ---")
+big_img = Image.new('RGB', (4096, 2560), color='white')
+big_out = ip(images=big_img, return_tensors='pt')
+big_grid = big_out['image_grid_thw'][0].tolist()
+factor = ip.patch_size * ip.merge_size
+print(f"Image source : {big_img.size}")
+print(f"grid_thw : {big_grid}")
+print(f"Dims resize : {big_grid[2] * factor}×{big_grid[1] * factor}")
+print(f"Pixels totaux après resize : "
+      f"{big_grid[1] * factor * big_grid[2] * factor}")
+
+# Section 5 — Test borne basse (image petite)
+print("\n--- TEST BORNE BASSE (image petite) ---")
+small_img = Image.new('RGB', (128, 64), color='white')
+small_out = ip(images=small_img, return_tensors='pt')
+small_grid = small_out['image_grid_thw'][0].tolist()
+print(f"Image source : {small_img.size}")
+print(f"grid_thw : {small_grid}")
+print(f"Dims resize : {small_grid[2] * factor}×{small_grid[1] * factor}")
+
+print("\n" + "=" * 70)
+print("FIN DUMP")
+print("=" * 70)