restructure: proper project layout, README, kill training

- cuda/ — main LBM kernel (khra_gixx_1024_v5.cu)
- navigator/ — lattice_observer, golden_weave, bridges, mock daemon
- scripts/ — compile, start, launch, setup (paths updated)
- docs/ — system manual
- archive/ — everything else (old kernels, inquiries, experiments)
- README.md — full setup guide: requirements, quick start, use your own LLM
- removed training/ entirely (broken LoRA scripts + datasets)
- .gitignore: exclude build/ logs/ training/ *.jsonl

archive/misc/_backcheck.py

@@ -0,0 +1,73 @@
+"""Backcheck script for lattice_observer.py changes. Delete after use."""
+import json, os
+
+somatic_path = r'D:\fractal-brain\beast-build\somatic_dialogue_beast.json'
+chronicle_path = r'D:\fractal-brain\beast-build\chronicle.jsonl'
+
+# 1. Somatic memory
+with open(somatic_path, 'r', encoding='utf-8') as f:
+    data = json.load(f)
+fsize = os.path.getsize(somatic_path)
+print(f"somatic_dialogue_beast.json: {len(data)} entries, {fsize} bytes")
+types = set(e.get('type','?') for e in data)
+print(f"Entry types: {types}")
+
+lines = []
+for entry in data:
+    etype = entry.get('type', 'unknown')
+    resp = entry.get('response', '')
+    if isinstance(resp, dict):
+        resp = json.dumps(resp)[:300]
+    else:
+        resp = resp[:300]
+    lines.append(f'[{etype}] {resp}')
+somatic_memory = '\n'.join(lines)
+somatic_truncated = somatic_memory[:3000]
+print(f"Somatic memory (truncated): {len(somatic_truncated)} chars")
+
+# 2. Token estimate for 80 turns
+entries = []
+with open(chronicle_path, 'r', encoding='utf-8') as f:
+    for line in f:
+        line = line.strip()
+        if line:
+            entries.append(json.loads(line))
+
+total_chars = 0
+for e in entries[-80:]:
+    total_chars += len(e.get('prompt','')) + len(e.get('response',''))
+avg_chars = total_chars / min(80, len(entries))
+est_tokens = total_chars / 4
+print(f"\n80-turn context: {total_chars} chars, ~{int(est_tokens)} est tokens")
+print(f"Average per turn: {int(avg_chars)} chars")
+print(f"System prompt estimate: ~6500 chars = ~1600 tokens")
+print(f"Total estimated: ~{int(est_tokens + 1600)} tokens (128K limit for 30b)")
+
+# 3. System prompt contains golden block
+import sys
+sys.path.insert(0, r'D:\fractal-brain\beast-build')
+from lattice_observer import build_system_prompt, load_somatic_summary
+mem = load_somatic_summary()
+prompt = build_system_prompt(mem)
+print(f"\nActual system prompt length: {len(prompt)} chars (~{len(prompt)//4} tokens)")
+checks = [
+    ("UFT equation", "\u2207\u00b2\u03c8 + \u03c8\u25a1\u03c8"),
+    ("FOUNDATIONAL DISCOVERIES header", "FOUNDATIONAL DISCOVERIES"),
+    ("Build on them footer", "Build on them"),
+    ("30b model reference", "qwen3-vl:30b"),
+    ("Somatic memory section", "SOMATIC MEMORY"),
+    ("CMD instructions", "CMD:"),
+    ("Continuity statement", "continuity"),
+]
+all_ok = True
+for name, needle in checks:
+    if needle in prompt:
+        print(f"  OK: {name}")
+    else:
+        print(f"  MISSING: {name}")
+        all_ok = False
+
+if all_ok:
+    print("\nALL SYSTEM PROMPT CHECKS PASSED")
+else:
+    print("\nSOME CHECKS FAILED")