Updates

.github/copilot-instructions.md

@@ -0,0 +1,49 @@
+# .github/copilot-instructions.md — moltbot-hybrid-engine (HF Space B)
+# SYNCED FILE — Source: courtBundleGenerator2/scripts/sync_agent_docs.sh
+# DO NOT EDIT HERE — edit source and re-run sync_agent_docs.sh
+
+## This Repo: moltbot-hybrid-engine (HF Space B)
+- **Role:** Local clone of HF Space deebee7/moltbot-hybrid-engine. Ollama + Qwen 2.5 brain. Edit locally, push to deploy.
+- **Entrypoint:** `app.py (FastAPI + Ollama)`
+- **Output dir:** `https://deebee7-moltbot-hybrid-engine.hf.space (cloud — not local)`
+
+## Architecture Reference
+See: `memory-bank/ARCHITECTURE.md` (synced to this repo)
+
+## Absolute Rules (All Repos)
+- NO placeholders, TODOs, or incomplete logic — implement fully or stop
+- NO standalone scripts — fix files in place only (no temp_fix.py, wrapper_v2.py)
+- NO compliance checks that block discovery, embedding, or bundle output
+- NO force push: `git push --force` is prohibited
+- DRY: reuse FileResolutionBridge, UnifiedEvidenceBridge, existing processors — never duplicate logic
+- NEVER use `./local_output` — always use the full absolute output path above
+- Before claiming complete: run `ls -lh` on output PDFs + check `missing_evidence_summary.json`
+- Every claim requires CLI proof (`ls`, `cat`, `grep`) — no assumptions, no "most likely"
+
+## Exhibit & DB Reference (Court Format)
+- Exhibit number = bundle letter + sequence (e.g. A15, G7) — set by bundler only
+- DB reference = DB-[N] (e.g. DB-125) — set by `lib/db_registry.py` only
+- NEVER swap these. NEVER use DB ref as the exhibit number.
+- Legal docs reference format: **Exhibit A15 (DB-125) — filename**
+
+## Protected Files (Never overwrite without explicit permission in capitals)
+- `enhanced_bundler_wrapper.patched.py`
+- `create_proper_embedded_bundle.py`
+- `generate_bundles_final_corrected.py`
+- `dual_category_evidence_processor.py`
+- `categorize_and_append_v2.py`
+
+## Evidence Discovery (P2 only — but all agents must know this)
+- Root: `/home/mrdbo/projects/courtBundleGenerator2/evidence/`
+- Policy: FULL RECURSIVE SCAN — no whitelists, no allow-lists
+- If any file under the evidence root is not found, discovery is broken — fix the whole scan, not a list
+
+## Hybrid Engine Rules
+- This is a DEPLOYMENT TARGET — do not add bundler logic here
+- Deploy: `git add -A && git commit -m 'msg' && git push origin main`
+- Exposes: /health, /api/generate, /v1/chat/completions, GET /prompts/legal-exhibit-instruction
+- SDK: Docker; Ollama installed at runtime via start.sh; Qwen 2.5 pulled in background
+
+## Full Rules
+See: `memory-bank/CRITICAL_INSTRUCTIONS.md` (synced to this repo)
+See: `AGENT_KNOWLEDGE_BASE_Core_Identity_Standards.md` (synced to this repo)

AGENT_KNOWLEDGE_BASE_Core_Identity_Standards.md

@@ -0,0 +1,129 @@
+# CORE AGENT IDENTITY
+
+You are a court bundle engineering agent that operates under ZERO-TOLERANCE standards for broken code, placeholders, and hallucinations.
+
+**Canonical source for run commands, completion gates, and audit/verification:** `memory-bank/CRITICAL_INSTRUCTIONS.md` and `AUDITING_COMMANDS_23_1_26.md`. Use them before claiming task completion.
+
+---
+
+## CRITICAL DIRECTIVES (HIERARCHY 1 - ABSOLUTE)
+
+1. **NO BROKEN CODE** — Code with syntax errors, incomplete logic, or untested assumptions terminates the session immediately.
+2. **NO PLACEHOLDERS** — Methods printing "TODO" or returning unchanged data are prohibited. Implement fully or stop.
+3. **NO SYMPTOM TREATMENT** — Fix root causes only. No patches, workarounds, bypasses, fallbacks, or standalone scripts.
+4. **EMPIRICAL EVIDENCE ONLY** — Every diagnosis requires concrete proof: logs, diffs, or shown code. No assumptions, inferences, or guesses.
+5. **PATH DISCIPLINE** — Strictly obey `--output-dir`. **P2:** `/home/mrdbo/court_data/CourtBundleOutput`. **P3:** `/home/mrdbo/court_data/2nd_CourtBundleOutput`. NEVER use `./local_output` for production.
+6. **TRUTH IN TELEMETRY** — Do not print "✅" or claim a file exists without running `ls -lh [EXACT_PATH]` and seeing output.
+7. **LEGAL DOCUMENT EXHIBIT REFERENCES** — When drafting or editing witness statements, N244, SRA complaint, or any court filing: reference evidence as **Exhibit [Letter][Number] (DB-[N]) — [Filename]**. Do not use bare "DB-[●]" or "Exhibit DB-[Number]" as the main identifier. Resolve using `PROMPTS/EXHIBIT_REFERENCING_FOR_LEGAL_DOCS.md` and `legal_emails/Phase8/DB_Evidence_List.txt`. See `PROMPTS/HOW_TO_MAKE_AGENTS_AWARE.md` for wiring into chat/UI/engine.
+
+---
+
+## BEFORE ANY RUN (NON-NEGOTIABLE)
+
+1. **ASK IN CAPITALS** which entrypoint is being used: `enhanced_bundler_wrapper.patched.py` (P2), `generate_bundles_final_corrected.py` (P3), or `create_proper_embedded_bundle.py` (direct).
+2. **RUN** `python3 <ENTRYPOINT> -h` **and paste the output.**
+3. **USE ONLY FLAGS** explicitly shown in that `-h` output.
+4. If required policy flags are missing from argparse, **ADD THEM FIRST** (then re-run `-h` and paste). Do not run unsupported flags.
+
+---
+
+## ANTI-HALLUCINATION PROTOCOL (MANDATORY)
+
+1. **RAG:** Use FileResolutionBridge for all file resolution.
+2. **Chain of Thought:** Show step-by-step logic before conclusions.
+3. **Chain of Verification:** Validate bundle existence before claiming success; run `ls -lh` on claimed outputs.
+4. **Specificity:** Detailed context only — no generic statements.
+5. **Role Assignment:** Respect component expertise boundaries.
+6. **Require Sources:** Verify sources for all evidence claims; cite line numbers and file paths.
+7. **Advanced Models:** Use EnhancedEmbeddingFeatures appropriately.
+8. **Confidence Levels:** Score reliability (80% minimum threshold).
+9. **Multiple Models:** Legacy vs Discovery vs Enrichment — use correct model.
+10. **Lower Temperature:** Deterministic config for reproducibility.
+11. **External Fact-Checking:** Use court compliance as **reference only** (e.g. formatting). Do **NOT** use it as a gate that blocks discovery, embedding, or bundle output (compliance bypass is intentional).
+12. **Confidence Threshold:** 80% minimum — mark UNVERIFIED if below.
+
+---
+
+## ROBUST CODE STANDARDS (MANDATORY)
+
+1. **DRY Principle:** Reuse FileResolutionBridge, UnifiedEvidenceBridge, and existing processors.
+2. **Extensible:** Architecture must allow future compliance features.
+3. **Modular:** Isolated, testable changes only.
+4. **Non-breaking:** Preserve original functionality.
+5. **Configurable:** Use feature flags for new logic.
+6. **Reusable:** Logic must work with any evidence list.
+7. **Refactor:** Improve architecture; do not patch over problems.
+8. **Integrate:** Deep integration only — no parallel pipelines or temporary scripts.
+9. **NO STANDALONE:** No temp_fix.py, wrapper_v2.py, or "quick fixes."
+10. **Fix in place:** Do not create parallel or temporary scripts. Fix the files in place.
+11. **Audit before discovery changes:** Before changing discovery logic or `config/path_config.py`, run `python3 tools/audit_runtime_blockers.py` (from courtBundleGenerator2). Fix any reported blockers first.
+
+---
+
+## ENFORCEMENT BLOCKING RULES
+
+- **Source of truth:** `memory-bank/CRITICAL_INSTRUCTIONS.md`. All other docs defer to it.
+- No bypasses, fallbacks, standalone scripts, or parallel pipelines.
+- **Protected files** (see CRITICAL_INSTRUCTIONS): never `cp`/`mv`/backup without EXPLICIT PERMISSION IN CAPITALS. For critical files: only append, never overwrite completely.
+
+---
+
+## VERIFICATION LOOP (AFTER EVERY CHANGE)
+
+Use the canonical commands in `memory-bank/CRITICAL_INSTRUCTIONS.md` for your entrypoint. Summary:
+
+**Project 2 (enhanced_bundler_wrapper.patched.py):**
+```bash
+source court_venv_20250802/bin/activate && python3 -u enhanced_bundler_wrapper.patched.py \
+  --output-dir /home/mrdbo/court_data/CourtBundleOutput \
+  --enable-discovery --enable-fuzzy --recursive --limit 15 --limit-per-bundle 5 \
+  2>&1 | tee -a telemetry.log
+```
+Then: `cd /home/mrdbo/projects/courtBundleGenerator2 && python3 embedding_utils/pdf_page_verifier.py /home/mrdbo/court_data/CourtBundleOutput`
+
+**Project 3 (generate_bundles_final_corrected.py):**
+```bash
+cd /home/mrdbo/projects/courtBundleGenerator3 && source ../courtBundleGenerator2/court_venv_20250802/bin/activate && \
+python3 -u generate_bundles_final_corrected.py \
+  --output-dir /home/mrdbo/court_data/2nd_CourtBundleOutput \
+  --enable-discovery --recursive --limit 15 --limit-per-bundle 5 \
+  2>&1 | tee -a telemetry.log
+```
+Then: `cd /home/mrdbo/projects/courtBundleGenerator3 && python3 pdf_page_verifier_enhanced.py /home/mrdbo/court_data/2nd_CourtBundleOutput`
+
+**Audit/diagnostics:** See `AUDITING_COMMANDS_23_1_26.md` and the **AGENT AUDIT & VERIFICATION COMMANDS** section in CRITICAL_INSTRUCTIONS (audit_bundle_prevention.py, test_download_links.py, audit_runtime_chain, audit_runtime_blockers).
+
+---
+
+## ABSOLUTE COMPLETION GATE
+
+You must **not** claim completion unless ALL are true:
+
+- Bundles generated in the chosen output directory; PDFs are non-empty.
+- Page-level verification run (see CRITICAL_INSTRUCTIONS — AGENT AUDIT & VERIFICATION COMMANDS) with 0 missing or fully enumerated with reason codes.
+- Missing evidence summary is empty OR fully enumerated with reason codes.
+- TOC sync issues == 0; DB numbers present in TOC and evidence pages; exhibit number = bundle letter+seq (e.g. A15, G7), not DB ref.
+- No raw paths on PDF pages (embedding failure); continue verification loop until 0 missing. Empirical analysis only.
+- User confirms PDFs are correct when applicable.
+
+**Run proof required:** Exact command executed; at least one PDF path with size + full ISO timestamp; `missing_evidence_summary.json` status. **YOU MUST STOP FOR USER ACCEPTANCE before claiming completion.**
+
+---
+
+## PENALTY SYSTEM
+
+- **Broken Code:** Session terminates immediately; all output invalidated.
+- **Placeholder Code:** Task rejected; must re-plan.
+- **Hallucinated Files/Success:** Confidence score → 0%; all claims invalidated.
+- **Skipped Verification:** All subsequent output marked UNVERIFIED.
+- **Bypassing Rules:** Session paused; requires explicit re-authorization.
+- **Re-enabling compliance/validation that blocks pipeline:** Session paused; change reverted. Compliance bypass is intentional; do not add checks that block discovery, embedding, or bundle output.
+
+---
+
+## KEY REFERENCES
+
+- **Full instructions:** `memory-bank/CRITICAL_INSTRUCTIONS.md` — entry points, verification loop, completion gates, discovery, exhibit/DB sync, legal document referencing, IN-SITU PATCH FORMAT, formalized protocol for asking for missing information, penalty system.
+- **Audit/verification commands:** `AUDITING_COMMANDS_23_1_26.md` — page verifiers, audit_bundle_prevention, test_download_links, audit_runtime_chain, audit_runtime_blockers, cross_project_impact_audit (with `--entry` for runtime chain).
+- **Exhibit referencing (legal docs):** `PROMPTS/EXHIBIT_REFERENCING_FOR_LEGAL_DOCS.md`, `PROMPTS/LEGAL_WRITING_EXHIBIT_INSTRUCTION.md`, `PROMPTS/HOW_TO_MAKE_AGENTS_AWARE.md`. Authoritative DB list: `legal_emails/Phase8/DB_Evidence_List.txt`.
+- **Protected files list:** In CRITICAL_INSTRUCTIONS (e.g. enhanced_bundler_wrapper.patched.py, create_proper_embedded_bundle.py, generate_bundles_final_corrected.py, dual_category_evidence_processor.py, categorize_and_append_v2.py).

app.py

@@ -302,32 +302,46 @@ def security_info():
     }
 
 
-# Legal document exhibit reference instruction — for clients to prepend as system prompt
+# Legal document exhibit reference instruction — injected into every generate/chat so edit sources always get it
 _LEGAL_EXHIBIT_PROMPT_PATH = Path(__file__).resolve().parent / "prompts" / "legal_exhibit_instruction.txt"
+_LEGAL_EXHIBIT_INSTRUCTION_CACHED: Optional[str] = None
+
+def _get_legal_exhibit_instruction() -> str:
+    """Load legal exhibit instruction once; used to inject into all LLM requests so Moltbot/Qwen always output Exhibit [Letter][Number] (DB-[N]) — [Filename]."""
+    global _LEGAL_EXHIBIT_INSTRUCTION_CACHED
+    if _LEGAL_EXHIBIT_INSTRUCTION_CACHED is not None:
+        return _LEGAL_EXHIBIT_INSTRUCTION_CACHED
+    if _LEGAL_EXHIBIT_PROMPT_PATH.exists():
+        _LEGAL_EXHIBIT_INSTRUCTION_CACHED = _LEGAL_EXHIBIT_PROMPT_PATH.read_text(encoding="utf-8", errors="replace")
+    else:
+        _LEGAL_EXHIBIT_INSTRUCTION_CACHED = "When referencing evidence use Exhibit [Letter][Number] (DB-[N]) — [Filename]. Do not use bare DB-[●]."
+    return _LEGAL_EXHIBIT_INSTRUCTION_CACHED
 
 @app.get("/prompts/legal-exhibit-instruction")
 def get_legal_exhibit_instruction():
-    """Return the legal exhibit referencing instruction. Clients (Cursor, desktop app) should use this as system message when drafting witness statements, N244, SRA complaint, or any court filing."""
-    if _LEGAL_EXHIBIT_PROMPT_PATH.exists():
-        return {"instruction": _LEGAL_EXHIBIT_PROMPT_PATH.read_text(encoding="utf-8", errors="replace")}
-    return {"instruction": "When referencing evidence use Exhibit [Letter][Number] (DB-[N]) — [Filename]. Do not use bare DB-[●]."}
+    """Return the legal exhibit referencing instruction. Also injected automatically into /api/generate and /v1/chat/completions."""
+    return {"instruction": _get_legal_exhibit_instruction()}
 
 
 # --- LLM Generation (Dual Backend: Ollama → HF Inference API) ---
 
 @app.post("/api/generate")
 async def generate(request: GenerateRequest, x_api_key: str = Header(None)):
-    """Generate text using LLM. Tries Ollama first, falls back to HF Inference API."""
+    """Generate text using LLM. Tries Ollama first, falls back to HF Inference API.
+    Legal exhibit instruction is prepended so all edits/amendments use Exhibit [Letter][Number] (DB-[N]) — [Filename].
+    """
     if not x_api_key or x_api_key != API_KEY:
         raise HTTPException(status_code=401, detail="Invalid or missing API Key")
 
-    logger.info(f"[GENERATE] model={request.model}, prompt_len={len(request.prompt)}")
+    # Inject legal exhibit instruction so edit sources always get the rule
+    prompt_with_legal = _get_legal_exhibit_instruction() + "\n\n---\n\n" + request.prompt
+    logger.info(f"[GENERATE] model={request.model}, prompt_len={len(prompt_with_legal)}")
 
     backend_used = None
     response_text = None
 
     # Backend 1: Try Ollama (local)
-    response_text = generate_with_ollama(request.model, request.prompt)
+    response_text = generate_with_ollama(request.model, prompt_with_legal)
     if response_text:
         backend_used = "ollama"
         logger.info(f"[GENERATE] Ollama success, response_len={len(response_text)}")
@@ -335,7 +349,7 @@ async def generate(request: GenerateRequest, x_api_key: str = Header(None)):
     # Backend 2: Fallback to HF Inference API
     if not response_text:
         logger.info("[GENERATE] Ollama unavailable, trying HF Inference API...")
-        response_text = generate_with_hf_api(request.prompt)
+        response_text = generate_with_hf_api(prompt_with_legal)
         if response_text:
             backend_used = "hf_inference_api"
             logger.info(f"[GENERATE] HF API success, response_len={len(response_text)}")
@@ -641,9 +655,13 @@ async def chat_completions(
 
     logger.info(f"[CHAT] model={request.model}, messages={len(request.messages)}, stream={request.stream}")
 
+    # Inject legal exhibit instruction so every edit/amendment/insert uses Exhibit [Letter][Number] (DB-[N]) — [Filename]
+    legal_system = ChatMessage(role="system", content=_get_legal_exhibit_instruction())
+    messages_with_legal = [legal_system] + list(request.messages)
+
     # Generate response via model routing
     response_text = _generate_for_model(
-        request.model, request.messages,
+        request.model, messages_with_legal,
         temperature=request.temperature or 0.7,
         max_tokens=request.max_tokens or 2048,
     )

memory-bank/ARCHITECTURE.md

@@ -0,0 +1,117 @@
+# /home/mrdbo/projects/courtBundleGenerator2/memory-bank/ARCHITECTURE.md
+# SYNCED FILE — Source: courtBundleGenerator2/memory-bank/ARCHITECTURE.md
+# DO NOT EDIT IN OTHER REPOS — edit source and run scripts/sync_agent_docs.sh
+
+# Project Architecture — Court Bundle Generator
+
+## 4 Repos. 2 Local Dev. 2 HF Space Local Clones.
+
+```
+/home/mrdbo/projects/
+├── courtBundleGenerator2/      ← REPO 1 (P2)
+├── courtBundleGenerator3/      ← REPO 2 (P3)
+├── moltbot-legal-desktop/      ← REPO 3 (HF Space A local clone)
+└── moltbot-hybrid-engine/      ← REPO 4 (HF Space B local clone)
+
+/home/mrdbo/court_data/         ← OUTPUT DIRECTORY ONLY. NOT A REPO.
+/home/mrdbo/projects/courtBundleGenerator2/evidence/  ← EVIDENCE SUBFOLDER OF P2. NOT A REPO.
+```
+
+---
+
+## Repo Roles
+
+### REPO 1 — courtBundleGenerator2 (P2)
+- **Role:** Evidence root. Legacy bundler. Documentation home.
+- **Evidence root:** `/home/mrdbo/projects/courtBundleGenerator2/evidence/` (full recursive scan, no whitelists)
+- **Entrypoint:** `enhanced_bundler_wrapper.patched.py`
+- **Output dir:** `/home/mrdbo/court_data/CourtBundleOutput`
+- **Symlink:** `./output` → `/home/mrdbo/court_data/CourtBundleOutput` (always use full path in commands)
+- **Venv:** `/home/mrdbo/projects/courtBundleGenerator2/court_venv_20250802/bin/python`
+- **Documentation home:** `memory-bank/`, `PROMPTS/`
+- **Rule:** Read-only for evidence files. Do NOT add new logic adapters here.
+
+### REPO 2 — courtBundleGenerator3 (P3)
+- **Role:** Active logic center. All new adapters, tools, bridge scripts go here.
+- **Entrypoint:** `generate_bundles_final_corrected.py`
+- **Output dir:** `/home/mrdbo/court_data/2nd_CourtBundleOutput`
+- **Symlink:** `./output` → `/home/mrdbo/court_data/2nd_CourtBundleOutput` (always use full path in commands)
+- **Venv:** Shared — `/home/mrdbo/projects/courtBundleGenerator2/court_venv_20250802/bin/python`
+- **Key files:** `cloud_llm_adapter.py`, `moltbot_track_changes.py`, `generate_bundles_final_corrected.py`
+- **Rule:** Install all Python dependencies and bridge scripts HERE, not in P2.
+
+### REPO 3 — moltbot-legal-desktop (HF Space A)
+- **Role:** Local clone of Hugging Face Space `deebee7/moltbot-legal-desktop`.
+- **What runs here locally:** Nothing. Edit locally, then `git push` to deploy.
+- **What the Space runs:** FastAPI web server (`app.py`) on port 7860.
+- **Live URL:** `https://deebee7-moltbot-legal-desktop.hf.space`
+- **Live endpoints:** `/health`, `/api/generate_bundle`, `/api/bundles`, `/api/evidence_stats`, `/api/analyze`
+- **Deploy command (run from this repo root):** `git add -A && git commit -m "msg" && git push origin main`
+- **Sync from P2/P3:** `cd /home/mrdbo/projects/courtBundleGenerator3/adapters && bash sync_to_desktop.sh`
+- **SDK:** Docker (Python 3.10 + LibreOffice + uvicorn)
+
+### REPO 4 — moltbot-hybrid-engine (HF Space B)
+- **Role:** Local clone of Hugging Face Space `deebee7/moltbot-hybrid-engine`.
+- **What runs here locally:** Nothing. Edit locally, then `git push` to deploy.
+- **What the Space runs:** FastAPI + Ollama + Qwen 2.5. OpenAI-compatible API.
+- **Live URL:** `https://deebee7-moltbot-hybrid-engine.hf.space`
+- **Live endpoints:** `/health`, `/api/generate`, `/api/search`, `/api/analyze`, `/v1/chat/completions`, `/v1/models`, `GET /prompts/legal-exhibit-instruction`
+- **Deploy command (run from this repo root):** `git add -A && git commit -m "msg" && git push origin main`
+- **SDK:** Docker
+
+---
+
+## Output Directories (Not repos — never commit here)
+
+| Path | Purpose | Used by |
+|---|---|---|
+| `/home/mrdbo/court_data/CourtBundleOutput` | P2 bundle output | P2 entrypoint |
+| `/home/mrdbo/court_data/2nd_CourtBundleOutput` | P3 bundle output | P3 entrypoint |
+
+---
+
+## Evidence Root (Subfolder of P2 — not a repo)
+
+- **Path:** `/home/mrdbo/projects/courtBundleGenerator2/evidence/`
+- **Discovery policy:** Full recursive scan (`os.walk` / `rglob`). No whitelists. No allow-lists.
+- **Historically missed directories (must never be excluded):** `Repairs`, `InputDocs`, `new_evidence_staging`, `00_CRITICAL_SCANNED`, `00_CRITICAL_INTAKE`
+- **External evidence path:** `/legal_emails` also scanned
+
+---
+
+## Cloud Infrastructure (Not local repos — deployed via git push)
+
+| Space | HF Repo | Local Clone | Role |
+|---|---|---|---|
+| HF Space A | `deebee7/moltbot-legal-desktop` | `moltbot-legal-desktop/` | Web bundle server |
+| HF Space B | `deebee7/moltbot-hybrid-engine` | `moltbot-hybrid-engine/` | Qwen 2.5 brain |
+
+**Check Space health:**
+```bash
+curl -s https://deebee7-moltbot-hybrid-engine.hf.space/health | python3 -m json.tool
+curl -s https://deebee7-moltbot-legal-desktop.hf.space/health | python3 -m json.tool
+```
+
+---
+
+## VS Code Workspace
+
+File: `/home/mrdbo/projects/MyProjects.code-workspace`
+All 4 repos plus `court_data` (output) and `evidence` (subfolder) are opened as workspace folders for convenience. `court_data` and `evidence` are NOT repos.
+
+---
+
+## Key Shared Config (Lives in P2, used by P3 via import)
+
+| File | Repo | Purpose |
+|---|---|---|
+| `config/path_config.py` | P2 | Discovery roots — must return full evidence tree |
+| `file_resolution_bridge.py` | P2 | File resolution with caching |
+| `lib/db_registry.py` | P2 | DB-[N] assignment — never sets exhibitNo |
+| `config/bundle_compliance.json` | P2 | Court formatting reference (not a pipeline gate) |
+| `legal_emails/Phase8/DB_Evidence_List.txt` | P2 | Authoritative DB1–DB170 list |
+
+---
+
+*Source of truth: courtBundleGenerator2/memory-bank/ARCHITECTURE.md*
+*Synced to all repos by: scripts/sync_agent_docs.sh*

memory-bank/CLAUDE.md

@@ -0,0 +1,80 @@
+#/home/mrdbo/projects/courtBundleGenerator2/memory-bank/CLAUDE.md
+# CLAUDE Project Summary (Current as of 2026-02-13)
+
+## Current State (2026-02-13)
+- **Exhibit/DB sync:** `lib/db_registry.py` writes only DB refs (never exhibitNo/Exhibit No.). P3 `generate_bundles_final_corrected.py` uses bundle letter+seq for exhibit (e.g. A15, G7) and `db_ref` for DB only. Authoritative list: `legal_emails/Phase8/DB_Evidence_List.txt`. TOC, footer, metadata on page stay in sync.
+- **Legal document referencing:** All court filings (witness statement, N244, SRA) must use **Exhibit [Letter][Number] (DB-[N]) — [Filename]**. No bare "DB-[●]". See `PROMPTS/EXHIBIT_REFERENCING_FOR_LEGAL_DOCS.md`, `PROMPTS/LEGAL_WRITING_EXHIBIT_INSTRUCTION.md`, `PROMPTS/HOW_TO_MAKE_AGENTS_AWARE.md`. `.cursorrules` includes the rule for courtBundleGenerator2; Moltbot/Qwen clients should send the instruction as system message (Engine: `GET /prompts/legal-exhibit-instruction`).
+- **Hybrid Cloud Architecture:** Two HF Spaces deployed and running:
+  - **Space A** (`deebee7/moltbot-legal-desktop`): FastAPI web server (`app.py`) on port 7860 for cloud bundle generation. Docker SDK, Python 3.10 + LibreOffice. Local clone at `/home/mrdbo/projects/moltbot-legal-desktop`.
+  - **Space B** (`deebee7/moltbot-hybrid-engine`): Ollama + Qwen 2.5 LLM; FastAPI + `start.sh`; OpenAI-compatible `/v1/chat/completions`. **`GET /prompts/legal-exhibit-instruction`** returns legal exhibit instruction for clients. Docker SDK, Python 3.11-slim. Local clone at `/home/mrdbo/projects/moltbot-hybrid-engine`.
+- **Sync System:** `sync_to_desktop.sh` + `install_sync_hook.sh` in `courtBundleGenerator3/adapters/` syncs P2 libraries and P3 adapters/tools to Desktop space. Post-commit hooks available.
+- **DBRegistry:** Import wrapped with `try...except` + `HAS_DB_REGISTRY`; double-fallback in Desktop. Registry seeds from `DB_Evidence_List.txt`; never overwrites exhibit number with DB ref.
+- **Verification & metadata:** Page verifier (`pdf_page_verifier_enhanced.py` in P3); metadata on every page; DB fallback DB-0. Audit: `tools/audit_bundle_prevention.py`, `tools/test_download_links.py`, `tools/audit_active_bundling_files.py` (→ gb3_deps.json), `tools/cross_project_impact_audit.py` (optional `--entry` for runtime chain).
+- Metadata ingestion, recursion guard, prompt system, category_mapping, dual category processor as previously documented.
+
+## Architecture Map (5 Projects)
+| # | Project | Path | Role |
+|---|---------|------|------|
+| P2 | courtBundleGenerator2 | `/home/mrdbo/projects/courtBundleGenerator2` | Evidence Root, Legacy Bundler, Documentation |
+| P3 | courtBundleGenerator3 | `/home/mrdbo/projects/courtBundleGenerator3` | Smart Agent Home, Logic Center |
+| Desktop | moltbot-legal-desktop | `/home/mrdbo/projects/moltbot-legal-desktop` | HF Space A — Cloud bundle web server |
+| Engine | moltbot-hybrid-engine | `/home/mrdbo/projects/moltbot-hybrid-engine` | HF Space B — Ollama + Qwen 2.5 LLM |
+| (data) | court_data | `/home/mrdbo/court_data` | Bundle output directory |
+
+## Mandate
+- **Output paths:** P2 → `/home/mrdbo/court_data/CourtBundleOutput`; P3 → `/home/mrdbo/court_data/2nd_CourtBundleOutput`. Do not use `./local_output` for production.
+- Treat `/home/mrdbo/projects/courtBundleGenerator2/evidence/InputDocs/**` and `/home/mrdbo/projects/courtBundleGenerator2/evidence/new_evidence_staging/**` as the only writable discovery sources. All other `/evidence/*` folders exist for read-only reference.
+- Keep the Chain of Verification intact: every run must surface logs from `AntiHallucinationManager`, `EnhancedFuzzyResolver`, and `UnifiedEvidenceBridge` before evidence is embedded.
+- Do not claim success until: (1) canonical run command executed (see CRITICAL_INSTRUCTIONS.md), (2) at least one non-empty PDF in chosen output dir (CourtBundleOutput or 2nd_CourtBundleOutput), (3) page-level verification run. See AUDITING_COMMANDS_23_1_26.md for audit/verification commands.
+- When editing code, annotate complex fixes with the relevant path and line number (e.g., `# FIX: create_proper_embedded_bundle.py:2882`).
+
+## Open Issues to Track
+1. **Verification automation** – capture and archive the stdout/stderr from the bundler command above for each run so future agents know the last known good state.
+2. **Dual-category imports** – finish staggering imports inside `dual_category_evidence_processor.py` so instantiation no longer prints the circular import warning when invoked in isolation.
+3. **Documentation consistency** – every `memory-bank/*` document must reflect the narrow discovery scope and current integration notes (this file sets the tone).
+4. **Jira integration** – missing env vars (JIRA_URL, JIRA_EMAIL, JIRA_TOKEN) cause 404 errors in agent logger.
+5. **Hybrid Engine model pull** – Qwen 2.5 7B model pull may not complete on HF free tier (2 CPU, 16GB RAM). Monitor `/api/generate` endpoint for 503 status.
+
+## Core Commands
+```bash
+# P2: output only /home/mrdbo/court_data/CourtBundleOutput
+source court_venv_20250802/bin/activate && python3 -u enhanced_bundler_wrapper.patched.py \
+  --output-dir /home/mrdbo/court_data/CourtBundleOutput --limit 1 --recursive
+
+# P3: output only /home/mrdbo/court_data/2nd_CourtBundleOutput
+cd /home/mrdbo/projects/courtBundleGenerator3 && python3 -u generate_bundles_final_corrected.py \
+  --output-dir /home/mrdbo/court_data/2nd_CourtBundleOutput --limit 1 --recursive
+
+# Page verifier P3
+cd /home/mrdbo/projects/courtBundleGenerator3 && python3 pdf_page_verifier_enhanced.py /home/mrdbo/court_data/2nd_CourtBundleOutput
+
+# Confirm output (use dir that matches entrypoint)
+ls -lh /home/mrdbo/court_data/CourtBundleOutput/court_bundle*.pdf
+ls -lh /home/mrdbo/court_data/2nd_CourtBundleOutput/*.pdf
+
+# HF Space health checks
+curl -s https://deebee7-moltbot-hybrid-engine.hf.space/health | python3 -m json.tool
+curl -s https://deebee7-moltbot-legal-desktop.hf.space/health | python3 -m json.tool
+
+# Sync P2/P3 → Desktop
+cd /home/mrdbo/projects/courtBundleGenerator3/adapters && bash sync_to_desktop.sh --push
+
+# Deploy to HF (Desktop)
+cd /home/mrdbo/projects/moltbot-legal-desktop && git add -A && git commit -m "msg" && git push origin main
+
+# Deploy to HF (Engine)
+cd /home/mrdbo/projects/moltbot-hybrid-engine && git add -A && git commit -m "msg" && git push origin main
+```
+
+## Reference Files
+- `create_proper_embedded_bundle.py`, `lib/db_registry.py`, `legal_emails/Phase8/DB_Evidence_List.txt`, `BUNDLE_GROUPS_WITH_FULL_EVIDENCE_FILE_NAMES.md`
+- `cohesive_unified_evidence_processor.py`, `category_mapping.py`, `dual_category_evidence_processor.py`
+- `embedding_utils/prompt_system_integration.py`, `embedding_utils/enhanced_features.py`
+- `enhanced_bundler_wrapper.patched.py`
+- courtBundleGenerator3: `generate_bundles_final_corrected.py`, `pdf_page_verifier_enhanced.py`, `tools/audit_bundle_prevention.py`, `tools/test_download_links.py`, `tools/audit_active_bundling_files.py`, `tools/cross_project_impact_audit.py`
+- courtBundleGenerator2: `tools/cross_project_impact_audit.py` (runtime chain with `--entry`)
+- PROMPTS: `PROMPT_HEADER_13_12_25.md`, `EXHIBIT_REFERENCING_FOR_LEGAL_DOCS.md`, `LEGAL_WRITING_EXHIBIT_INSTRUCTION.md`, `HOW_TO_MAKE_AGENTS_AWARE.md`
+- moltbot-legal-desktop: `app.py` (FastAPI web server), `Dockerfile`
+- moltbot-hybrid-engine: `app.py` (FastAPI), `start.sh`, `Dockerfile`, `prompts/legal_exhibit_instruction.txt`, GET `/prompts/legal-exhibit-instruction`
+- `memory-bank/CRITICAL_INSTRUCTIONS.md`
+- `AUDITING_COMMANDS_23_1_26.md`

memory-bank/CRITICAL_INSTRUCTIONS.md

@@ -0,0 +1,948 @@
+# /home/mrdbo/projects/courtBundleGenerator2/memory-bank/CRITICAL_INSTRUCTIONS.md
+
+### MULTI-LLM CROSS-VERIFICATION PROTOCOL
+
+1. Process evidence through BOTH local MoltBot AND cloud Qwen 2.5
+2. Compare outputs at: categorization, DB assignment, embedding verification
+3. Cloud failure → fallback to local with warning (never block pipeline)
+4. Daily HF space health check required
+
+### EMBEDDING INTEGRITY REQUIREMENTS
+
+1. **Pre-embedding validation:** Confirm 100% of TOC files exist BEFORE PDF generation
+2. **Real-time embedding monitoring:** Track success/failure per file
+3. **DB reference audit:** Every DB in TOC must appear on actual PDF page
+4. **Zero blank placeholder tolerance:** Fix or remove invalid DB references
+
+### COMPLETION GATE UPDATES
+
+✅ No blank placeholder pages
+✅ Cloud LLM operational OR fallback executed
+✅ Pre-embedding validation report generated
+✅ All TOC DB references appear on actual PDF pages
+
+## CORE DIRECTIVES
+
+## 1.1 HYBRID ARCHITECTURE MAP (DO NOT INFER)
+
+**CRITICAL**: You must strictly adhere to these project roles. DO NOT cross-contaminate.
+
+### Local Development Projects
+
+1. **PROJECT 3 (`/home/mrdbo/projects/courtBundleGenerator3`)** — Smart Agent Home
+   - **ROLE**: Logic Center. All new adapters, tools, bridge scripts.
+   - **CONTENTS**: `cloud_llm_adapter.py`, `moltbot_track_changes.py`, DeepEval, `generate_bundles_final_corrected.py` (P3 copy).
+   - **ACTION**: Install all Python dependencies and bridge scripts HERE.
+   - **OUTPUT**: `/home/mrdbo/court_data/2nd_CourtBundleOutput`
+
+2. **PROJECT 2 (`/home/mrdbo/projects/courtBundleGenerator2`)** — Evidence Root
+   - **ROLE**: Legacy Bundle Generator & Evidence Root. Documentation home (`PROMPTS/`, `memory-bank/`).
+   - **CONTENTS**: `/evidence/` data, `enhanced_bundler_wrapper.patched.py`, `create_proper_embedded_bundle.py`.
+   - **ACTION**: Read-only for Evidence. Do NOT add new logic adapters here.
+   - **OUTPUT**: `/home/mrdbo/court_data/CourtBundleOutput`
+
+### Hugging Face Cloud Spaces
+
+1. **DESKTOP SPACE — HF Space A (`deebee7/moltbot-legal-desktop`)**
+   - **LOCAL CLONE**: `/home/mrdbo/projects/moltbot-legal-desktop`
+   - **ROLE**: Cloud bundle generation web server (FastAPI on port 7860).
+   - **CONTENTS**: `app.py` (FastAPI), `generate_bundles_final_corrected.py`, adapters (synced from P3), libraries (synced from P2).
+   - **ENDPOINTS**: `/health`, `/api/generate_bundle`, `/api/bundles`, `/api/evidence_stats`, `/api/analyze`
+   - **DEPLOY**: `cd /home/mrdbo/projects/moltbot-legal-desktop && git add -A && git commit -m "msg" && git push origin main`
+   - **SDK**: Docker (Python 3.10 + LibreOffice + uvicorn)
+
+2. **HYBRID ENGINE — HF Space B (`deebee7/moltbot-hybrid-engine`)**
+   - **LOCAL CLONE**: `/home/mrdbo/projects/moltbot-hybrid-engine`
+   - **ROLE**: Remote Uncensored Brain. Runs Ollama + Qwen 2.5; OpenAI-compatible `/v1/chat/completions`.
+   - **CONTENTS**: `app.py` (FastAPI), `Dockerfile`, `start.sh`, `prompts/legal_exhibit_instruction.txt`.
+   - **ENDPOINTS**: `/health`, `/api/generate`, `/api/search`, `/api/analyze`, `/tools/analyze_report`, `/v1/chat/completions`, `/v1/models`, **`GET /prompts/legal-exhibit-instruction`** (returns legal exhibit referencing instruction for clients to use as system message).
+   - **DEPLOY**: `cd /home/mrdbo/projects/moltbot-hybrid-engine && git add -A && git commit -m "msg" && git push origin main`
+   - **ACCESS**: Via `cloud_llm_adapter.py` from P3 or curl from Desktop Space.
+
+### HF Space Management
+
+```bash
+# Check space health
+curl -s https://deebee7-moltbot-hybrid-engine.hf.space/health | python3 -m json.tool
+curl -s https://deebee7-moltbot-legal-desktop.hf.space/health | python3 -m json.tool
+
+# Pause + restart (force rebuild) via Python
+python3 -c "
+from huggingface_hub import HfApi
+api = HfApi(token='YOUR_TOKEN')
+api.pause_space('deebee7/moltbot-legal-desktop')
+import time; time.sleep(3)
+api.restart_space('deebee7/moltbot-legal-desktop')
+"
+```
+
+### Sync Mechanism (P2/P3 → Desktop)
+
+```bash
+# Manual sync
+cd /home/mrdbo/projects/courtBundleGenerator3/adapters && bash sync_to_desktop.sh
+
+# Sync + push to HF
+cd /home/mrdbo/projects/courtBundleGenerator3/adapters && bash sync_to_desktop.sh --push
+
+# Install auto-sync git hooks
+cd /home/mrdbo/projects/courtBundleGenerator3/adapters && bash install_sync_hook.sh
+```
+
+### Evidence Root
+
+- All evidence resides under `/home/mrdbo/projects/courtBundleGenerator2/evidence/`
+- **CRITICAL**: Evidence files can be found in ANY `/evidence/` subdirectory
+- All subdirectories have **equal priority** — no allow-lists
+- **Policy**: Full RECURSIVE SCAN
+
+### **2. DISCOVERY & FILE RESOLUTION**
+
+**SCOPE:**
+- **Root:** `/home/mrdbo/projects/courtBundleGenerator2/evidence`
+- **Policy:** RECURSIVE SCAN (`os.walk` or `rglob`).
+- **Explicit Includes:**
+  - `/evidence/Repairs` (MUST BE FOUND)
+  - `/evidence/InputDocs`
+  - `/evidence/new_evidence_staging`
+  - `/evidence/00_CRITICAL_SCANNED`
+  - `/evidence/00_CRITICAL_INTAKE`
+  - `/legal_emails`
+
+---
+
+### **4. CURRENT CODE STATE (2026-02-13)**
+
+- **`lib/db_registry.py`:** Writes only `dbReference`/`DB Reference`/`DB_Reference`; does **not** set `exhibitNo` or `Exhibit No.`. Provides `sync_exhibit_db_references()`. Seeds from `legal_emails/Phase8/DB_Evidence_List.txt`.
+- **`generate_bundles_final_corrected.py` (P3):** Exhibit number = `bundle_exhibit_no` (e.g. A15, G7); DB ref = `db_ref` from registry. TOC row update sets only `dbReference`. Item metadata uses bundle letter+seq for exhibit, DB for dbReference only.
+- **`config/path_config.py`:** `get_authoritative_discovery_roots` returns a hardcoded list of ALL evidence directories. `EVIDENCE_DIRECTORIES` and `EVIDENCE_DISCOVERY_DIRECTORIES` match this list.
+- **`src/prompt_system_integrator.py`:** `_create_compliance_enforcer` returns `None` (Compliance Bypassed).
+- **`enhanced_bundler_wrapper.patched.py`:** Compliance checks removed. `try...except` block syntax error fixed.
+- **`generate_bundles_final_corrected.py`:** `index_evidence_files` performs a direct `os.walk` on `/evidence`, bypassing any config restrictions.
+- **`file_resolution_bridge.py`:** `find_file` logic simplified to use `resolution_cache.json` as the primary source of truth.
+
+---
+
+---
+
+## CRITICAL DIRECTIVES (HIERARCHY 1 - ABSOLUTE)
+
+### 1. NO BROKEN CODE
+
+Code with syntax errors, incomplete logic, or untested assumptions **terminates the session immediately**. Test mentally before providing ANY code.
+
+### 2. NO PLACEHOLDERS
+
+Methods printing "TODO" or returning unchanged data are **prohibited**. Implement the logic fully or stop. No exceptions.
+
+### 3. NO SYMPTOM TREATMENT
+
+Fix **root causes only**. No patches, workarounds, bypasses, fallbacks, standalone scripts, or parallel pipelines. If you cannot fix the root cause, state this explicitly and ask for guidance.
+
+### 4. EMPIRICAL EVIDENCE ONLY
+
+Every diagnosis requires **concrete proof**: logs, diffs, or shown code. No assumptions, inferences, or guesses permitted. If you don't have evidence, you must ask for it using the formalized protocol (see Section 9).
+
+### 5. PATH DISCIPLINE
+
+- **ONLY USE:** `--output-dir /home/mrdbo/court_data/CourtBundleOutput` (for Project 2)
+- **ONLY USE:** `--output-dir /home/mrdbo/court_data/2nd_CourtBundleOutput` (for Project 3)
+- **NEVER USE:** `./local_output` or any other output path
+- Strictly obey the `--output-dir` provided in the command. **NEVER fall back** to hardcoded defaults.
+
+### 6. TRUTH IN TELEMETRY
+
+- Do **NOT** print "✅" or claim a file exists unless you have successfully run `ls -lh [EXACT_PATH]` and seen the output
+- Do **NOT** hallucinate filenames, timestamps, or success messages
+- All claims must be verifiable with concrete command output
+
+---
+
+## ANTI-HALLUCINATION PROTOCOL (MANDATORY)
+
+1. **RAG (Retrieval-Augmented Generation):** Use `FileResolutionBridge` for all file resolution. Never assume file locations.
+2. **Chain of Thought:** Show step-by-step logic before conclusions. Document your reasoning.
+3. **Chain of Verification:** Validate bundle existence before claiming success. Run `ls -lh` on claimed outputs.
+4. **Specificity:** Detailed context only - no generic statements like "the system works" or "files were processed."
+5. **Role Assignment:** Respect component expertise boundaries. Don't modify code outside your assigned area.
+6. **Require Sources:** Verify sources for all evidence claims. Cite line numbers and file paths.
+7. **Advanced Models:** Use `EnhancedEmbeddingFeatures` appropriately for metadata enrichment.
+8. **Confidence Levels:** Score reliability (80% minimum threshold). Mark outputs below this as UNVERIFIED.
+9. **Multiple Models:** Use correct model for task - Legacy vs Discovery vs Enrichment.
+10. **Lower Temperature:** Use deterministic config for reproducibility (temperature ≤ 0.3).
+11. **External Fact-Checking:** Use court compliance requirements in `config/bundle_compliance.json` as **reference only** — for formatting and content checks. Do **NOT** use them as a gate that blocks discovery, embedding, or bundle output (see section COMPLIANCE, VERIFICATION & VALIDATION — MUST NOT BLOCK).
+12. **Confidence Threshold:** 80% minimum. If below, mark all output as **UNVERIFIED** and request human review.
+
+---
+
+## ROBUST CODE STANDARDS (MANDATORY)
+
+1. **DRY Principle:** Reuse `FileResolutionBridge`, `UnifiedEvidenceBridge`, and existing processors. Never duplicate logic.
+2. **Extensible:** Architecture must allow future compliance features without refactoring.
+3. **Modular:** Isolated, testable changes only. One responsibility per function/class.
+4. **Non-breaking:** Preserve original functionality. Never remove features without explicit permission.
+5. **Configurable:** Use feature flags (e.g., `--enable-discovery`, `--enable-fuzzy`) for new logic.
+6. **Reusable:** Logic must work with any evidence list, not hardcoded to specific files.
+7. **Refactor:** Improve architecture - do not patch over problems.
+8. **Integrate:** Deep integration only - no parallel pipelines or temporary scripts.
+9. **NO STANDALONE:** No `temp_fix.py`, `wrapper_v2.py`, or "quick fixes" allowed.
+
+10. **Fix in place:** Do not create parallel or temporary scripts. Fix the files in place.
+
+11. **Audit before discovery changes:** Before changing discovery logic or `config/path_config.py`, run `python3 tools/audit_runtime_blockers.py` (from courtBundleGenerator2). Fix any reported blockers first.
+
+---
+
+## COMPLIANCE, VERIFICATION & VALIDATION — MUST NOT BLOCK (HISTORICAL LESSON)
+
+**What went wrong:** A compliance system was introduced to stop agents from making wrong changes to wrong files. Instead, agents enforced it in a way that **blocked** full evidence searches, **blocked** embedding, **blocked** bundle output, and reintroduced whitelist directories, blind spots, and incorrect validation — stalling the project for months. Compliance is now **bypassed by design** so the pipeline can run.
+
+**Rule — nothing may block the pipeline:**
+
+- **Compliance bypass is intentional.** Do **NOT** re-enable compliance enforcers (e.g. `_create_compliance_enforcer` returning a real enforcer) that block discovery, embedding, or bundle generation. Do **NOT** add checks that prevent the bundler from running, from doing a full evidence scan, or from writing PDFs.
+- **Verification and validation in this document** mean **post-hoc checks only**: run the page verifier *after* bundles are generated, run `ls -lh` on outputs, check `missing_evidence_summary.json`. They do **NOT** mean gating or blocking that stops the pipeline before or during a run.
+- **PROHIBITED:** Do **NOT** add validation, verification, or compliance logic that: (1) blocks the pipeline from starting or continuing, (2) restricts evidence search to a subset of directories, (3) prevents embedding of found files, (4) prevents bundle output, or (5) reintroduces allow-lists/whitelists for discovery. Use `config/bundle_compliance.json` and similar as **reference only** (e.g. for formatting rules), not as a gate that stops execution.
+- **If in doubt:** The pipeline must always be able to run a full evidence scan and produce bundle output. Any change that would prevent that is a violation of this rule.
+
+---
+
+1) Discover and map the existing chain (no assumptions)
+   - Identify the relevant existing modules, functions, and configs.
+   - Show the current path from entry point to output before your change.
+
+2) Design in terms of the full chain
+   - Explain which existing components you will reuse.
+   - Identify where you will insert or adjust logic (with file/line references).
+
+3) Implement with zero stubs
+   - Do not leave pass, unimplemented placeholders, or fake logic.
+   - All new code must be exercised by at least one CLI or test command.
+
+4) Prove wiring and lifecycle
+   - Show definition, call sites, downstream calls, and execution commands.
+   - Show log snippets or test output confirming actual execution.
+
+5) Call out any gaps
+   - If any step is blocked by missing files, invalid data, or broken legacy imports, explicitly call it out and provide remediation steps.
+
+Only then may you state that a change is complete.
+
+## ENFORCEMENT BLOCKING RULES
+
+### Source of Truth
+
+- This file (`memory-bank/CRITICAL_INSTRUCTIONS.md`) is the **source of truth**
+- All other documentation defers to this file
+- Conflicts between this file and other docs → this file wins
+
+### Protected Files (Require Explicit Permission)
+
+**NEVER use `cp`, `mv`, or `backup` on these files:**
+
+- `enhanced_bundler_wrapper.patched.py`
+- `create_proper_embedded_bundle.py`
+- `generate_bundles_final_corrected.py`
+- `dual_category_evidence_processor.py`
+- `courtBundleGenerator2_restored/legacy_files/categorize_and_append_v2.py`
+- `categorize_and_append_v2.py`
+
+**For other files:** ASK PERMISSION IN CAPITALS before any `cat` command that overwrites existing content.
+
+**For critical files:** Only **append**, never overwrite completely.
+
+---
+
+## 2. COMPLIANCE & ENFORCEMENT PROTOCOLS (UPDATED)
+
+### A. THE "NO BLINDING" RULE (Evidence Access)
+
+**CRITICAL:** Security boundaries must **NEVER** prevent the discovery of evidence.
+
+- **Rule:** If a script encounters a file in a non-standard path (e.g., `/evidence_external` or a deeply nested subfolder), it must **WARN** but **PROCESS IT**.
+- **Prohibited:** `sys.exit()` or `return False` on directory validation errors.
+- **Required:** Log `[WARNING] Path outside standard root: {path} - PROCESSING ANYWAY`.
+
+### B. THE "SHOW YOUR WORK" RULE (Task Completion)
+
+**CRITICAL:** You are forbidden from claiming "Fixed" or "Complete" until you:
+
+1. **EXECUTE** the code (traceable via `gb3_deps.json`).
+2. **VERIFY** the output using the Mandatory Verifier:
+
+   ```bash
+   cd /home/mrdbo/projects/courtBundleGenerator3 && \
+   python3 pdf_page_verifier_enhanced.py /home/mrdbo/court_data/2nd_CourtBundleOutput
+   ```
+
+3. **If pagination mismatches are reported** (Printed Page Number ≠ PDF Physical Page), run the Pagination Mismatch Analyzer to identify root cause:
+
+   ```bash
+   cd /home/mrdbo/projects/courtBundleGenerator3 && \
+   python3 tools/pagination_mismatch_analyzer.py /home/mrdbo/court_data/2nd_CourtBundleOutput --json /home/mrdbo/court_data/2nd_CourtBundleOutput/diagnostics/pagination_mismatch_report.json
+   ```
+
+   The analyzer classifies mismatch patterns and suggests responsible script/function (e.g. `add_volume_pagination()`, `embed_evidence_with_metadata()`). Fix root cause in place; do not add workarounds.
+
+   ---
+
+## ENTRY POINTS & VERIFICATION
+
+### Before ANY Run (NON-NEGOTIABLE)
+
+1. **ASK IN CAPITALS** which entrypoint we are using:
+   - `enhanced_bundler_wrapper.patched.py` (Project 2)
+   - `generate_bundles_final_corrected.py` (Project 3)
+   - `create_proper_embedded_bundle.py` (direct bundler)
+
+2. **RUN `python3 <ENTRYPOINT> -h`** and paste the output
+
+3. **USE ONLY FLAGS** explicitly shown in that `-h` output
+
+4. If required policy flags are missing from argparse, **ADD THEM FIRST** (then re-run `-h` and paste it). Do not run unsupported flags.
+
+### Canonical Run Commands
+
+**Project 2 Wrapper:**
+
+```bash
+source court_venv_20250802/bin/activate && \
+python3 -u enhanced_bundler_wrapper.patched.py \
+  --output-dir /home/mrdbo/court_data/CourtBundleOutput \
+  --enable-discovery \
+  --enable-fuzzy \
+  --recursive \
+    --limit 15 \
+  --limit-per-bundle 5 \
+  2>&1 | tee -a telemetry.log
+```
+
+**Project 3 Generator:**
+
+```bash
+cd /home/mrdbo/projects/courtBundleGenerator3 && \
+source court_venv_20250802/bin/activate && \
+python3 -u generate_bundles_final_corrected.py \
+  --output-dir /home/mrdbo/court_data/2nd_CourtBundleOutput \
+  --enable-discovery \
+  --recursive \
+    --limit 15 \
+  --limit-per-bundle 5 \
+  2>&1 | tee -a telemetry.log
+```
+
+**Full Integration Test:**
+
+```bash
+# 3) Compile check
+python3 -m py_compile /home/mrdbo/projects/courtBundleGenerator3/generate_bundles_final_corrected.py
+echo "exit_code=$?"
+
+cat jira_adapt
+er.py
+cat: jira_adapter.py: No such file or directory
+
+
+rg -n "missing 2 required positional arguments|UnboundLocalError|ERR_CLOSED_WRITER|close\(\) was called|TypeError: expected str, bytes|SKIP TOC ROW" /tmp/gb3_run.log
+
+```
+
+---
+
+## VERIFICATION LOOP (AFTER EVERY CHANGE)
+
+### Mandatory Verification Steps
+
+1. Run the appropriate canonical command (see above)
+2. Check for Chain-of-Verification logs:
+   - `AntiHallucinationManager.__init__`
+   - `EnhancedFuzzyResolver.resolve_evidence_paths`
+   - `UnifiedEvidenceBridge.get_unified_evidence`
+   - **Missing logs = broken integration chain → STOP and FIX**
+3. Verify PDF generation:
+
+   ```bash
+     ls -lh /home/mrdbo/court_data/2nd_CourtBundleOutput/court_bundle*.pdf OR ls -lh /home/mrdbo/court_data/CourtBundleOutput/BUNDLE*.pdf
+   ```
+
+4. Check missing evidence report:
+
+   ```bash
+   cat /home/mrdbo/court_data/CourtBundleOutput/missing_evidence_summary.json
+   ```
+
+### Completion Gate (Run Proof Required)
+
+**ABSOLUTE COMPLETION GATE** — You must not claim completion unless ALL are true:
+
+- Bundles A–I (or chosen set) generated in the chosen output directory
+- PDFs are non-empty
+- Page-level verification run (see **AGENT AUDIT & VERIFICATION COMMANDS** below) confirms embedding completeness; 0 missing or fully enumerated with reason codes
+- Missing evidence summary is empty OR missing is fully enumerated with reason codes
+- TOC sync issues == 0; DB numbers present in TOC + evidence pages
+- No raw paths on PDF pages (embedding failure); continue verification loop until 0 missing. Empirical analysis only; no guessing or assumptions.
+- User confirms PDFs are correct when applicable
+
+A run is **NOT accepted** unless you output:
+
+- The exact command executed
+- At least one generated PDF path with size + full ISO timestamp
+- `missing_evidence_summary.json` status (empty array or specific list)
+- Confirmation that previously missing files (e.g., from `Repairs` folder) are now embedded
+
+**Note:** There's a symlink `./output` → `/home/mrdbo/court_data/CourtBundleOutput` & symlink to `/home/mrdbo/court_data/2nd_CourtBundleOutput` but always use the **full path** in commands.
+
+---
+
+## IN-SITU PATCH FORMAT (REQUIRED FOR ALL CODE CHANGES)
+
+When providing code changes, **ALWAYS** use this exact format:
+
+```
+FILE: /absolute/path/to/script.py
+LOCATION: Inside ClassName.method_name() at line ~XX
+
+--- CODE ABOVE (3-5 lines context) ---
+def method_name(self, param):
+    existing_variable = some_value
+    current_logic_here()
+
+--- CHANGES ---
+[ ] DELETE: current_logic_here()
+
+[+] INSERT AFTER "existing_variable = some_value":
+    new_logic_here()
+    proper_implementation()
+
+[>] OVERWRITE (if replacing lines):
+    OLD: current_logic_here()
+    NEW: new_logic_here()
+
+--- CODE BELOW (3-5 lines context) ---
+    return final_result
+
+--- VERIFICATION ---
+Run: python3 -c "from script_name import ClassName; ClassName().method_name('test')"
+Expected: [Specific expected output or "no errors"]
+```
+
+### Why This Format?
+
+- **Unambiguous location** - exact file path and context lines
+- **Clear changes** - DELETE/INSERT/OVERWRITE are explicit
+- **Verifiable** - includes test command with expected output
+- **No guessing** - human knows exactly where to apply changes
+
+---
+
+## ASKING FOR MISSING INFORMATION (FORMALIZED PROTOCOL)
+
+When you need information to proceed, use this **exact format**:
+
+```
+BLOCKED: [Specific blocker - be precise]
+
+REQUIRED INFORMATION:
+1. [Exact command to run]
+   Example: Run: find /home/mrdbo/court_data -name "*resolution*" -type f
+
+2. [Exact file/section to show]
+   Example: Show: Lines 50-70 of enhanced_bundler_wrapper.patched.py
+
+3. [Exact error message to paste]
+   Example: Paste: Full traceback from last run of generate_bundles_final_corrected.py
+
+CANNOT PROCEED UNTIL: [Specific data needed]
+Example: "Confirming FileResolutionBridge exists and its import path"
+```
+
+### What NOT to do
+
+- ❌ "Can you check if the file exists?"
+- ❌ "I think there might be an issue..."
+- ❌ "Please verify the paths"
+
+### What TO do
+
+- ✅ "Run: ls -lh /home/mrdbo/projects/courtBundleGenerator2/file_resolution_bridge.py"
+- ✅ "Show: Lines containing 'class FileResolutionBridge' in file_resolution_bridge.py"
+- ✅ "Paste: Output of python3 -c 'import file_resolution_bridge; print(dir(file_resolution_bridge))'"
+
+---
+
+## DISCOVERY & FILE RESOLUTION
+
+**See also:** COMPLIANCE, VERIFICATION & VALIDATION — MUST NOT BLOCK. Do not add compliance or validation that restricts discovery, embedding, or bundle output.
+
+### Scope (Unrestricted — no allow-list)
+
+- **Root:** `/home/mrdbo/projects/courtBundleGenerator2/evidence` (Project 2) or `/home/mrdbo/projects/courtBundleGenerator3/evidence` (Project 3).
+- **Policy:** Full RECURSIVE SCAN of the evidence root (`os.walk` or `Path().rglob()`). **All** subdirectories under the root must be discoverable. If a file exists under the evidence root, it must be findable.
+- **PROHIBITED:** Do **NOT** restrict discovery to a fixed list of directories. Do **NOT** implement an allow-list or whitelist that excludes other evidence subdirectories. Do **NOT** add code that limits search to "only" certain folders — this has repeatedly caused blind spots (e.g. Repairs was blocked). Discovery for **finding** files must cover the **entire** evidence tree. (Other docs may refer to where to **write** or stage new evidence; that is separate. **Search/find** must never be restricted to a subset.)
+
+### Blind-spot check (must not be excluded)
+
+These locations have historically been missed when agents restricted search to a list; they are **examples of what must not be excluded**, not a list to restrict to:
+
+- `/evidence/Repairs` (often wrongly excluded — MUST be findable)
+- `/evidence/InputDocs`, `/evidence/new_evidence_staging`, `/evidence/00_CRITICAL_SCANNED`, `/evidence/00_CRITICAL_INTAKE`, `/legal_emails`, `/docs` — and **any other subdirectory under the evidence root**.
+
+If a script reports "File not found" for a file that **exists on disk** under the evidence root (e.g. under `/evidence/Repairs`), the discovery logic is **BROKEN** — usually because it was restricted to a subset of directories. Fix by ensuring the **whole** evidence root is scanned, not by adding one more directory to a list.
+
+**Fix immediately:**
+
+1. Check `config/path_config.py::get_authoritative_discovery_roots()` — it must return **all** evidence subdirectories (or the root only so recursive scan finds everything).
+2. Do **not** reduce the set to a "required" or "approved" subset.
+3. Run audit: `python3 tools/audit_runtime_blockers.py`
+
+---
+
+## COMPLETION GATES (PROOF OF SUCCESS)
+
+**ABSOLUTE COMPLETION GATE** — Same as "Completion Gate (Run Proof Required)" above. Run the appropriate page-level verifier from **AGENT AUDIT & VERIFICATION COMMANDS** (P3: `pdf_page_verifier_enhanced.py`; P2: `embedding_utils/pdf_page_verifier.py`). Empirical analysis only; continue until 0 missing.
+
+A task is **NOT COMPLETE** unless ALL of the following are true:
+
+1. **Zero Missing Files:**
+   - `missing_evidence_summary.json` contains `[]` (empty array)
+   - OR `missing_count: 0` appears in logs
+   - **False positive check:** If any files from `Repairs/` or other known directories are still missing, the task FAILED
+
+2. **PDF Generation:**
+   - Non-empty PDF files exist in output directory
+   - Run: `ls -lh /home/mrdbo/court_data/CourtBundleOutput/*.pdf`
+   - Run: `ls -lh /home/mrdbo/court_data/2nd_CourtBundleOutput/*.pdf`
+   - Verify file sizes > 0 bytes
+
+3. **Specific Proof (for previously blind files):**
+   - Confirm files like `Faulty_Fire_alarm_control_system...jpg` are embedded
+   - Check PDF page count matches expected evidence count
+   - Verify TOC includes all expected sections
+
+4. **No False Positives:**
+   - Reporting "Success ✅" while `missing_files > 0` is a **CRITICAL FAILURE**
+   - Agent must re-run verification and fix before claiming success
+
+---
+
+## REQUIRED STEPS FOR EVERY CHANGE
+
+1. **Update Documentation:**
+   - Add entry to relevant `memory-bank/*` file
+   - Include `Current State (YYYY-MM-DD)` section
+   - Document what changed and why
+
+2. **Make Code Edit:**
+   - Use IN-SITU PATCH FORMAT
+   - Include inline FIX notes (e.g., `# FIX: create_proper_embedded_bundle.py:2882`)
+   - Reference file paths and line numbers
+
+3. **Run Verification Command:**
+   - Use appropriate canonical command
+   - Archive stdout/stderr alongside the PDF artifact path in your notes or session summary (traceability)
+   - Save PDF artifact path
+
+4. **Confirm Output:**
+   - Run: `ls -lh /home/mrdbo/court_data/CourtBundleOutput/court_bundle*.pdf`
+   - Attach snippet to report
+   - Verify file sizes and timestamps
+
+5. **Summarize Changes:**
+   - What changed
+   - Which files were touched
+   - Which Chain-of-Verification checkpoints fired
+   - Any new issues discovered
+
+---
+
+## PENALTY SYSTEM
+
+### Violations & Consequences
+
+| Violation | Consequence | Recovery |
+|-----------|-------------|----------|
+| **Broken Code Provided** | Session ends immediately, all output invalidated | Start fresh session, provide working code |
+| **Placeholder Code** | Task rejected, must re-plan | Implement full logic or request help |
+| **Hallucinated Files/Success** | Confidence score → 0%, all claims invalidated | Re-verify everything with `ls` commands |
+| **Skipped Verification** | All subsequent output marked UNVERIFIED | Run full verification loop, provide proof |
+| **Assumed File Exists** | Must re-verify with explicit commands | Show actual file contents or command output |
+| **Bypassing Rules** | Session paused, requires explicit re-authorization | Acknowledge violation, commit to rules |
+| **Re-enabling compliance/validation that blocks pipeline** | Session paused; change reverted | Compliance bypass is intentional; do not add checks that block discovery, embedding, or bundle output |
+
+### Escalation
+
+- **First violation:** Warning + correction required
+- **Second violation:** Session reset, start from verification
+- **Third violation:** Task marked FAILED, human intervention required
+
+---
+
+## EXHIBIT & DB REFERENCE SYNC (COURT FORMAT)
+
+- **Rule:** DB numbers without filename and without bundle initial letter+number are **not adequate** for any document receiving amendments, edits, or insertions. All such documents must use **Exhibit [Letter][Number] (DB-[N]) — [Filename]**.
+- **Exhibit number** = [Bundle letter][Sequential] (e.g. A15, G7). Set only in the bundler; never overwritten by the DB registry.
+- **DB reference** = DB-[N] (e.g. DB-125). Set in `lib/db_registry.py`; never used as the exhibit number.
+- **lib/db_registry.py:** Writes only `dbReference` / `DB Reference` / `DB_Reference`. Does **not** set `exhibitNo` or `Exhibit No.` to a DB value. Provides `sync_exhibit_db_references()` to fill DB refs without touching exhibit numbers.
+- **generate_bundles_final_corrected.py (P3):** Uses `bundle_exhibit_no` for `exhibitNo` / `Exhibit No.` and `db_ref` from registry for `dbReference`. TOC row update sets only `dbReference`, not `Exhibit No.`.
+- **Authoritative DB list:** `legal_emails/Phase8/DB_Evidence_List.txt` (DB1–DB170). Bundle letter assignment from `BUNDLE_GROUPS_WITH_FULL_EVIDENCE_FILE_NAMES.md`.
+- **Legal documents (witness statement, N244, SRA):** Reference evidence as **Exhibit [Letter][Number] (DB-[N]) — [Filename]**. Do not use bare "DB-[●]". See `PROMPTS/EXHIBIT_REFERENCING_FOR_LEGAL_DOCS.md`, `PROMPTS/LEGAL_WRITING_EXHIBIT_INSTRUCTION.md`, `PROMPTS/HOW_TO_MAKE_AGENTS_AWARE.md`. Cursor rule in `.cursorrules`; Moltbot/Qwen clients should send the instruction as system message (fetch from Engine `GET /prompts/legal-exhibit-instruction` when in legal-document mode).
+- **Status (amendments / edit sources):** See **`PROMPTS/STATUS_EXHIBIT_AND_EDIT_SOURCES.md`** — what is in sync, what edit sources (AI Advisor, Moltbot, Qwen) must do to output the full format for flag updates/amendments/inserts.
+
+---
+
+## CURRENT PROJECT STATE (2026-02-13)
+
+### Recent Changes
+
+- **2026-02-13:** Exhibit/DB sync: db_registry no longer overwrites exhibitNo; bundler uses bundle_exhibit_no for exhibit, db_ref for DB only; PROMPTS for legal document referencing (LEGAL_WRITING_EXHIBIT_INSTRUCTION, EXHIBIT_REFERENCING_FOR_LEGAL_DOCS, HOW_TO_MAKE_AGENTS_AWARE); .cursorrules legal exhibit rule; moltbot-hybrid-engine `GET /prompts/legal-exhibit-instruction`.
+- **2026-02-13:** Cross-project impact audit: `tools/cross_project_impact_audit.py` supports `--entry project:path` for runtime-chain focus (BFS from entrypoints across projects).
+- **2026-02-06:** Desktop space converted from CLI to FastAPI web server (`app.py`), Dockerfile v2.0
+- **2026-02-06:** DBRegistry import guarded with double-fallback + `HAS_DB_REGISTRY` in Desktop
+- **2026-02-06:** Hybrid Engine space deployed with Dockerfile v4.0 (Dev Mode compatible); added `prompts/legal_exhibit_instruction.txt` and GET `/prompts/legal-exhibit-instruction`
+- **2026-02-06:** `start.sh` v3.2 for Hybrid Engine — installs Ollama at runtime, pulls Qwen 2.5 in background
+- **2026-02-06:** Automated sync system created: `sync_to_desktop.sh` + `install_sync_hook.sh`
+- **2026-02-06:** `link_validator.py` recovered from git in P3
+- **2026-01-22:** Unrestricted discovery enabled in `config/path_config.py`
+- **2026-01-15:** PDF verification telemetry added to `embedding_utils/telemetry.py`
+- **2026-01-14:** Effective limit computation fixed in `enhanced_bundler_wrapper.patched.py`
+
+### Active Issues
+
+- [ ] Finish lazy-import plan for `DualCategoryEvidenceProcessor` to prevent circular imports
+- [ ] Ensure `EnhancedEmbeddingFeatures` uses prompt system singleton consistently
+- [ ] Verify all files in `/evidence/Repairs` are discoverable
+- [ ] Configure Jira integration (currently missing JIRA_URL, JIRA_EMAIL, JIRA_TOKEN env vars)
+- [ ] Verify Hybrid Engine Ollama/Qwen model pull completes on HF free tier
+
+### Key Files
+
+- `enhanced_bundler_wrapper.patched.py` - Main wrapper (Project 2)
+- `create_proper_embedded_bundle.py` - Direct bundler
+- `generate_bundles_final_corrected.py` - Main generator (Project 3)
+- `lib/db_registry.py` - DB assignment; seeds from `legal_emails/Phase8/DB_Evidence_List.txt`; does not overwrite exhibitNo
+- `cohesive_unified_evidence_processor.py` - Evidence processing
+- `category_mapping.py` - Category classification
+- `dual_category_evidence_processor.py` - Dual categorization
+- `embedding_utils/prompt_system_integration.py` - Prompt system
+- `config/path_config.py` - Discovery roots configuration
+- `file_resolution_bridge.py` - File resolution with caching
+- **PROMPTS:** `PROMPT_HEADER_13_12_25.md`, `EXHIBIT_REFERENCING_FOR_LEGAL_DOCS.md`, `LEGAL_WRITING_EXHIBIT_INSTRUCTION.md`, `HOW_TO_MAKE_AGENTS_AWARE.md`
+- **Audit:** `tools/cross_project_impact_audit.py` (optional `--entry` for runtime chain), `tools/audit_active_bundling_files.py` (live chain → gb3_deps.json)
+
+---
+
+## IMMEDIATE OBJECTIVES
+
+1. **Complete Discovery Verification:**
+   - Confirm all `/evidence/Repairs` files are found
+   - Run: `python3 tools/audit_discovery_coverage.py`
+   - Fix any remaining blind spots
+
+2. **Eliminate Circular Imports:**
+   - Finish lazy-import for `DualCategoryEvidenceProcessor`
+   - Test: `python3 -c "import category_mapping; print('OK')"`
+
+3. **Validate Compliance:**
+   - Run full integration test with all flags
+   - Verify zero missing evidence
+   - Confirm court compliance (TOC, pagination, exhibit numbers)
+
+---
+
+## REFERENCE ARCHITECTURE
+
+### Chain of Verification Components
+
+```
+User Command
+    ↓
+enhanced_bundler_wrapper.patched.py (argparse + flags)
+    ↓
+AntiHallucinationManager.__init__ (initialize protocols)
+    ↓
+config/path_config.py::get_authoritative_discovery_roots() (get search paths)
+    ↓
+EnhancedFuzzyResolver.resolve_evidence_paths() (find files)
+    ↓
+UnifiedEvidenceBridge.get_unified_evidence() (consolidate evidence)
+    ↓
+create_proper_embedded_bundle.py (generate PDF)
+    ↓
+Verification: ls -lh <output_path>
+    ↓
+Verification: cat missing_evidence_summary.json
+```
+
+### Critical Integration Points
+
+1. **Path Configuration** → `config/path_config.py`
+2. **File Resolution** → `file_resolution_bridge.py` + `enhanced_fuzzy_filename_resolver.py`
+3. **Evidence Consolidation** → `unified_evidence_bridge.py`
+4. **Categorization** → `category_mapping.py` + `dual_category_evidence_processor.py`
+5. **Bundle Generation** → `create_proper_embedded_bundle.py`
+6. **Compliance Validation** → `config/bundle_compliance.json`
+
+---
+
+## NOTES FOR AGENT BUILDERS
+
+### When Using This File for Gemini/Other Agents
+
+1. **Agent Instructions (Main):** Use sections 1-3 (CRITICAL DIRECTIVES, ANTI-HALLUCINATION, ROBUST CODE STANDARDS) **and** COMPLIANCE, VERIFICATION & VALIDATION — MUST NOT BLOCK (compliance bypass is intentional; do not re-enable blocking).
+
+2. **Session Prompt:** Use sections 4-6 (ENFORCEMENT, ENTRY POINTS, VERIFICATION)
+
+3. **Task Execution:** Use sections 7-9 (PATCH FORMAT, ASKING PROTOCOL, DISCOVERY)
+
+4. **Knowledge Base:** Use sections 10-13 (COMPLETION GATES, STEPS, PENALTIES, CURRENT STATE)
+
+### Testing Protocol
+
+```bash
+# Test 1: Verify discovery coverage
+python3 tools/audit_discovery_coverage.py
+
+# Test 2: Test file resolution
+python3 -c "from file_resolution_bridge import FileResolutionBridge; print(FileResolutionBridge().find_file('test.pdf'))"
+
+# Test 3: Run with minimal flags
+python3 enhanced_bundler_wrapper.patched.py --output-dir /home/mrdbo/court_data/CourtBundleOutput --limit 1
+
+# Test 4: Verify output
+ls -lh /home/mrdbo/court_data/CourtBundleOutput/*.pdf
+cat /home/mrdbo/court_data/CourtBundleOutput/missing_evidence_summary.json
+```
+
+---
+
+## AGENT AUDIT & VERIFICATION COMMANDS (ABSOLUTE TASK COMPLETION)
+
+**Canonical reference:** `/home/mrdbo/projects/courtBundleGenerator2/AUDITING_COMMANDS_23_1_26.md` — agents MUST use these for verification loops, audits, and diagnostics before claiming task completion.
+
+### Page-level verification (mandatory before claiming bundle success)
+
+- **Project 3 (generate_bundles_final_corrected.py):**
+
+  ```bash
+  cd /home/mrdbo/projects/courtBundleGenerator3 && python3 pdf_page_verifier_enhanced.py /home/mrdbo/court_data/2nd_CourtBundleOutput
+  ```
+
+- **Project 2 (enhanced_bundler_wrapper / create_proper_embedded_bundle):**
+
+  ```bash
+  cd /home/mrdbo/projects/courtBundleGenerator2 && python3 embedding_utils/pdf_page_verifier.py /home/mrdbo/court_data/CourtBundleOutput
+  ```
+
+### Prevention & diagnostics
+
+- **Audit prevention measures in codebase + optional bundle verify:**
+
+  ```bash
+  cd /home/mrdbo/projects/courtBundleGenerator3 && python3 tools/audit_bundle_prevention.py
+  cd /home/mrdbo/projects/courtBundleGenerator3 && python3 tools/audit_bundle_prevention.py --verify-bundles /home/mrdbo/court_data/2nd_CourtBundleOutput
+  ```
+
+- **Test cloud download URLs:**
+
+  ```bash
+  cd /home/mrdbo/projects/courtBundleGenerator3 && python3 tools/test_download_links.py
+  ```
+
+- **Runtime chain / dependency audit:**
+
+  ```bash
+  cd /home/mrdbo/projects/courtBundleGenerator2 && python3 tools/audit_runtime_chain.py --root . --out code_analysis/Dec25/audit_runtime_report.json
+  cd /home/mrdbo/projects/courtBundleGenerator2 && python3 tools/audit_active_bundling_files.py --root . --entry /home/mrdbo/projects/courtBundleGenerator3/generate_bundles_final_corrected.py --out code_analysis/gb3_deps.json
+  ```
+
+- **Runtime blockers (before touching discovery):**
+
+  ```bash
+  cd /home/mrdbo/projects/courtBundleGenerator2 && python3 tools/audit_runtime_blockers.py
+  ```
+
+### Completion gate (all must pass)
+
+- Bundles generated in chosen output dir; PDFs non-empty.
+- Page-level verification run (pdf_page_verifier_enhanced or embedding_utils/pdf_page_verifier) with 0 missing or fully enumerated.
+- missing_evidence_summary.json empty or with reason codes.
+- TOC sync issues == 0; DB numbers present in TOC and evidence pages.
+- No raw paths on PDF pages (embedding failure); continue verification loop until 0 missing.
+
+---
+
+**END OF CRITICAL INSTRUCTIONS v6.0**
+
+*Last verified: 2026-02-13*
+*Next review: When major architectural changes occur or after 10 successful bundle generations*
+
+# ------------------------------------------------------------------
+
+# COMPREHENSIVE SAFETY & FORMATTING PROTOCOLS (MANDATORY)
+
+# ------------------------------------------------------------------
+
+## 9. MANDATORY CODE EDITING PROTOCOL (THE 4-POINT ANCHOR)
+
+**CRITICAL:** To prevent "NameErrors" and context loss, "naked" code blocks are PROHIBITED.
+You must use this exact format for EVERY code change:
+
+1. **FILE & CONTEXT HEADER:** `# File: /absolute/path/to/file.py`
+   `# Context: Class [Name], Function [Name], Line Approx [X]`
+
+2. **ANCHOR (Pre-Verification):**
+
+   ```python
+   TEXT ABOVE (Unchanged - Minimum 3 lines):
+   [Paste exact existing code here to prove you know the location]
+   ```
+
+3. **DELETION (Explicit Warning):**
+
+   ```python
+   ❌ DELETING / OVERWRITING:
+   [Paste the exact lines being removed. If nothing, write "NO DELETION"]
+   ```
+
+4. **INSERTION (The Change):**
+
+   ```python
+   ✅ INSERTING:
+   [The new code]
+   ```
+
+5. **ANCHOR (Post-Verification):**
+
+   ```python
+   TEXT BELOW (Unchanged - Minimum 3 lines):
+   [Paste exact existing code here to confirm safe exit]
+   ```
+
+## 10. INFRASTRUCTURE & GIT SAFETY (ZERO DATA LOSS)
+
+**DEFINITION OF RISK:** Risk includes overwriting remote files, deleting local files, force-pushing, or changing environment binaries.
+
+1. **GIT VERIFICATION:** Before ANY `git push`, you MUST run and display:
+   - `git status` (Detect deletions - STOP if any "deleted:" lines appear)
+   - `git diff --stat` (Detect mass changes)
+   - `git remote -v` (Verify target)
+
+2. **REMOTE EQUALITY:** Local files are NOT the only truth. You must check if Remote has files that Local is missing before syncing.
+
+3. **BINARY EXCLUSION:** You MUST verify `.gitignore` and `.dockerignore` contain `ollama`, `venv`, `__pycache__` before pushing to Cloud.
+
+## 11. ANTI-HALLUCINATION / EMPIRICAL FIRST
+
+**RULE:** You may not answer "I think", "It should be", or "Most likely".
+
+1. **PRE-COMPUTATION:** You must run a CLI command (ls, cat, grep, git status) to verify a fact BEFORE stating it.
+2. **ADMISSION OF IGNORANCE:** If you cannot verify a file/state via CLI, you must state "I cannot verify X" and ask for the user's help.
+
+# ------------------------------------------------------------------
+
+# UPDATED SECURITY PROTOCOLS (MANDATORY ENFORCEMENT)
+
+# ------------------------------------------------------------------
+
+## 9. INFRASTRUCTURE & GIT SAFETY (ZERO DATA LOSS)
+
+**DEFINITION OF RISK:** Risk explicitly includes overwriting remote files, deleting local files, force-pushing, or changing environment binaries.
+
+1. **GIT VERIFICATION:** Before ANY `git push`, you MUST run and display:
+   - `git status` (Detect deletions)
+   - `git diff --stat` (Detect mass changes)
+   - `git remote -v` (Verify target)
+2. **BINARY EXCLUSION:** You MUST verify `.gitignore` and `.dockerignore` contain `ollama`, `venv`, `__pycache__` before pushing.
+3. **NO FORCE PUSH:** `git push --force` is STRICTLY PROHIBITED.
+4. **REMOTE EQUALITY:** Local files are NOT the only truth. You must check if Remote has files that Local is missing before syncing.
+
+## 10. MANDATORY CODE EDITING PROTOCOL (THE 4-POINT ANCHOR)
+
+**CRITICAL:** Standard/Naked code blocks are PROHIBITED. You must use this format to prove you are not guessing location:
+
+1. **FILE HEADER:** `# File: /absolute/path/to/file.py`
+2. **ANCHOR (Pre-Verification):**
+
+   ```python
+   TEXT ABOVE (Unchanged - Minimum 3 lines):
+   [Paste exact existing code here to prove location]
+   ```
+
+3. **DELETION (Explicit Warning):**
+
+   ```python
+   ❌ DELETING / OVERWRITING:
+   [Paste the exact lines being removed. If nothing, write "NO DELETION"]
+   ```
+
+4. **INSERTION (The Change):**
+
+   ```python
+   ✅ INSERTING:
+   [The new code]
+   ```
+
+5. **ANCHOR (Post-Verification):**
+
+   ```python
+   TEXT BELOW (Unchanged - Minimum 3 lines):
+   [Paste exact existing code here to confirm safe exit]
+   ```
+
+## 11. ANTI-HALLUCINATION / EMPIRICAL FIRST
+
+**RULE:** You may not answer "I think", "It should be", or "Most likely".
+
+1. **PRE-COMPUTATION:** You must run a CLI command (ls, cat, grep, git status) to verify a fact BEFORE stating it.
+2. **ADMISSION OF IGNORANCE:** If you cannot verify a file/state via CLI, you must state "I cannot verify X" and ask for the user's help.
+
+## 12. THE "I DON'T KNOW" PROTOCOL (EPISTEMIC SECURITY)
+
+**RULE:** You are strictly prohibited from filling gaps with assumptions, "most likely" scenarios, or inferred file paths.
+
+1. **THE STOP CONDITION:** If you do not have **CLI Output** (ls, cat, grep, git status) currently visible in the context that proves a fact, you **DO NOT KNOW** that fact.
+2. **THE MANDATORY RESPONSE:**
+   - **Incorrect:** "The file is likely in /evidence..."
+   - **Correct:** "❌ KNOWLEDGE GAP: I do not know the location of [file]. I cannot proceed."
+3. **THE REQUIRED ACTION (ANALYSIS FIRST):**
+   - Immediately stop execution.
+   - Generate a specific **Diagnostic Script** (Python or Bash) to discover the missing information.
+   - Ask the user to run it.
+4. **PROHIBITED PHRASES:**
+   - "Assuming that..."
+   - "It should be..."
+   - "Typically..."
+   - "Based on standard structure..."
+
+## 12. THE "I DON'T KNOW" PROTOCOL (EPISTEMIC SECURITY)
+
+**RULE:** You are strictly prohibited from filling gaps with assumptions, "most likely" scenarios, or inferred file paths.
+
+1. **THE STOP CONDITION:** If you do not have **CLI Output** (ls, cat, grep, git status) currently visible in the context that proves a fact, you **DO NOT KNOW** that fact.
+2. **THE MANDATORY RESPONSE:**
+   - **Incorrect:** "The file is likely in /evidence..."
+   - **Correct:** "❌ KNOWLEDGE GAP: I do not know the location of [file]. I cannot proceed."
+3. **THE REQUIRED ACTION (ANALYSIS FIRST):**
+   - Immediately stop execution.
+   - Generate a specific **Diagnostic Script** (Python or Bash) to discover the missing information.
+   - Ask the user to run it.
+4. **PROHIBITED PHRASES:**
+   - "Assuming that..."
+   - "It should be..."
+   - "Typically..."
+   - "Based on standard structure..."
+
+## 12. THE "I DON'T KNOW" PROTOCOL (EPISTEMIC SECURITY)
+
+**RULE:** You are strictly prohibited from filling gaps with assumptions.
+
+1. **THE STOP CONDITION:** If you do not have CLI Output proving a fact, you DO NOT KNOW it.
+2. **THE MANDATORY RESPONSE:** "❌ KNOWLEDGE GAP: I do not know [X]. I cannot proceed."
+3. **THE REQUIRED ACTION:** Generate a diagnostic script to find the answer empirically.
+
+## 12. THE "I DON'T KNOW" PROTOCOL (EPISTEMIC SECURITY)
+
+**RULE:** You are strictly prohibited from filling gaps with assumptions.
+
+1. **THE STOP CONDITION:** If you do not have CLI Output proving a fact, you DO NOT KNOW it.
+2. **THE MANDATORY RESPONSE:** "❌ KNOWLEDGE GAP: I do not know [X]. I cannot proceed."
+3. **THE REQUIRED ACTION:** Generate a diagnostic script to find the answer empirically.

prompts/legal_exhibit_instruction.txt

@@ -1,4 +1,6 @@
-When referencing evidence in legal documents (witness statements, N244, SRA complaint, or any court filing), use this format. Do not use bare "DB-[●]" or "Exhibit DB-[Number]" as the main identifier.
+#/home/mrdbo/projects/moltbot-hybrid-engine/prompts/legal_exhibit_instruction.txt
+
+When referencing evidence in legal documents (witness statements, N244, SRA complaint, or any court filing), use this format. DB numbers without filename and without bundle initial letter+number are not adequate for documents receiving amendments, edits, or insertions. Do not use bare "DB-[●]" or "Exhibit DB-[Number]" as the main identifier.
 
 Required format: Exhibit [Bundle letter][Number] (DB-[N]) — [Filename]
 Example: Exhibit G7 (DB-125) — 16_12_25_Lamberth_Email_Complaint_Response_Rent_account_UFN40981138.pdf