chore: remove accidentally committed file

NICK_REVIEW_OUTSTANDING.md

@@ -1,144 +0,0 @@
-# Nick's Review - Outstanding Items for Discussion
-
-**PR #11699 - Dynamic LLM Registry**
-
----
-
-## 🔴 CRITICAL (Merge Blockers)
-
-### 1. Thundering Herd Problem
-**Location:** Overall architecture  
-**Issue:** At scale (100 pods x 10 threads = 1000+ simultaneous DB locks on registry refresh)  
-**Nick's suggestion:** Use Redis cache - one service writes, everyone reads from cache  
-**Options:**
-- A) Full Redis cache layer (preferred by Nick)
-- B) Single-writer pattern with Redis lock (quicker interim)
-- C) Rate limiting/jitter (band-aid)
-
-**Status:** Not started  
-**Estimate:** 6-8 hours for full Redis cache
-
----
-
-## 🟠 DESIGN QUESTIONS (Need Decisions)
-
-### 2. Route Organization
-**Location:** `api/features/admin/llm_routes.py`  
-**Issue:** Routes are under `/admin/llm` but models are shared between directories  
-**Nick's question:** Should routes be under `/llm` instead of `/admin` since models aren't admin-specific?  
-**Current:** `/api/llm/admin` (admin) and `/api/llm` (public)  
-**Status:** Already fixed routing, but organizational question remains
-
-### 3. Disabled Models in Validation
-**Location:** `data/llm_registry/schema_utils.py`  
-**Issue:** Current design allows disabled models to pass validation, runtime fallback handles it  
-**Nick's question:** "is that a good user experience?"  
-**Options:**
-- Permissive (current): Old graphs with disabled models still work, runtime fallback
-- Strict: Fail fast if disabled model selected, force user to update  
-**Status:** Design decision needed - depends on migration UX
-
-### 4. Dedicated Registry Worker
-**Location:** `api/ws_api.py`  
-**Issue:** Dedicated async worker constantly waiting for rare events (registry updates every few days)  
-**Nick's suggestion:** Generic "refresh page" notification system that could be reused for updates, registry changes, etc.  
-**Status:** Bigger refactor - defer or propose alternative?
-
-### 5. Try-Catch Pattern Consistency
-**Location:** `api/features/admin/llm_routes.py` lines 417-425  
-**Issue:** Extensive try-catch blocks wrapping every route  
-**Nick's question:** "are all these try catches actually helping us? does this follow the existing pattern?"  
-**Action:** Check existing route patterns in codebase to match style  
-**Status:** Need investigation
-
----
-
-## 🟡 CODE QUALITY (Should Fix)
-
-### 6. Dangerous Continue on Error
-**Location:** `data/block_cost_config.py` lines 112-115  
-```python
-logger.warning("Failed to query model migration overrides: %s. Proceeding with default costs.", exc)
-```
-**Nick's concern:** "this seems like a dangerous place to continue by default"  
-**Action:** Should probably raise or at least make it more obvious this is a fallback  
-**Status:** Easy fix - 5 minutes
-
-### 7. Backwards Compat Re-exports
-**Location:** `data/llm_registry/__init__.py` lines 10-15  
-```python
-# Re-export for backwards compatibility
-from backend.data.llm_registry.notifications import (...)
-```
-**Nick's question:** "why? what backwards compatibility are we providing? update the whole system to support it"  
-**Action:** Check if these are actually used elsewhere or just cruft  
-**Status:** Need to grep codebase for usage
-
-### 8. Module-Level Re-exports
-**Location:** `data/llm_registry/__init__.py` lines 16-40  
-**Issue:** Large block of re-exports from sub-modules  
-**Nick's question:** "do you actually import from the module level or is this an autogenerated file we don't use and will have to upkeep?"  
-**Action:** Check if anything imports from `backend.data.llm_registry` vs direct sub-modules  
-**Status:** Need to grep codebase for import patterns
-
-### 9. _json_to_dict Util Duplication
-**Location:** `data/llm_registry/registry.py`  
-**Nick's question:** "is this not a util already? also if its prisma json it should probably take that in as the type"  
-**Action:** Check if existing util exists, consolidate if possible  
-**Status:** Quick fix if util exists - 10 minutes
-
-### 10. Dataclasses vs Pydantic
-**Location:** `data/llm_registry/registry.py`  
-**Issue:** Registry uses `@dataclass` instead of Pydantic models  
-**Nick's request:** "these should be pydantic base models or a justification as to why not"  
-**Status:** Bigger refactor - need justification or migrate  
-**Estimate:** 1-2 hours to migrate, or provide justification for dataclasses
-
-### 11. operation_id Change
-**Location:** `api/features/store/routes.py`  
-```python
-operation_id="getV2GetCreatorDetails"
-```
-**Nick's question:** "why this change?"  
-**Action:** Explain if needed for OpenAPI collision fix, or remove if not  
-**Status:** Quick check - 5 minutes
-
-### 12. o-series Detection
-**Location:** `blocks/llm.py`  
-**Issue:** Hardcoded regex `r"(^|/)o\d"` to detect o-series models  
-**Nick's request:** Should come from DB model metadata, not hardcoded pattern  
-**Action:** Add `supportsParallelToolCalls` field or similar to model schema  
-**Status:** Medium refactor - 1 hour
-
----
-
-## ✅ FIXED TODAY
-
-1. ✅ Inline imports moved to top-level (execution_analytics_routes.py, manager.py, graph.py)
-2. ✅ Empty state UX message removed
-3. ✅ gpt-4o hardcode → DEFAULT_LLM_MODEL env var
-4. ✅ Route prefix narrowed from `/api` to `/api/llm`
-5. ✅ Removed unnecessary import alias
-6. ✅ Capabilities migrated from Provider to Model
-7. ✅ Removed commented code and obvious comments in llm.py
-
----
-
-## SUMMARY
-
-**Must discuss:**
-- Thundering herd solution approach (Redis cache vs alternatives)
-- Route organization philosophy
-- Disabled model validation UX strategy
-- Dedicated worker vs generic notification system
-
-**Quick wins to do after call:**
-- Dangerous continue on error (#6)
-- Check backwards compat re-exports usage (#7, #8)
-- operation_id justification (#11)
-
-**Bigger refactors to decide:**
-- Dataclasses → Pydantic (#10)
-- o-series detection from DB (#12)
-- _json_to_dict consolidation (#9)
-- Try-catch pattern review (#5)