docs: add navigation index for all investigation documents

Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>

Author: cursoragent

INDEX.md

@@ -0,0 +1,144 @@
+# Investigation Index
+
+## Quick Navigation
+
+### 🚀 Start Here
+- **[INVESTIGATION_README.md](./INVESTIGATION_README.md)** - Overview of all documents
+
+### 📋 Executive Summary  
+- **[SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md)** - Quick TL;DR for team (5 min read)
+
+### 🔍 Full Investigation
+- **[INVESTIGATION_OPUS_4_7_HAIKU_REGRESSION.md](./INVESTIGATION_OPUS_4_7_HAIKU_REGRESSION.md)** - Complete analysis (15 min read)
+
+### 💻 Technical Deep-Dive
+- **[CODE_PATTERNS_ANALYSIS.md](./CODE_PATTERNS_ANALYSIS.md)** - Code patterns & fixes (20 min read)
+
+### 📊 Visual Explanation
+- **[VISUAL_TIMELINE.md](./VISUAL_TIMELINE.md)** - Diagrams & timeline (10 min read)
+
+---
+
+## Reading Paths
+
+### Path 1: Quick Brief (10 minutes)
+1. [INVESTIGATION_README.md](./INVESTIGATION_README.md) - Overview
+2. [SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md) - Action items
+
+**Result:** Understand what happened and what needs to be done.
+
+### Path 2: Implementation (30 minutes)
+1. [SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md) - Context
+2. [CODE_PATTERNS_ANALYSIS.md](./CODE_PATTERNS_ANALYSIS.md) - Specific fixes
+3. [VISUAL_TIMELINE.md](./VISUAL_TIMELINE.md) - Before/after comparison
+
+**Result:** Ready to implement the hotfix.
+
+### Path 3: Complete Understanding (60 minutes)
+1. [INVESTIGATION_README.md](./INVESTIGATION_README.md) - Start here
+2. [VISUAL_TIMELINE.md](./VISUAL_TIMELINE.md) - See the timeline
+3. [INVESTIGATION_OPUS_4_7_HAIKU_REGRESSION.md](./INVESTIGATION_OPUS_4_7_HAIKU_REGRESSION.md) - Full details
+4. [CODE_PATTERNS_ANALYSIS.md](./CODE_PATTERNS_ANALYSIS.md) - Technical depth
+5. [SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md) - Action plan
+
+**Result:** Complete context for planning long-term fix.
+
+---
+
+## Documents by Audience
+
+### For Engineering Managers
+- [SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md) - What happened, impact, action items
+- [VISUAL_TIMELINE.md](./VISUAL_TIMELINE.md) - Timeline and cost impact
+
+### For Developers Implementing Fix
+- [CODE_PATTERNS_ANALYSIS.md](./CODE_PATTERNS_ANALYSIS.md) - Exact code changes needed
+- [SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md) - Files to modify
+
+### For Architects Planning Long-term Fix
+- [INVESTIGATION_OPUS_4_7_HAIKU_REGRESSION.md](./INVESTIGATION_OPUS_4_7_HAIKU_REGRESSION.md) - Root cause analysis
+- [CODE_PATTERNS_ANALYSIS.md](./CODE_PATTERNS_ANALYSIS.md) - Architectural fix strategy
+
+### For Customer Success / Support
+- [VISUAL_TIMELINE.md](./VISUAL_TIMELINE.md) - What customers experienced
+- [SUMMARY_FOR_TEAM.md](./SUMMARY_FOR_TEAM.md) - Timeline of events
+
+---
+
+## Key Findings Summary
+
+### The Issue
+Upgrading from `v1.82.3-stable.patch.4` to `v1.82.3-stable.opus-4.7` caused cost spikes on Haiku 4.5 Bedrock due to "unsupported thinking metadata" errors.
+
+### Root Cause
+Broad pattern matching (`"claude-haiku-4" in model`) added for Opus 4.7 support inadvertently matched Haiku 4.5, causing thinking metadata to be sent to models that don't support it.
+
+### Immediate Fix
+Explicitly exclude Haiku 4.5 before checking broader Claude 4.x patterns.
+
+### Long-term Fix
+Fix `supports_reasoning()` provider name mismatch and remove all hardcoded model version checks.
+
+---
+
+## Files to Review in LiteLLM Repo
+
+Priority order:
+
+1. **`litellm/llms/bedrock/common_utils.py`**
+   - Function: `is_claude_4_5_on_bedrock()`
+   - Lines: 565-597
+
+2. **`litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py`**
+   - Function: `_supports_extended_thinking_on_bedrock()`
+
+3. **`litellm/llms/anthropic/chat/transformation.py`**
+   - Functions: `_is_opus_4_7_model()`, thinking logic
+   - Lines: 172-210, 237-248
+
+4. **`litellm/llms/bedrock/chat/converse_transformation.py`**
+   - Function: `get_supported_openai_params()`
+   - Lines: 565-580
+
+---
+
+## Related GitHub Resources
+
+### PRs
+- [#25867](https://github.com/BerriAI/litellm/pull/25867) - Day 0 Opus 4.7 support (merged)
+- [#25876](https://github.com/BerriAI/litellm/pull/25876) - Opus 4.7 hotfix (merged April 16)
+- [#24053](https://github.com/BerriAI/litellm/pull/24053) - Claude 4+ thinking support (open)
+
+### Discussions
+- [#22555](https://github.com/BerriAI/litellm/discussions/22555) - Day 0 Model Release Gaps
+
+### Issues
+Referenced in investigation but no specific issue filed yet for this regression.
+
+---
+
+## Timeline
+
+- **April 16, 2026 5:19pm** - PR #25867 merged
+- **April 16, 2026 7:19pm** - PR #25876 merged, v1.82.3-stable.opus-4.7 released
+- **April 17, 2026 7:32pm** - User reports cost spike on Slack
+- **April 17, 2026 7:47pm** - Investigation requested
+- **April 17, 2026 [Now]** - Investigation complete
+
+---
+
+## Next Steps
+
+- [ ] LiteLLM team validates findings
+- [ ] Deploy immediate hotfix
+- [ ] Notify affected users
+- [ ] Plan architectural fix
+- [ ] Update CI/CD requirements
+- [ ] Document lessons learned
+
+---
+
+**Investigation Status:** ✅ Complete  
+**Branch:** `cursor/investigate-opus-4-7-haiku-regression-5249`  
+**Created by:** Cursor Cloud Agent  
+**Date:** April 17, 2026