docs: update README with all new features, add .env.example

README updates:
- Added 11 recent fixes/features with details
- Bug fixes: mutation confidence, RL boundary, circular import, enriched events
- New features: CI pipeline, SQLite, sessions, multi-LLM, fallback, decay profiles
- Updated limitations and roadmap sections
- All 638 tests passing

Added .env.example:
- Template for all configuration options
- Sections: storage, LLM provider, belief ecology, thresholds
- Includes OpenAI/Anthropic API key placeholders

Updated .gitignore:
- Block .env and .env.local from git

Author: moonrunnerkc

.env.example

@@ -0,0 +1,61 @@
+# ABES Configuration
+# Copy to .env and fill in your values
+
+# =============================================================================
+# STORAGE
+# =============================================================================
+# Options: memory (default), sqlite
+STORAGE_BACKEND=memory
+
+# SQLite database path (only used if STORAGE_BACKEND=sqlite)
+DATABASE_URL=sqlite+aiosqlite:///./data/abes.db
+
+# =============================================================================
+# LLM PROVIDER
+# =============================================================================
+# Options: ollama (default), openai, anthropic, none
+LLM_PROVIDER=ollama
+
+# If true, falls back to showing raw beliefs when LLM is unavailable
+LLM_FALLBACK_ENABLED=true
+
+# --- Ollama (default) ---
+OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_MODEL=llama3.1:8b-instruct-q4_0
+
+# --- OpenAI ---
+OPENAI_API_KEY=
+OPENAI_MODEL=gpt-4o-mini
+OPENAI_BASE_URL=https://api.openai.com/v1
+
+# --- Anthropic ---
+ANTHROPIC_API_KEY=
+ANTHROPIC_MODEL=claude-3-haiku-20240307
+
+# =============================================================================
+# BELIEF ECOLOGY
+# =============================================================================
+# Decay profile: aggressive (0.99), moderate (0.995), conservative (0.999), persistent (0.9999)
+DECAY_PROFILE=moderate
+
+# Or set decay rate directly (overrides profile)
+# DECAY_RATE=0.995
+
+# Embedding model (sentence-transformers)
+EMBEDDING_MODEL=all-MiniLM-L6-v2
+
+# =============================================================================
+# THRESHOLDS
+# =============================================================================
+CONFIDENCE_THRESHOLD_DECAYING=0.3
+TENSION_THRESHOLD_MUTATION=0.5
+CLUSTER_SIMILARITY_THRESHOLD=0.7
+REINFORCEMENT_SIMILARITY_THRESHOLD=0.7
+MAX_ACTIVE_BELIEFS=10000
+
+# =============================================================================
+# LLM GENERATION
+# =============================================================================
+LLM_TEMPERATURE=0.7
+LLM_MAX_TOKENS=1024
+LLM_CONTEXT_BELIEFS=15

.gitignore

@@ -1,6 +1,11 @@
 # ABES spec (internal only)
 abes_spec_md.md
 
+# Environment files (contains secrets)
+.env
+.env.local
+.env.*.local
+
 # Python
 __pycache__/
 *.py[cod]

README.md

@@ -269,29 +269,54 @@ d4dd4f5c4a777eac6d2954cd25030f74c9a4e7f27275f1c5e19fa21795d247c5  results/decay_
 
 ## Recent Fixes (2026-01-30)
 
-All tests now pass. Here's what was fixed:
+All 638 tests passing. Here's what was fixed and added:
+
+### Bug Fixes
 
 1. **Mutation engineer confidence calculation**
-   - Updated test to match intended design: mutated beliefs preserve relative confidence with a tension-based penalty
-   - Original belief at 0.4 confidence with 0.7 tension becomes 0.3 (floor value)
-   - This preserves the relative strength of beliefs rather than resetting to neutral
+   - Mutated beliefs preserve relative confidence with tension-based penalty
+   - Formula: `original - 0.1 - (tension * 0.1)`, floor 0.3
 
 2. **RL environment action decoding**
    - Fixed boundary condition in test assertion (> changed to >=)
-   - Positive actions correctly increase parameters to their expected values
 
 3. **Circular import in storage/core modules**
    - Moved imports to function level with TYPE_CHECKING guard
-   - Direct script imports now work without errors
 
 4. **ContradictionDetectedEvent enriched**
-   - Added contradicting_belief_id, belief_content, contradicting_content, similarity_score fields
-   - Events now carry full context for debugging and UI display
+   - Added: contradicting_belief_id, belief_content, contradicting_content, similarity_score
+
+### New Features
+
+5. **GitHub Actions CI Pipeline** (`.github/workflows/ci.yml`)
+   - Runs pytest on push/PR to main
+   - Includes lint checks with ruff
+
+6. **SQLite Persistence** (`STORAGE_BACKEND=sqlite`)
+   - Beliefs survive server restarts
+   - Async support via aiosqlite
+
+7. **Session Isolation** (`session_id` on beliefs)
+   - Multi-user support ready
+   - Filter beliefs by session
+
+8. **Multiple LLM Providers**
+   - Ollama (default), OpenAI, Anthropic
+   - Set via `LLM_PROVIDER` env var
+
+9. **LLM Fallback Mode**
+   - If LLM fails, returns raw beliefs
+   - Set `LLM_FALLBACK_ENABLED=true`
+
+10. **Configurable Decay Profiles**
+    - `DECAY_PROFILE`: aggressive (0.99), moderate (0.995), conservative (0.999), persistent (0.9999)
+
+11. **Configurable Embedding Model**
+    - `EMBEDDING_MODEL` env var (default: all-MiniLM-L6-v2)
 
-### Known Issues Being Worked On
+### Remaining Known Issues
 
 - Contradiction detection uses embeddings and antonym lists, not full semantic understanding
-- The decay factor default (0.995) might be too aggressive for some use cases
 
 ---