v1.7.0 — The Great Unification

[akarlaraytu]

v1.7.0 — The Great Unification

Chimera's safety layer now runs on CSL-Core — the same formally verified policy engine that blocked 22/22 adversarial attacks against GPT-4.1, GPT-4o, Claude Sonnet 4, and Gemini 2.0 Flash.

This release delivers on the v1.6.0 roadmap promise: migrating SymbolicGuardian from hardcoded Python rules to deterministic, Z3-verified .csl policy files — with zero breaking changes.

---

What Changed

🛡️ CSL-Core Integration

• SymbolicGuardianV4 (e-commerce) and SymbolicGuardianV6 (quant) now delegate all rule enforcement to CSL-Core's compiled policy engine.
• Trading and pricing rules moved to human-readable .csl policy files under policies/.
• All policies are Z3-verified at startup — mathematical proof of logical consistency before a single action is evaluated.
• Sub-millisecond policy evaluation (previously ~50ms with the old rule engine).

📜 Policy Files

• policies/ecommerce_guard.csl — Ad spend caps, price change limits, margin thresholds with safety buffers.
• policies/quant_guard.csl — Portfolio health checks, position-aware constraints (SELL requires long, SHORT requires flag), computed amount limits.

🔧 Architecture

• New src/csl_guardian.py — Drop-in wrapper maintaining identical __init__ signatures and validate_action() / repair_action() interfaces.
• Hybrid design: CSL-Core handles deterministic rule enforcement, Python handles computed fields (future price, margin, binary search for max position size) and repair logic.
• Graceful fallback to legacy Python validation if csl-core is not installed.

🧹 Codebase Cleanup

• Migrated all LangChain imports from deprecated langchain.* to langchain-classic / langchain-core across 8 files.
• Cleaned requirements.txt: removed ~60 transitive dependencies, added missing langchain-classic, organized into categories.
• Removed stale v1.6.0 deprecation warnings from components.py.

---

Why This Matters

Before (v1.5.x)	After (v1.7.0)
Rules buried in 200+ lines of Python if/else	Rules in 30-line .csl files, readable as English
No consistency checking — contradictions found at runtime	Z3 proves consistency at load time
Rule changes require Python edits + testing	Edit .csl, re-verify with cslcore verify
No formal connection to safety benchmarks	Same engine that passed 22/22 adversarial attack benchmark

---

Migration

pip install csl-core    # That's it. Policies auto-load from policies/

Zero changes needed in calling code. SymbolicGuardianV4() and SymbolicGuardianV6() work exactly as before — same class names, same parameters, same return format.

---

Verified & Tested

• ✅ Streamlit Adaptive Strategy Lab — agent proposes, CSL validates, repair suggests fixes
• ✅ Colosseum multi-agent arena — all gladiators governed by CSL policies
• ✅ Quant backtest suite — 30-day recency test passes with CSL-Core enforcement
• ✅ Live paper trader — no disruption to active trading stream

---

What's Next

• Causal engine integration with CSL policy feedback loops
• Multi-agent Colosseum with per-agent .csl policies
• Temporal logic constraints (LTL/CTL) for multi-step reasoning
• Formal verification of causal-policy interaction properties

---

Full Changelog: v1.6.0...v1.7.0

CSL-Core Engine: chimera-protocol/csl-core · pip install csl-core

---

This discussion was created from the release v1.7.0 — The Great Unification.