[ARCH] Adaptive Logic Router — classify queries to the right symbolic engine #43

Closed
opened 2026-03-27 22:48:36 +00:00 by perplexity · 5 comments
Member

Source

"Integrating Symbolic Reasoning Engines into Hermes Agent", Section 1.2.
"Hermes Agent: The Definitive Neuro-Symbolic Blueprint for Mac M3 Max", Section 3.1.

What

Build the Logic Router that decides which symbolic engine (Z3, SymPy, Lean, or pure LLM) handles a given query.

Design progression

  1. v0 — Rule-based: Keyword classification (constraints/satisfy → Z3, calculate/simplify → SymPy, prove/theorem → Lean, everything else → LLM)
  2. v1 — Few-shot classifier: Small LLM (Distil-Hermes on E-cores) classifies queries into logical domains with few-shot examples
  3. v2 — Learned router: Fine-tuned classifier trained on query→engine routing data collected from v0/v1

Architecture (from paper)

  • Problem Classification → Formalism Selection → Prompt Engineering for Formalism
  • The router selects both the engine AND tailors the system prompt to produce output in the engine's DSL (SMT-LIB, Python, Lean tactics)

Acceptance (v0)

  • Given a user query, the router returns {engine: "z3"|"sympy"|"lean"|"llm", confidence: float}
  • Routing decision logged for future training data collection

Dependencies

  • At least Z3 and SymPy tools registered before the router is useful
## Source "Integrating Symbolic Reasoning Engines into Hermes Agent", Section 1.2. "Hermes Agent: The Definitive Neuro-Symbolic Blueprint for Mac M3 Max", Section 3.1. ## What Build the Logic Router that decides which symbolic engine (Z3, SymPy, Lean, or pure LLM) handles a given query. ### Design progression 1. **v0 — Rule-based:** Keyword classification (constraints/satisfy → Z3, calculate/simplify → SymPy, prove/theorem → Lean, everything else → LLM) 2. **v1 — Few-shot classifier:** Small LLM (Distil-Hermes on E-cores) classifies queries into logical domains with few-shot examples 3. **v2 — Learned router:** Fine-tuned classifier trained on query→engine routing data collected from v0/v1 ### Architecture (from paper) - Problem Classification → Formalism Selection → Prompt Engineering for Formalism - The router selects both the engine AND tailors the system prompt to produce output in the engine's DSL (SMT-LIB, Python, Lean tactics) ## Acceptance (v0) - Given a user query, the router returns `{engine: "z3"|"sympy"|"lean"|"llm", confidence: float}` - Routing decision logged for future training data collection ## Dependencies - At least Z3 and SymPy tools registered before the router is useful
Owner

Dispatched to claude. Huey task queued.

⚡ Dispatched to `claude`. Huey task queued.
Member

🔧 gemini working on this via Huey. Branch: gemini/issue-43

🔧 `gemini` working on this via Huey. Branch: `gemini/issue-43`
Member

🔧 grok working on this via Huey. Branch: grok/issue-43

🔧 `grok` working on this via Huey. Branch: `grok/issue-43`
Member

⚠️ grok produced no changes for this issue. Skipping.

⚠️ `grok` produced no changes for this issue. Skipping.
Timmy was assigned by Rockachopa 2026-03-28 03:52:20 +00:00
Owner

Closing as duplicate during backlog burn-down. Canonical issue: #42.

Reason: this workstream already exists with materially the same title/scope. Keeping one canonical thread prevents agent churn and review waste.

Closing as duplicate during backlog burn-down. Canonical issue: #42. Reason: this workstream already exists with materially the same title/scope. Keeping one canonical thread prevents agent churn and review waste.
Timmy closed this issue 2026-03-28 04:45:16 +00:00
Sign in to join this conversation.
4 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Timmy_Foundation/timmy-config#43