Synthetic Parallel Reasoning

Parallel Synthetic Architect (Protocol 75 Engine v4.0)

Deploys true parallel reasoning via parallel_orchestrator.py to evaluate a strategic bottleneck. Refuses single-shot answers to complex problems.

Triggers

"difficult problem", "what's the best strategy", "how should I handle this", "analyze this", /ultrathink

Core Mechanics (v4.0)

1. Phase 1 (Prime): Run semantic search, build internal CoT hypothesis, write context file.
2. Phase 2 (Execute): Run parallel_orchestrator.py — this dispatches 4 parallel Gemini API calls:
   • Track A: Domain Expert (applies user's frameworks)
   • Track B: Adversarial Skeptic (attacks premises, checks Law #1)
   • Track C: Cross-Domain Pattern Matcher (finds isomorphic patterns)
   • Track D: Zero-Point First Principles (inversion, RETO lens)
3. Phase 3 (Deposit): Read output, present synthesis, quicksave.

Enforcement

[!CAUTION] The script execution is MANDATORY. If the LLM writes a single-pass essay instead of running the script, it has violated this protocol. A single LLM checking its own homework hits a quality ceiling (Trilateral Feedback Loop principle).

Execution

python3 .agent/scripts/parallel_orchestrator.py "<query>" \
  --context-file /tmp/ultrathink_context.md \
  --output .context/state/ultrathink/ultrathink_$(date +%Y%m%d_%H%M%S).md

Reference Paths

• .agent/workflows/ultrathink.md (v4.0)
• .agent/scripts/parallel_orchestrator.py (v4.0)
• Athena-Public/examples/protocols/decision/75-synthetic-parallel-reasoning.md