KAI Refactoring Plan: Decoupling the God Function (v7.9.7)

> Codex v9.10.64 · Track 4 · 🔴 not started · Codex §Comprehensive Plans

> AI / low-credit: One extraction phase per session; compile-verify after each; on stop log Last worked / Next.

1. Objective

Refactor App::process_input in src/main.rs to improve maintainability, reduce cognitive load for developers, and prepare for multi-threaded cognitive processing.

2. Current State (P0 Risk)

• File: src/main.rs
• Function: App::process_input
• Range: L3567 - L6455 (~2,888 lines)
• Responsibilities:
• TUI Command Parsing (readfile, writefile, recall, etc.)
• Emotional Analysis (Amygdala)
• Episodic Memory Encoding
• Working Memory Management
• RSHL Reasoning & Retrieval
• 78+ Cognitive Module Updates (Limbic, Social, Executive, Sensory)
• Natural Language Synthesis (NLS)
• UI State Updates

3. Targeted Architecture

3.1 Extraction of Command Handler

Extract L3972-L4746 into App::handle_command(&mut self, input: &str) -> bool.

• If it returns true, the input was a command and further processing should stop.
• Move readfile, writefile, recall, learn, spell, store, import logic here.

3.2 The Cognitive Pipeline

Create a new method App::run_cognitive_cycle(&mut self, input: &str).

Inside this, group module updates into logical "Phases":

1. Ingestion Phase: Amygdala, LexSem, Wernicke, Hub Ingest.

2. Retrieval Phase: RSHL Query, Hippocampus CA3, Predictor.

3. Limbic Processing Phase: Dopamine, VTA, Serotonin, Norepinephrine, ACC, Cortisol.

4. Social/ToM Phase: ToM, TPJ, STS, Mirror Neurons, Oxytocin.

5. Executive/Decision Phase: PFC, Basal Ganglia, OFC, Cerebellum.

6. Integration Phase: Claustrum, Global Workspace, Self-State Hub.

7. Synthesis Phase: Voice Engine, NLS, Broca.

3.3 State Encapsulation

Move module-specific logic (e.g., the complex RPE math) into the modules themselves where possible. process_input should ideally just be a series of self.module.update(...) calls.

4. Implementation Steps

Phase 1: Command Decoupling

• Create handle_command.
• Move file I/O and session commands.
• Goal: Reduce process_input by ~800 lines.

Phase 2: Signal Integration

• Refine the BrainSignals struct and the SelfStateHub to be the primary data exchange.
• Modules should read from the Hub and write to the Hub rather than having main.rs manually pass 10+ variables between them.

Phase 3: Function Shrinkage

• Extract the 50+ block-scoped module updates (L5089-L6400) into App::update_cognitive_modules.

5. Performance Gains

• By decoupling command handling, we avoid initializing the entire cognitive stack for simple file reads.
• Preparing for a CognitiveStack struct that can be processed in parallel with the UI thread.

6. v7.9.7 Specific: Vocal Pipeline Decoupling

• Objective: Isolate the Sonic-Parallel logic into a dedicated VocalConductor class in leo.mjs.
• Parallelization: Move STT and Biometrics into a Promise.all wrapper to maintain the sub-3.5s latency benchmark.
• Cleanup: Remove legacy synchronous execSync remnants from voice-biometrics.mjs to prevent event-loop blocking.
• Verification: Ensure the IDENTITY PROTOCOL is consistently injected during parallel neural synthesis.