Research & Publications
Research reports and defensive publications from the NeuralMind project. All reports are verified against the live codebase.
Budget-Neutral Synaptic Recall: Displacement Injection with Learned Per-Edge Decay
Companion to the synapse learning loop publication, covering the retrieval side: recalled associations enter the agent’s context by boosting present hits and displacing the weakest vector hits one-for-one, so the token budget never grows — with per-edge forgetting rates learned from reinforcement history.
Hebbian Co-Activation with Long-Term Potentiation and Hub-Normalized Spreading Activation
The NeuralMind synapse learning loop: debounced file-edit co-activation strengthens weighted edges, wall-clock half-life decay forgets, long-term potentiation preserves habitual associations, and hub-normalized spreading activation recalls. Local-first, telemetry-free.
NeuralMind + Claude Teams: Procedures, Token Measurement, and Amnesia Prevention
Deep-dive research report: how to use NeuralMind with Claude Code teams, measure token reduction, and eliminate agent amnesia via committed team memory bundles. Verified against the live codebase. DeepSeek v4 Pro QA-corrected.
How to Use NeuralMind with OpenAI Codex: Setup, Token Measurement, Savings
Step-by-step guide: install NeuralMind, register the MCP server with Codex (or any MCP client), measure L0–L3 token budgets, and verify real dollar savings. Verified against the live codebase.
Quality-Weighted Merge with Conflict-Driven Decay
A method for resolving conflicting assertions in a distributed edge-weight graph. Each edge carries a quality score from reinforcement, recency, and conflict rate. The loser is excluded, forcing consensus.