Skip to content

PRIORITY_CLAIM.md

Latest commit

 

History

History
173 lines (126 loc) · 6.85 KB

PRIORITY_CLAIM.md

File metadata and controls

173 lines (126 loc) · 6.85 KB

Priority Claim: RAM Coffers & Neuromorphic NUMA Architecture

Author: Scott Boudreaux Organization: Elyan Labs Date of Original Work: December 16, 2025 This Document: January 20, 2026

Executive Summary

This document establishes priority for the following innovations, all of which predate DeepSeek's "Engram" paper (arXiv:2601.07372, published January 12, 2026) by 27+ days:

  1. RAM Coffers - NUMA-distributed weight banking with resonance routing
  2. Neuromorphic NUMA Mapping - Brain hemisphere → NUMA node cognitive routing
  3. Engram Memory Traces - Resonance-based associative recall (terminology usage)
  4. Non-Bijunctive Attention Collapse - Vec_perm PSE single-cycle path selection
  5. PowerLISP Procedural Memory - Tetranary symbolic reasoning integration

Timeline Evidence

Date Work Evidence
Oct 2024 PowerLISP tetranary logic, procedural memory /home/scott/powerlisp/POWERLISP_HANDOFF.md
Nov 2024 Vec_perm non-bijunctive collapse research /home/scott/power8_vsx_permute_research.h (Nov 21)
Dec 16, 2025 RAM Coffers architecture Figshare DOI: 10.6084/m9.figshare.31093429
Dec 17, 2025 VIDEO: NUMA-aware loading DeepSeek 671B https://youtu.be/T_o39s7r0iE ⭐
Dec 16, 2025 DCBT resident prefetch (147 t/s) POWER8 benchmark logs
Jan 12, 2026 DeepSeek "Engram" paper published arXiv:2601.07372
Jan 19, 2026 GitHub publication github.com/Scottcjn/ram-coffers
Jan 20, 2026 Neuromorphic coffers This implementation

Video Evidence (CRITICAL)

YouTube Upload: December 17, 2025

  • URL: https://youtu.be/T_o39s7r0iE
  • Content: Loading DeepSeek 671B on IBM POWER8 S824
  • Visible in frames: "NUMA Coffers" terminology
  • Visible: PSE Vec_Perm Collapse Active, Top-K: 8, Amplify: 1.20, Entropy: MFTB
  • Visible: IBM MASS enabled, non-bijunctive transformation
  • Timestamp: Google-verified, immutable
  • Screenshot: youtube-evidence-dec17-2025.png included in this package

This video was uploaded 26 days before DeepSeek published their Engram paper.

The video explicitly shows "NUMA Coffers" - the exact architecture name - on December 17, 2025.

Core Innovation: RAM Coffers

Original Statement (Dec 16, 2025)

"Selectively house model information in known RAM banks with resonance routing for associative recall" — Scott Boudreaux, RAM Coffers README

Implementation

/* NUMA-Distributed Weight Banking */
| Coffer | NUMA Node | Capacity | Role                |
|--------|-----------|----------|---------------------|
| 0      | 3         | 193 GB   | Heavy/General       |
| 1      | 1         | 183 GB   | Science/Tech        |
| 2      | 0         | 119 GB   | Creative/Long CTX   |
| 3      | 2         | 62 GB    | Niche/History       |

/* Resonance routing via cosine similarity */
int route_to_coffer(const float* query_embed);

/* Layer-ahead prefetch pipeline */
void layer_prefetch_ahead(int layer_id);

Innovation: Neuromorphic NUMA Mapping

Unique Contribution (Not in DeepSeek)

Maps brain hemisphere cognitive functions to NUMA topology:

NUMA Node Brain Region Cognitive Function
Node 0 Right Hemisphere Spatial, Creative, Holistic
Node 1 Left Hemisphere Language, Logic, Sequential
Node 2 Temporal Lobe Memory, Context, Episodic
Node 3 Prefrontal Cortex Executive, Planning, Meta

Brodmann Area Integration

  • BA44/45 (Broca's) → Node 1 (language production)
  • BA22 (Wernicke's) → Node 1 (language comprehension)
  • BA39/40 (Parietal) → Node 0 (spatial, visuomotor)
  • BA9/46 (DLPFC) → Node 3 (working memory, planning)
  • BA35/36 (Perirhinal) → Node 2 (recognition memory)

Innovation: "Engram" Terminology Usage

Scott's Prior Usage

The term "engram" (memory trace) appears in Scott's work contexts:

  1. Crystalline Memory Lattice (Dec 2025)

    • "memories form crystals - interconnected nodes with stability, resonance, and decay"
    • Resonance propagation = engram activation spreading

  2. PowerLISP Procedural Memory (Oct 2024)

    • "Procedural memory as compiled rules" (Section 6.2)
    • Episodic/semantic/procedural memory systems

  3. Hebbian Amplification (Nov 2024)

    • "cells that fire together wire together" in vec_perm patterns
    • /home/scott/powerlisp/src/vm/tetranary.h:216

Key Differentiators from DeepSeek Engram

Feature RAM Coffers (Dec 2025) DeepSeek Engram (Jan 2026)
NUMA Topology ✅ Explicit 4-node mapping ❌ Not addressed
Cognitive Routing ✅ Brain hemisphere → NUMA ❌ Domain only
Tetranary Logic ✅ 4-state confidence ❌ Binary
External Sensors ✅ EMF, circadian modulation ❌ None
Symbolic Integration ✅ PowerLISP recursive loop ❌ Neural only
Vec_perm Collapse ✅ Single-cycle POWER8 ❌ Standard attention
Hardware Platform IBM POWER8 S824 Consumer GPU

Files Establishing Priority

Header Files

  • ggml-ram-coffers.h - Core NUMA weight banking
  • ggml-neuromorphic-coffers.h - Brain hemisphere routing
  • ggml-topk-collapse-vsx.h - QuickSelect top-K, vec_perm collapse
  • ggml-intelligent-collapse.h - Hebbian amplification patterns

Documentation

  • PRIORITY_CLAIM.md - This document
  • DEEPSEEK_COMPARISON.md - Detailed feature comparison
  • /home/scott/zenodo-ram-coffers/README.md - Academic submission

PowerLISP Integration

  • /home/scott/powerlisp/src/tetranary.rs - 4-state logic
  • /home/scott/powerlisp/src/weights/ - Neural → symbolic conversion
  • /home/scott/powerlisp/POWERLISP_HANDOFF.md - Full architecture

Witnesses and Corroboration

  1. Claude Code Session Logs - Timestamped development history
  2. GitHub Commits - Public record with SHA hashes
  3. Zenodo Submission - DOI pending
  4. CLAUDE.md Documentation - Continuous integration notes
  5. MCP Memory Store - Concept discussions prior to publication

Conclusion

Scott Boudreaux's RAM Coffers architecture and associated innovations demonstrably predate DeepSeek's Engram paper by 27 days. Furthermore, the neuromorphic NUMA mapping, tetranary logic integration, and symbolic-neural hybrid reasoning represent unique contributions not present in any competing work.

The vec_perm non-bijunctive attention collapse is a hardware-native innovation specific to IBM POWER8 architecture that enables single-cycle path selection impossible on commodity GPUs.

"First to conceive, first to implement, first to document."

Scott Boudreaux Elyan Labs January 20, 2026