A personal collection in progressWorking drafts · Study review notes retained

Research methods

Sixty-nine questions for dependable local AI

The research catalog covers model efficiency, CPU systems, memory, tools, and evaluation. A question belongs in the portfolio before it earns a result.

R640 & AI2 min readProgram design
Research questions catalogued
69
Draft research designs
64
Engineering-preparation records
4

A broad, structured agenda

The catalog includes quality-constrained quantization, mixed precision, MoE caching, KV memory, NUMA placement, speculative decoding, retrieval, local routing, verified coding, checkpoint recovery, and research-integrity methods. Each question has a documented purpose, next step, and acceptance criteria.

The source records 64 draft designs, four preparation-complete records, and one preparation-blocked record. These are documented planning and engineering stages—not 69 completed scientific experiments.

Preparation is useful, and it has limits

MM-16’s record reports a 31/31 engineering fixture, while MM-44 reports a 45/45 engineering receipt. Both scientific studies remained NOT_TESTED. Recovery and evaluation-integrity preparations retained unresolved independent checks.

The operating plan admits no new scientific run and asserts no accepted scientific result for these 69 cards. The catalog preserves the full questions while keeping that boundary visible.

What the next evidence must establish

A useful study needs a fixed baseline and treatment, a meaningful independent unit, a protected test set, a primary metric, uncertainty, and a stopping rule. Changes to resources or runtime must not quietly change the question.

Explore the research catalog for the complete agenda, grouped by area. Its status date is October 14; the website is a source-based portfolio rather than a live supervisor.

What we are trying to achieve

The aim is a research program that improves useful work per unit of time, memory, and effort. The questions explore both speed and trustworthiness: a faster system is only valuable when its output still meets the task’s requirements.

Sources & reading notes

  1. R640 Autonomous Operating Plan v1AI research annex; all MM-01–MM-69 cards; coverage and open gates. Drive record · access may be restricted Source record retained privately.

This is a dated, source-based research account, not a live machine report. Historical results, engineering preparation, research designs, open review gates, and measured pilots remain distinct.

← Return to R640 & AI