AI LAB / RESEARCH BACKLOG

Questions before conclusions.

A place for experiments with a defined question, reproducible method, and honest account of what the evidence supports.

All experiments below are planned. Local CPU and Metal trials have occurred, but the structured comparison described here has not been published.

PLANNED

CPU and Metal comparison

Repeat equivalent local inference workloads and record processing speed, generation speed, memory, and task quality.

PLANNED

Independent code audits

Compare implementation review with a separate audit using a reproducible task and explicit acceptance criteria.

PLANNED

Retrieval experiments

Investigate a sanitized technical-document collection, retrieval quality, and source attribution.

PLANNED

Assistant profiles

Evaluate whether task-specific system instructions help engineering, code review, and tutoring.

How experiments will be documented

Each published experiment should state the question, configuration, method, observations, interpretation, and next step. Hypotheses remain labeled until tested. Unsuccessful runs can be useful findings when the conditions are clear.