Projects

What is being worked on now.

This page separates the research we are actively pursuing from related experiments and the tools that support the work.

Active

Current Agalmic Research projects.

These are the projects receiving substantial attention now.

A01

Attention Allocation Under Cognitive Abundance

Study how limited human evaluation should be used when AI can produce more candidates than people can review.

Where it stands: Track A v0.2 ended outcome-blind because admissible historical sources could not meet its frozen precision floor. In a separate review-value benchmark, disagreement was only a weak general signal, while a fixed near-boundary rule preserved 95% of pooled recommendation-side changes across fresh hold-outs with a median 45% reduction in simulated third-review work.

Next question: Do not build a learned allocator. Design a closer-to-operational validation of the fixed near-boundary rule using sequential review state and per-stratum preservation floors.

Programme source ↗ · v0.1 feasibility result ↗ · v0.2 source-feasibility result ↗

active

A02

Personal Capability Practice

Test whether AI assistance can improve learning and performance in real work without leaving the human unable to understand or defend the result.

Where it stands: The current Cognitive Kernel experiment is ready to move from prompt testing to matched real-world tasks. The useful evidence now is whether it improves outcomes while reducing rework and unnecessary human attention.

Next question: Run the frozen pilot on real task pairs and compare completion, rework, human attention and understanding before adding more machinery.

Practice ↗ · Cognitive Kernel protocol ↗

active

Independent projects that help answer specific questions.

These projects can provide experiments or evidence without becoming part of Agalmic Research itself.

R01

Nemosyne / Moneta

Nemosyne is an independent project that can provide concrete experiments on representation, verification and discovery without becoming part of Agalmic Research.

How it relates: Its useful role is specific: test measurable questions about how representations and workflows affect people. It cannot establish the wider economic or institutional claims of Agalmic Research.

Next: Use Nemosyne only where the intervention and outcome can be observed and measured, while keeping project ownership separate.

nemosyne.world ↗ · Nemosyne source ↗

Supporting

Tools and earlier work.

These remain useful, but they are not presented as research programmes in their own right.

S01

Expert-Attention Preservation

An earlier benchmark tested what gets lost when scarce review effort is allocated by simple ranking. Its main question now feeds the broader Attention Allocation study.

Current state: At a 20% review budget, ranking by post-review score preserved about 60.8% of accepted papers but only 10.0% of high-disagreement cases, compared with roughly 20% under random allocation.

Next: Keep the result and reuse its diagnostics where useful rather than maintaining a separate research programme.

Phase-1 findings ↗

absorbed

S02

Research Progress Instrumentation

A lightweight event log for recording how candidate ideas move through selection, testing and use.

Current state: The instrument exists, but it should receive more development only when it helps an active project answer a real question.

Next: Use it only where it reduces uncertainty or manual work in active research.

Instrumentation source ↗

supporting