RESEARCH CORRESPONDENCE
For researchers and research groups
A short description of where the project may intersect with work in artificial intelligence, systems, empirical software engineering, and human–computer interaction.
What we're working on
Our current question concerns the cost of coordination in language-model agent systems. The study will compare a single fixed-model agent, a routed single agent, routed multi-agent execution, and a supervised multi-agent configuration. The focus is verified task completion, whole-workflow cost, failures, and the effects of human review.
The one-page research brief summarizes the project for an initial reading; the public working note develops the question; the methods document sets out the proposed comparisons and their limitations. No comparative results are claimed.
Where criticism would be useful
- Evaluation design. Are there better ways to distinguish model routing from coordination, or to avoid unequal resource allocation?
- Task selection. Which tasks expose genuine decomposition benefits without favoring one experimental condition?
- Human intervention. How should review burden, risk reduction and delayed completion be measured together?
- Reproducibility. What minimum release package would make a small pilot meaningful to other researchers?
Possible forms of exchange
Relevant feedback may take the form of a short methodological comment, an additional reading recommendation, a proposed baseline, a replication suggestion, or a discussion of whether the question fits an existing research program. We do not ask for endorsements, guaranteed supervision, admission, or a statement that this work is already publishable.
Specificity helps. When contacting us, mention the research note, the aspect of the protocol you are reacting to, and any publicly accessible paper or benchmark that seems relevant.
Contact
Correspondence currently goes through the existing NOAERTH contact page. Choose the closest relevant topic and begin your message with “Scholar research correspondence” so it can be routed appropriately. This link is an organizational contact route, not a claim of a dedicated academic office.
About outreach
The same public research note can be shared with multiple research groups, but meaningful academic correspondence should be selective. University names and faculty affiliations are not endorsement signals. A useful message must refer to a specific research overlap, explain the current limitations, and give the recipient a clear way to decline further contact.