MetaScience, Scientific Inquiry, and Agents

Denolle Lab research-group seminar | Fall 2026

A thirteen-week reading seminar reconstructing how research actually proceeds, and translating those lessons into agent designs and evaluations that can be checked.

About this seminar

MetaScience, Scientific Inquiry, and Agents
Denolle Lab research-group pilot | Fall 2026
13 Tuesdays, September 8-December 1, 1:00-2:00 PM Pacific

Purpose

Reconstruct how experimental, observational, historical, and theoretical research actually proceeds, then translate those lessons into scientifically grounded agent designs and evaluations. The adequacy of a hypothesis-generation/testing loop is a question to investigate, not a conclusion the seminar assumes.

One proposal runs alongside the rest: that agents for science let a researcher work competently across more fields, making individual polymathy practical again. It is an open question here. Meeting 12 states the hypothesis, the historical evidence bearing on it, and the observations that would count against it.

This is an exploratory quarter for the group. Nothing here is a settled syllabus, and the reading library is meant to grow. How to contribute covers adding a note, a reading, or a correction.

How the paper pairings work

Each meeting uses two scholarly papers with a specific relationship: argument/counterpoint, discovery/diagnostic scrutiny, capability/calibration, or practice/framework. A pairing is not necessarily a disagreement. Read the assigned portions; rotating discussants read both in depth. No participant is expected to read every optional extension.

Meeting 13 additionally carries two discussant-led papers, because it merges what were previously separate architecture and evaluation meetings. Each is taken by one rotating discussant; everyone else reads the abstract. The reading load stays at two papers per participant throughout.

For each paper distinguish its scientific claim, the evidence offered, its limits, and the evidence it provides about the research process. A philosophical argument, a retrospective autobiography, an archival history, a primary research report, and a benchmark paper are not interchangeable records of actual practice. Where chronology is not documented, label reconstruction as inference.

Schedule

Meeting	Tuesday, 2026	Theme	Pair
1	Sep 8	Is there a scientific method?	Platt 1964 / Cleland 2002
2	Sep 15	Exploratory experimentation is not theory-free	Steinle 1997 / Karaca 2013
3	Sep 22	From coastal traces to earthquake histories	Atwater 1987 / Nelson et al. 1996
4	Sep 29	Plate tectonics: concept formation and a discriminating test	Wilson 1965 / Sykes 1967
5	Oct 6	New sensing capability versus trustworthy measurement	Lindsey et al. 2019 / Lindsey et al. 2020
6	Oct 13	Explore freely; distinguish exploration from confirmation	Tukey 1962 / Nosek et al. 2018
7	Oct 20	Field and marine research under real constraints	Powell 2007 / Becker et al. 2019
8	Oct 27	Serendipity, anomaly, and scientific interestingness	Moore 2025 / Yaqub 2018
9	Nov 3	Hypothesis testing when several models fit	Oreskes et al. 1994 / Beven & Freer 2001
10	Nov 10	Scientific novelty: new to whom, compared with what?	Uzzi et al. 2013 / Fontana et al. 2020
11	Nov 17	Scientific advance: cumulative work and field-shaping change	Wu et al. 2019 / Petersen et al. 2025
12	Nov 24	Scientific polymathy: breadth, specialization, and borrowed competence	Burke 2020 / Kitcher 1990
13	Dec 1	What architecture follows from practice, and what would show it worked?	Boiko et al. 2023 / Chen et al. 2025

All times are local Pacific time, including the daylight-saving transition. The scheduled November 24 meeting is retained; no holiday break has been inserted.

Standing questions

1. What was known at the beginning, and what was the initial question?
2. Which observation, representation, instrument, or model changed the next action?
3. When did the eventual hypothesis appear, and what was exploratory or confirmatory?
4. Where did surprise, failed expectation, opportunism, or missing evidence matter?
5. What is documented about the trajectory, and what disappears from the publication narrative?
6. Which action could an agent perform, and how would we know it had performed it well?

Quarter outputs

The group develops an inquiry-action vocabulary, exploration policy, trace-inference worksheet, measurement-provenance checklist, adaptive observation plan, Novelty Dossier, Scientific Advance Profile, Polymathy Profile, and finally an agent architecture and evaluation protocol produced together. The working rubrics are design proposals to revise, not validated scales.

Before specifying any agent, consult the prior art chapter: it catalogues the systems and benchmarks that already exist and, for each, what its scores do not establish. The point of building anything here is to add to that record rather than repeat it.

Reading access and public use

Every assigned paper has a DOI, publisher, repository, or official proceedings link. Some full texts require institutional access. This book does not redistribute journal PDFs or claim that freely readable papers may automatically be republished. Use the publisher’s download link or institutional library for personal reading, and check the license before putting third-party files on the public site.

The reading library collects all links. The bibliography is generated from a reusable BibTeX file.

Version and scope

Working curriculum v0.3, revised September 5, 2026. The reading audit replaces indirect or underspecified pairings with focused cases, adds implemented agents and benchmarks, and separates required pairs from optional extensions. v0.3 merges the former architecture and evaluation meetings into meeting 13 and adds meeting 12 on scientific polymathy; see the reading audit for the rationale. Current evidence does not establish that one architecture is best for all scientific inquiry.