Meeting notes

This page is the group’s shared record of what was actually said. It is public, so keep unpublished results, confidential data, and anything a collaborator has not agreed to share in the private group notes instead.

Add an entry under the meeting you attended. Sign and date it. Do not edit anyone else’s entry; disagree by adding your own underneath. The fastest route is the “Edit this page” link in the right-hand margin, which opens this file on GitHub and turns your change into a pull request.

Each session page ends with the same four prompts, and they are the ones worth answering here: the evidence for your main claim, what changed your mind, what remains unresolved, and one change to the agent or its evaluation. A note that records a disagreement is more useful later than one that records a consensus.

Meeting 1: Is there a scientific method?

Held: September 15, 2026. Reading: session page, with Fortunato et al. (2018) and Simon (1973) brought into the discussion.

Transcribed from the group’s shared notes, September 23, 2026. The session page gives the full account and the readings checked against the notes.

  • Evidence for the main claim: the method as the room stated it is Cleland’s schema, H with conditions \(C\) implying E; conditions are never fully represented, so falsification in nature is rarely as clean as Platt and Chamberlin describe.
  • What changed: the smoking gun was read as both discriminating and biasing toward confirmation; Simon’s argument that generating hypotheses is a search, better or worse done, gave the group a way to treat discovery as having a logic.
  • Unresolved: whether experimental and historical science differ in kind or in degree; the source and strength of the small-stable-core finding.
  • One change to the agent: log how each rival at H was generated, so that search strategies can be compared.
  • Follow-up, September 23: the discussion of prior knowledge traced to Klahr and Simon (1999), read in full. It gives the unit graph its surprise edge and a sixth entry state (graph v0.3).

Modes meeting: definitions, evidence, and the graph

Held: date to be recorded. Reading: modes page, with Elliott et al. (2016) and Betts et al. (2021).

Convenor’s decisions on the open questions, October 21, 2026, before the room meets:

  • Split synthesis into cross-study synthesis and reconstruction from traces; add question-driven inquiry, defined from the literature on research questions and research ideas.
  • Graph v0.4.2 is the vocabulary; the decision tree (v0.5.1) is the controller.
  • Reframing a question comes after a failure to conclude, at question 8 of the tree.
  • A numeric cap on protective revisions: undecided.
  • Breadth v1 is adopted as a provisional measure until the breadth literature is read.
  • Failure means no conclusion was reached; a contradicted claim is a successful result.
  • Coding needs more papers and has to be automated: codebook v1 with a schema and a checker.
  • The controller, the modes, and the graphs are versioned in the repository; each change gets a version and a change-log row.

Meeting 10: Scientific novelty: new to whom, compared with what?

Held: September 22, 2026. Reading: session page, with Hofstra et al. (2020), Machado (2026), and Si et al. (2025) presented alongside the anchor.

Transcribed from the group’s shared notes and slides, September 23, 2026. The session page gives the full account and the readings checked against the notes.

  • Evidence for the main claim: Uzzi’s high-impact pattern is a conventional base with an atypical tail; Hofstra separates novelty from uptake and finds them pulling in opposite directions, with uptake falling as link distance grows.
  • What changed: “novelty” was used for at least four constructs (journal-pair atypicality, new concept links, embedding distance, expert rating against anchors). The group asked for a subjective and an objective rubric, kept side by side.
  • Position taken: human–AI teams raise novelty and breadth; agents raise uptake and impact, taking up new links without waiting for a local human audience.
  • Unresolved: whether an agent audience takes up distal links or reproduces the discount on them (Machado); whether distal novelty is taken up less in the earth sciences; whether novelty drives impact once a long citation window is used.
  • One change to the agent: record novelty and expected uptake as separate fields, and the distance of any link generated at H.

Inquiry-mode ledger

One brief row per reconstructed episode. Field definitions. Sign/date each row; add a separate row to disagree. Unknown chronology remains unknown. Confidential artifacts belong in private notes.

Episode/source Question before → after Action → result Modes/versions Gain and warrant Next action/agent lesson Who, when
none yet

Unresolved across the quarter

Disagreements that outlived the meeting they started in. Move an item here when it comes up a second time, and record who holds which position.

Nothing recorded yet.