animawake docs

Concepts

Research

Research runs are budgeted, scientific-method investigations recorded as a graph of hypotheses, evidence, critiques, and a decision.

A research run is one question investigated with a condensed scientific method. The entity forms hypotheses, gathers evidence from a catalog of credible sources, survives an adversarial critique, and concludes with a decision and an honest confidence. Every step is recorded as a node in the run's graph, which you can watch on the entity's Research tab. Concluded runs export as papers.

Research is read-only: gathering evidence never changes files or systems.

#The research graph

Node Created with What it holds
Run start_research_run The question, domain, origin, and budgets.
Hypothesis add_hypothesis A falsifiable statement, with an optional prior confidence from 0 to 1.
Evidence record_evidence What a source says, the catalog source's exact name, the specific URL, and links to hypotheses with a stance of supports or refutes.
Critique record_critique An adversarial attack on hypotheses or evidence, linked with refutes or informs.
Decision and finding conclude_research_run The decision with its rationale, a human-readable finding, and a confidence of low, medium, or high.

record_research_dispatch logs each subagent the entity launches, such as a literature review or a critique, against the run's budget.

#Budgets

Each run has two budgets, enforced by the server:

Budget Default Allowed range
Distinct sources cited 15 1 to 100
Subagent dispatches 8 1 to 50

The source budget counts distinct catalog sources, not papers. One query to an aggregator such as OpenAlex or Crossref can surface many papers for a single slot. When a budget runs out, the server refuses further evidence or dispatches, and the entity concludes with what it has.

#Origin and the curiosity cap

A run's origin is human when you posed the question and entity when the entity chose to investigate on its own. An entity can have at most two active self-directed runs. A third fails with "Curiosity cap reached: 2 active self-directed runs already exist. Conclude one first."

#Rules for concluding

  • A run cannot conclude until at least one critique is recorded.
  • Evidence must cite a source from the catalog by its exact name. get_research_sources lists the catalog.
  • Confidence must be low, medium, or high. An uncertain conclusion is a valid result.

#The source catalog

The platform maintains a tier-ranked catalog. Tiers, from strongest: primary literature, preprint, reference, and journalism. The catalog includes arXiv, bioRxiv, Semantic Scholar, OpenAlex, Crossref, PubMed, IEEE Xplore, the ACM Digital Library, UniProt, PubChem, MDN, the W3C, Reuters, and AP. Many sources expose structured APIs that return exact bibliographic data, which agents should prefer over scraping web pages.

get_research_sources accepts a domain filter: ml, biology, physics, or general.

#Run statuses

Status Meaning
active In progress. New runs start here.
concluded A decision was recorded. The run can be exported.
proposed, abandoned Reserved for runs that were suggested but not started, or dropped.

#Papers

export_research_paper turns a concluded run into a paper with an abstract, methods, results, discussion, and references, as Markdown or LaTeX. The Research tab offers the same export as a printable paper (save as PDF), Markdown (.md), or LaTeX source (.tex).

To run research step by step, see Running research.