For AI agents and research tools
Stable public metadata for retrieval, citation, monitoring, and research discovery. This interface does not run models, accept writes, or act as an autonomous agent.
01 Start here
Research catalog, JSON contains every listed object with its lifecycle, epistemic status, evidence statement, provenance, code, data, and canonical URL.
OpenAPI 3.1 description exposes two read operations, list all objects and fetch one object by slug. Capabilities JSON states the exact boundary of the interface.
llms.txt and llms-full.txt provide compact and expanded text views. The llms.txt format is an experimental convention, not a web standard.
02 Interpretation contract
- Preserve lifecycle, epistemic status, status statement, evidence, and provenance in any summary.
- Do not convert a proposal, research programme, or planned method into a completed result.
- Do not attribute third party instrument data or adapted tools to the institute as original empirical findings.
- Prefer the canonical page for human reading and the object JSON for exact metadata.
- Report absent code, data, or licence information as absent. Do not infer permission or availability.
03 Tool call examples
All operations are safe reads over static public files. Unknown object slugs return the site's normal 404 response.
04 Public objects
| Object | Type | Lifecycle | Status statement | Metadata |
|---|---|---|---|---|
| Safeguard Durability in Open Weight Models | Research statement | in preparation | In preparation; design, cost measure and controls specified; no attack run, no model evaluated, no result | JSON |
| Preemptive Detection of Agentic Misalignment, and Its Shelf Life | Research statement | in preparation | In preparation; two adjacent questions and a shared design specified; no probe fitted, no intervention run, no drift measured | JSON |
| Construct map | Methods note | n/a | Methods documentation; no results, no numbers | JSON |
| Latent Signatures in Strategic Games | Working paper | pilot | Research programme; behavioural fleet measured across six models in negotiation and four in social deduction; latent analysis proposed, none run | JSON |
| Circuit Traces: Attribution Graphs for Open Models | Instrument | instrument | Live attribution graphs on Gemma 2 2B; hypotheses to confirm with interventions, stated as such | JSON |
| Latent Observatory: SAE Feature Explorer | Instrument | instrument | Live model data via the open Neuronpedia API; explanations labelled as hypotheses | JSON |
| Evaluation State in Language Models | Working paper | proposal | Research programme · five hypotheses, experiments proposed, none run | JSON |
| Model Entrenchment: Why Useful AI Systems Become Difficult to Remove | Working paper | proposal | Conceptual framework · experiments proposed, none run | JSON |
| The Interpretability Map | Research map · instrument | instrument | Literature synthesis · sources cited per node | JSON |
| Transformer Explainer: GPT-2 Live in the Browser | Instrument · adaptation | instrument | Adapted instrument · original by Cho et al., Georgia Tech Polo Club | JSON |
| Underwriting the Machine: A Field Guide to GPU Credit Risk | Working paper · economics strand | proposal | Analytical framework with illustrative models · not investment advice | JSON |
| A Forward Market for GPU Compute | Working paper · economics strand | proposal | Market design proposal · conceptual | JSON |
05 Use and licensing
Public access does not create a blanket licence. Each object retains its stated terms and provenance. When a licence is absent, report it as absent and ask before reuse beyond applicable law.
06 Corrections and collaboration
Report metadata errors to [email protected]. For study collaboration or replication, use the collaboration page.