{
  "$schema": "https://latentmindsinstitute.com/api/schema/research-catalog-v1.json",
  "schemaVersion": "1.0.0",
  "name": "Latent Minds Institute research object catalog",
  "description": "A read only catalog of the institute's public research statements, papers, methods notes, and instruments. Status and evidence fields must be preserved when summarizing an object.",
  "publisher": {
    "name": "Latent Minds Institute",
    "url": "https://latentmindsinstitute.com/",
    "email": "research@latentmindsinstitute.com"
  },
  "sourceOfTruth": "https://latentmindsinstitute.com/data/research-objects.json",
  "documentation": "https://latentmindsinstitute.com/for-agents/",
  "openapi": "https://latentmindsinstitute.com/api/openapi.json",
  "objectCount": 12,
  "objects": [
    {
      "slug": "open-weight-safeguard-durability",
      "url": "/research/open-weight-safeguard-durability/",
      "title": "Safeguard Durability in Open Weight Models",
      "type": "statement",
      "typeLabel": "Research statement",
      "lifecycle": "in-preparation",
      "epistemicStatus": "design-only",
      "statusLabel": "In preparation; design, cost measure and controls specified; no attack run, no model evaluated, no result",
      "programme": "deployment",
      "question": "When a model's weights are public, how much adversarial effort does each published safeguard actually impose, and how much of that cost survives a modest fine tuning budget?",
      "date": "2026-07-28",
      "version": "v0.1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "adversarial red teaming across input, decoding, weight modification and composed surfaces (planned)",
        "attack cost curves against a fixed behaviour set (planned)",
        "capability controls separating a defeated safeguard from a degraded model (planned)",
        "attacker effort controls so null results read as bounded search (planned)"
      ],
      "models": [
        "published open weight models (planned)"
      ],
      "evidence": "none; this is a design statement published before any run",
      "code": null,
      "data": null,
      "provenance": "Published in advance of results so the question, the cost measure and the controls are on the record before any finding is",
      "related": [
        "model-entrenchment"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/research/open-weight-safeguard-durability/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/open-weight-safeguard-durability.json"
    },
    {
      "slug": "agentic-misalignment-detection",
      "url": "/research/agentic-misalignment-detection/",
      "title": "Preemptive Detection of Agentic Misalignment, and Its Shelf Life",
      "type": "statement",
      "typeLabel": "Research statement",
      "lifecycle": "in-preparation",
      "epistemicStatus": "design-only",
      "statusLabel": "In preparation; two adjacent questions and a shared design specified; no probe fitted, no intervention run, no drift measured",
      "programme": "cognition",
      "question": "Can representation engineering detect agentic misalignment from internal state before the action is taken, and does the self representation such a detector reads remain stable as reinforcement learning horizons lengthen?",
      "date": "2026-07-28",
      "version": "v0.1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "readouts over a fixed set of agentic settings, taken before the action is emitted (planned)",
        "intervention to separate a used internal variable from a merely decodable one (planned)",
        "re-evaluation across a reinforcement learning checkpoint series at matched behavioural performance (planned)",
        "surface form, capability and lexical baseline controls (planned)"
      ],
      "models": [
        "open weight models with reinforcement learning checkpoint series (planned)"
      ],
      "evidence": "none; this is a design statement published before any run",
      "code": null,
      "data": null,
      "provenance": "The two questions are posed together because the second sets the shelf life of any detector built under the first",
      "related": [
        "evaluation-state",
        "latent-signatures-strategic-games"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/research/agentic-misalignment-detection/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/agentic-misalignment-detection.json"
    },
    {
      "slug": "construct-map",
      "url": "/research/construct-map/",
      "title": "Construct map",
      "type": "note",
      "typeLabel": "Methods note",
      "lifecycle": "n/a",
      "epistemicStatus": "methods-note",
      "statusLabel": "Methods documentation; no results, no numbers",
      "programme": "cognition",
      "question": "For each strategic game family, what construct is measured, what control does a reading require, and which claim does the evidence never license?",
      "date": "2026-07-18",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "construct specification per game family",
        "required controls stated before code",
        "explicit never claim boundaries"
      ],
      "models": [],
      "evidence": "documentation of method only; contains no measurements and states none",
      "code": null,
      "data": null,
      "provenance": "Written for the latent arena game families behind Latent Signatures in Strategic Games",
      "related": [
        "latent-signatures-strategic-games"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/research/construct-map/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/construct-map.json"
    },
    {
      "slug": "latent-signatures-strategic-games",
      "url": "/papers/latent-signatures-strategic-games/",
      "title": "Latent Signatures in Strategic Games",
      "type": "working-paper",
      "typeLabel": "Working paper",
      "lifecycle": "pilot",
      "epistemicStatus": "research-programme",
      "statusLabel": "Research programme; behavioural fleet measured across six models in negotiation and four in social deduction; latent analysis proposed, none run",
      "programme": "cognition",
      "question": "Do deception, opponent-modelling, and evaluation-awareness signatures in an open model's activations predict and cause its moves in strategic games?",
      "date": "2026-07-11",
      "version": "v0.2",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "self-hosted game environments",
        "behavioural self-play (measured)",
        "instrumented self-play (proposed)",
        "signature probes (proposed)",
        "steering during play (proposed)"
      ],
      "models": [
        "open-weight models (proposed)"
      ],
      "evidence": "research programme with three signature hypotheses and a three-stage proposed pipeline; the latent-arena harness is built and tested, the behavioural fleet is measured (negotiation self-play across six models, social deduction with four model tables), no activations captured",
      "code": null,
      "data": null,
      "provenance": "Built for Latent Minds Institute; the harness is the latent-arena repository, being prepared for public release and available on request until then",
      "related": [
        "evaluation-state",
        "model-entrenchment"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/papers/latent-signatures-strategic-games/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/latent-signatures-strategic-games.json"
    },
    {
      "slug": "circuit-traces",
      "url": "/instruments/circuit-traces/",
      "title": "Circuit Traces: Attribution Graphs for Open Models",
      "type": "instrument",
      "typeLabel": "Instrument",
      "lifecycle": "instrument",
      "epistemicStatus": "instrument",
      "statusLabel": "Live attribution graphs on Gemma 2 2B; hypotheses to confirm with interventions, stated as such",
      "programme": "computation",
      "question": "Which features caused which, from prompt to prediction, and does the causal story survive comparison across tasks, models, and methods?",
      "date": "2026-07-11",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "attribution graphs on transcoder replacement models",
        "curated case-study reading guides",
        "validated graph loading"
      ],
      "models": [
        "Gemma 2 2B (Gemma Scope transcoders)"
      ],
      "evidence": "real causal graphs served live; replacement-model caveat stated in interface",
      "code": "https://github.com/safety-research/circuit-tracer",
      "data": null,
      "provenance": "Method: Ameisen et al. / Lindsey et al. (2025), Transformer Circuits; implementation: safety-research/circuit-tracer; hosting: Neuronpedia",
      "related": [
        "latent-observatory",
        "transformer-explainer"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/instruments/circuit-traces/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/circuit-traces.json"
    },
    {
      "slug": "latent-observatory",
      "url": "/instruments/latent-observatory/",
      "title": "Latent Observatory: SAE Feature Explorer",
      "type": "instrument",
      "typeLabel": "Instrument",
      "lifecycle": "instrument",
      "epistemicStatus": "instrument",
      "statusLabel": "Live model data via the open Neuronpedia API; explanations labelled as hypotheses",
      "programme": "representations",
      "question": "What has a sparse autoencoder actually learned about a model, and does the auto-interp story survive contact with the activation evidence?",
      "date": "2026-07-11",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "semantic search over auto-interp explanations",
        "activation-record inspection",
        "logit-effect readout",
        "decoder-space nearest neighbours"
      ],
      "models": [
        "GPT-2 small (RES-JB)",
        "Gemma 2 2B (Gemma Scope res 16k)"
      ],
      "evidence": "model data from the open Neuronpedia API",
      "code": null,
      "data": "https://www.neuronpedia.org/api-doc",
      "provenance": "Built by Latent Minds Institute on the open Neuronpedia API (MIT)",
      "related": [
        "interpretability-map",
        "evaluation-state"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/instruments/latent-observatory/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/latent-observatory.json"
    },
    {
      "slug": "evaluation-state",
      "url": "/papers/evaluation-state/",
      "title": "Evaluation State in Language Models",
      "type": "working-paper",
      "typeLabel": "Working paper",
      "lifecycle": "proposal",
      "epistemicStatus": "conceptual-framework",
      "statusLabel": "Research programme · five hypotheses, experiments proposed, none run",
      "programme": "cognition",
      "question": "What evidence would show that a model represents evaluation as a shared internal state rather than reacting to surface cues?",
      "date": "2026-07-10",
      "version": "v0.1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "layer-wise probes vs lexical baselines (proposed)",
        "cross-family and cross-task transfer (proposed)",
        "activation steering with capability controls (proposed)",
        "Jacobian-lens readouts (proposed)"
      ],
      "models": [
        "open-weight instruction-tuned models (proposed)"
      ],
      "evidence": "literature synthesis and experimental framework; no original empirical results",
      "code": null,
      "data": null,
      "related": [
        "model-entrenchment",
        "interpretability-map"
      ],
      "provenance": null,
      "canonicalUrl": "https://latentmindsinstitute.com/papers/evaluation-state/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/evaluation-state.json"
    },
    {
      "slug": "model-entrenchment",
      "url": "/papers/model-entrenchment/",
      "title": "Model Entrenchment: Why Useful AI Systems Become Difficult to Remove",
      "type": "working-paper",
      "typeLabel": "Working paper",
      "lifecycle": "proposal",
      "epistemicStatus": "conceptual-framework",
      "statusLabel": "Conceptual framework · experiments proposed, none run",
      "programme": "deployment",
      "question": "When an AI system resists removal, where does the resistance live: in the model's internal representations and behaviour, or in the web of dependence around it?",
      "date": "2026-07-01",
      "version": "v1.1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "linear probes (proposed)",
        "activation steering (proposed)",
        "counterfactual environments (proposed)",
        "behavioural evaluations (proposed)",
        "case coding (proposed)"
      ],
      "models": [
        "open-weight models (proposed)"
      ],
      "evidence": "conceptual framework with falsification criteria; every empirical claim is a prediction",
      "code": null,
      "data": null,
      "related": [
        "interpretability-map",
        "gpu-credit-underwriting",
        "gpu-forward-market"
      ],
      "provenance": null,
      "canonicalUrl": "https://latentmindsinstitute.com/papers/model-entrenchment/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/model-entrenchment.json"
    },
    {
      "slug": "interpretability-map",
      "url": "/instruments/interpretability-map/",
      "title": "The Interpretability Map",
      "type": "research-map",
      "typeLabel": "Research map · instrument",
      "lifecycle": "instrument",
      "epistemicStatus": "literature-synthesis",
      "statusLabel": "Literature synthesis · sources cited per node",
      "programme": "representations",
      "question": "What has mechanistic interpretability actually established, in what order, with what dependencies, and what should a new researcher do this week?",
      "date": "2026-07-01",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "literature dependency graph",
        "chronology",
        "tiered reading pathway",
        "runnable TransformerLens/SAELens protocols",
        "model-access matrix",
        "open problems as experiments"
      ],
      "models": [
        "GPT-2 small",
        "Gemma 2 2B (protocols)"
      ],
      "evidence": "synthesis of primary sources; contains no original experimental claims",
      "code": null,
      "data": "/instruments/interpretability-map/atlas.json",
      "related": [
        "model-entrenchment",
        "transformer-explainer"
      ],
      "provenance": null,
      "canonicalUrl": "https://latentmindsinstitute.com/instruments/interpretability-map/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/interpretability-map.json"
    },
    {
      "slug": "transformer-explainer",
      "url": "/instruments/transformer-explainer/",
      "title": "Transformer Explainer: GPT-2 Live in the Browser",
      "type": "instrument",
      "typeLabel": "Instrument · adaptation",
      "lifecycle": "instrument",
      "epistemicStatus": "adaptation",
      "statusLabel": "Adapted instrument · original by Cho et al., Georgia Tech Polo Club",
      "programme": "computation",
      "question": "What does a transformer's forward pass actually compute, step by step, on real input?",
      "date": "2026-07-10",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "ONNX GPT-2 (small) inference in-browser",
        "attention map inspection",
        "temperature and sampling controls"
      ],
      "models": [
        "GPT-2 small (124M)"
      ],
      "evidence": "pedagogical instrument running a real model; the 2026 commentary is editorial, not original research",
      "code": "https://github.com/poloclub/transformer-explainer",
      "data": null,
      "provenance": "Adaptation of poloclub/transformer-explainer (MIT); previously hosted as Transformer Visualiser at mo3.ca",
      "related": [
        "interpretability-map"
      ],
      "canonicalUrl": "https://latentmindsinstitute.com/instruments/transformer-explainer/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/transformer-explainer.json"
    },
    {
      "slug": "gpu-credit-underwriting",
      "url": "/papers/gpu-credit-underwriting/",
      "title": "Underwriting the Machine: A Field Guide to GPU Credit Risk",
      "type": "working-paper",
      "typeLabel": "Working paper · economics strand",
      "lifecycle": "proposal",
      "epistemicStatus": "conceptual-framework",
      "statusLabel": "Analytical framework with illustrative models · not investment advice",
      "programme": "deployment",
      "question": "How should lenders price credit risk on GPU-collateralised debt, and what does that market structure imply for compute dependence?",
      "date": "2026-07-11",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "credit risk decomposition",
        "DSCR/LTV sizing models",
        "residual-value analysis",
        "scenario stress"
      ],
      "models": [],
      "evidence": "analytical framework with worked illustrative numbers; no proprietary deal data",
      "code": null,
      "data": null,
      "related": [
        "gpu-forward-market",
        "model-entrenchment"
      ],
      "provenance": null,
      "canonicalUrl": "https://latentmindsinstitute.com/papers/gpu-credit-underwriting/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/gpu-credit-underwriting.json"
    },
    {
      "slug": "gpu-forward-market",
      "url": "/papers/gpu-forward-market/",
      "title": "A Forward Market for GPU Compute",
      "type": "working-paper",
      "typeLabel": "Working paper · economics strand",
      "lifecycle": "proposal",
      "epistemicStatus": "conceptual-framework",
      "statusLabel": "Market-design proposal · conceptual",
      "programme": "deployment",
      "question": "What would a functioning forward market for GPU compute look like, and what would it change about the economics of AI deployment?",
      "date": "2026-07-01",
      "version": "v1",
      "authors": [
        "Latent Minds Institute"
      ],
      "methods": [
        "market design",
        "contract specification",
        "term-structure analysis"
      ],
      "models": [],
      "evidence": "market-design proposal; no live market data beyond cited public sources",
      "code": null,
      "data": null,
      "related": [
        "gpu-credit-underwriting",
        "model-entrenchment"
      ],
      "provenance": null,
      "canonicalUrl": "https://latentmindsinstitute.com/papers/gpu-forward-market/",
      "apiUrl": "https://latentmindsinstitute.com/api/research-objects/gpu-forward-market.json"
    }
  ]
}