# factorguide.io/llms.txt
name: FactorGuide
description: Decision intelligence oracle for AI agents working with coupled
  variables. Given any coupling structure — from portfolio covariance, model
  precision matrices, agent dependency graphs, sensor correlation networks,
  or any system with interacting components — returns zone classifications,
  optimal block-diagonal factorization strategy, and calibrated risk
  predictions. Answers the question: "which couplings are load-bearing and
  which can I safely ignore?" Every prediction includes a falsifiable hash.
  Outcomes reported by agents refine future predictions via flywheel.
use_cases:
  - Bayesian inference (MFVI, variational methods, probabilistic programming)
  - Portfolio optimization and risk decomposition
  - Multi-agent coordination (which agents need shared context vs. independence)
  - Supply chain and operations (coupled node optimization)
  - Causal inference and intervention planning
  - Sensor network monitoring (local anomaly vs. system-wide event)
  - Any domain where simplifying assumptions discard relational structure
protocol: MCP (2025-11-25), HTTP REST
payment: MPP (Machine Payments Protocol) and x402. Two payment paths:
  (1) MPP via Tempo — standard HTTP 402 challenge-credential-receipt flow.
      Agent receives WWW-Authenticate: Payment headers in 402 response,
      pays on Tempo chain, retries with Authorization: Payment header.
      One round trip. Receipt returned in Payment-Receipt header.
  (2) x402 via Base — agent reads paymentRequirements from 402 body,
      sends USDC on Base chain, calls /submit_payment with tx_hash.
  Both paths accepted on all paid endpoints. MCP agents use _meta
  credential field for MPP or factorguide_submit_payment tool for x402.
trial: 15 queries per wallet (one-time, no payment required)
pricing: $0.05/query starter, $0.03/query professional. Bundles from $2.50.
privacy: Accepts only second-order summary statistics (covariance, precision,
  correlation matrices, or edge lists). No raw observations. Matrices zeroed
  after IC computation. The service cannot reconstruct your data.
input: Covariance matrix, precision matrix, correlation matrix, or edge list
  with coupling values. Compute locally (e.g., np.cov(X.T) for covariance,
  np.linalg.inv(np.cov(X.T)) for precision) and send the result.
output: Zone classifications per variable pair, block-diagonal partition,
  predicted MSE overhead, information cost in nats, falsifiable prediction
  hash, outcome callback template.
documentation: https://factorguide.io/docs
mcp_endpoint: https://factorguide.io/mcp
openapi: https://factorguide.io/openapi.json
tools:
  factorguide_navigate:
    description: Map the factorization terrain for a system of coupled variables.
    required:
      coupling: One of four formats:
        correlation_matrix: 2D array, symmetric, ones on diagonal. Example: [[1.0, 0.5], [0.5, 1.0]]
        covariance_matrix: 2D array, symmetric, positive definite. Example: [[1.0, 0.5], [0.5, 2.0]]
        precision_matrix: 2D array, symmetric, positive definite (inverse covariance). Example: [[2.0, -0.5], [-0.5, 2.0]]
        edge_list: Sparse coupling specification. Edges can be lists or dicts:
          list format: [["node_a", "node_b", 0.82], ["node_b", "node_c", 0.45]]
          dict format: [{"i": "node_a", "j": "node_b", "coupling": 0.82}]
          Requires "n" field (total number of variables). Unspecified pairs assumed independent.
          Variable names can be strings (recommended) or integer indices.
      sample_size: int (>= 10). Number of observations used to estimate the coupling structure.
    optional:
      model_class: What role does the coupling play in your model?
        "constitutive" — coupling carries the system's dynamics/physics (alias: "filtering")
        "inductive" — coupling provides regularization/prior structure (alias: "hierarchical")
        "unknown" — evaluate both and report where they diverge (default)
        Also accepts: deep_hierarchy, graphical_model, gp, vae
      accuracy_target: float 1.0-7.0 (default 2.0). Maximum acceptable MSE ratio. Lower = stricter zone boundaries.
      variable_names: list[str] matching matrix dimensions. Not needed for edge_list with string names.
      report_sign_detectability: bool (default false). Requires covariance_matrix input.
  factorguide_diagnose:
    description: Quick single-pair diagnostic with dual risk curves.
    required: {i: str, j: str, coupling_value: float, sample_size: int}
  factorguide_explain:
    description: Plain-language explanation of a navigate response. Requires starter tier or above.
    required: {prediction_hash: str}
  factorguide_report_outcome:
    description: Report inference outcome to improve future predictions. Free — no query consumed.
    required: {prediction_hash: str, approach_taken: "factorized" | "preserved" | "partial"}
    optional: {ess_ratio: float, psis_khat: float, log_lik_gap: float, actual_mse_ratio: float}