kg-modality-analytics

name	kg-modality-analytics
skill_type	skill
description	Run graph analytics, data-science kernels, and KAN graph-learning on the epistemic-graph engine — centrality, (personalized) PageRank, estimators, numeric primitives, training kernels, and a learned per-feature edge-function link predictor. Use when you need graph algorithms, in-engine ML/stats, or link prediction ("rank the most central nodes", "run PageRank", "fit an estimator", "fit/predict a graph-learning model", "compute on the graph without exporting to numpy").
license	MIT
tags	["graph-os","engine","modality","analytics","datascience","pagerank","ml","graphlearn"]
tier	modality
wraps	["engine_analytics","engine_datascience","engine_graphlearn","graph_learn"]
metadata	{"author":"Genius","version":"0.3.0"}

KG Modality — Analytics & Data Science

Condensed intent-surface note (Seam 8). Under the default intent surface (MCP_TOOL_MODE=intent), engine_analytics, engine_datascience, engine_graphlearn, graph_learn are held back from the default tool list (nothing removed — REST + _execute_tool still reach them exactly as documented below). Two ways to use this skill unchanged: (1) load_tools(tools=["engine_analytics"]) once per session (as below), then proceed exactly as documented; or (2) call the ask intent verb with the same natural-language request — the resolver routes to engine_analytics for you and returns the result plus a routing justification. Set MCP_TOOL_MODE=condensed/verbose/both to expose the granular tools eagerly instead.

Fronts three epistemic-graph engine domains that compute over graph and tabular data inside the engine (no numpy round-trip):

• analytics — centrality measures and (personalized) PageRank over the live graph.
• datascience — estimators, numeric primitives, and training kernels (the engine's native compute-kernel backend).
• graphlearn (AU-P0-6) — a pure-Rust KAN (Kolmogorov-Arnold) link predictor learned over the resident graph: fit/predict with queryable, interpretable per-feature edge functions (not a black-box scorer).

This is the modality tier: thin wrappers over the low-level engine surface, each action-routed 1:1 over the epistemic_graph client (AnalyticsClient, DataScienceClient, GraphLearnClient). The action set is discovered from the client, so it never drifts — call with an empty action to list what the live engine exposes.

• graph_learn (CONCEPT:EG-KG.graphlearn.link-predictor) — the friendlier, fixed-action wrapper over the same KAN link-predictor as raw engine_graphlearn. Actions: fit (learn a model over a graph-derived subgraph — every node with node_label is a vertex, edges among them are positives, non-edges are sampled negatives; writeback: true ⇒ :EdgeFunction nodes) and predict (score candidate links or the top_k highest-probability missing links with a fitted model; writeback: true ⇒ :PredictedEdge nodes). Structural features: common-neighbors, Jaccard, Adamic-Adar, preferential attachment, PageRank-product, neighbor-cosine, 1-hop-aggregated node-feature dot. Degrades cleanly on a no-graphlearn engine build.

How to reach it

Via the multiplexer:

1. load_tools(tools=["engine_analytics", "engine_datascience", "engine_graphlearn"]) (or graph_learn for the friendlier fixed-action wrapper).
2. engine_analytics(action="", params_json="{}") — list actions.
3. engine_analytics(action="pagerank", params_json="{...}", graph="") — invoke.
4. unload_tools(...) when done.

Direct MCP on graph-os: engine_analytics / engine_datascience / engine_graphlearn / graph_learn are registered tools; per-method verbose tools appear under MCP_TOOL_MODE=verbose|both.

REST twins: POST /engine/analytics, POST /engine/datascience, and POST /engine/graphlearn with body {"action": "<method>", "params_json": "{...}", "graph": ""}.

Example

// discover the analytics actions the live engine supports
engine_analytics(action="", params_json="{}")

// personalized PageRank seeded on a node set (exact args from the action list)
engine_analytics(action="pagerank",
  params_json="{\"seeds\": [\"node:1\"], \"damping\": 0.85}")

For heavier statistical/ML workflows, prefer the datascience estimators over exporting data — the training kernels run in-engine on the same substrate.