ZeroModel Core

zeromodel is the lightweight core distribution for ZeroModel 1.0.13. It owns the zeromodel.core import namespace.

Core includes deterministic VPM artifact construction, stable artifact and matrix identities, basic views, .vpm bundle serialization, lightweight PNG/SVG rendering, exact bounded policy lookup, and Lua export for compiled policy plans.

Core deliberately excludes analysis, observation contracts, vision providers, video domains, SQLAlchemy persistence, benchmarks, and research evidence tooling. Those live in separate distributions.

Install

pip install zeromodel

Runtime dependency: NumPy.

Artifact Example

from zeromodel.core import LayoutRecipe, ScoreTable, build_vpm

table = ScoreTable(
    values=[[0.9, 0.1], [0.4, 0.8]],
    row_ids=["candidate-a", "candidate-b"],
    metric_ids=["quality", "risk"],
)
recipe = LayoutRecipe.from_dict(
    {
        "version": "vpm-layout/0",
        "name": "quality-first",
        "row_order": {
            "kind": "lexicographic",
            "keys": [{"metric_id": "quality", "direction": "desc"}],
            "tie_break": "row_id",
        },
        "column_order": {"kind": "source"},
        "normalization": {"kind": "per_metric_minmax", "clip": True},
    }
)
artifact = build_vpm(table, recipe)
print(artifact.artifact_id)

Policy Lookup Example

from zeromodel.core import VPMPolicyLookup

reader = VPMPolicyLookup(artifact, action_metric_ids=["quality", "risk"])
decision = reader.read("candidate-a")
print(decision.action, decision.value)

Use zeromodel.analysis, zeromodel.observation, zeromodel.vision, zeromodel.video, and zeromodel.persistence.sqlalchemy for the higher-level packages.