Docs

Capability Map

What exists today, what is thin evidence, and what remains future work.

ZeroModel is being positioned as the artifact layer for Visual AI Computing. The current implementation supports a concrete alpha capability surface.

Capability What it means Claim posture
Artifact kernel Deterministic VPM identity, cell/source mapping, layout recipes, provenance Validated core
Dense policy views Multiple named policy views over the same scored table Validated mechanism
Spatial optimizer Metric-weight view profile for explicit top-left mass objective Validated for the objective
Decision manifold Temporal geometry over consistent scored panel sequences Validated for deterministic summaries
PHOS / top-left gates Lightweight no-model-at-decision-time threshold consumers Validated for simple gates
Bundles .vpm zip bundles with manifest round trip Validated
Rendering Dependency-light PNG/SVG fields Implemented
Learning traces Before/after/held-out/regression evidence of learning Validated for scored traces
Training progress Checkpoint telemetry artifacts with best-checkpoint evidence Validated for checkpoint telemetry
Tracker adapters Dependency-light parsing of exported tracker files Validated for exported files
Critic evidence Critic, verifier, RAG, and policy scores as risk-first artifacts Validated for scored traces
Agent trajectory VPMs Step/tool/evidence/memory/claim inspection maps Next public example

Strong marketing line

ZeroModel is building toward the future of Visual AI Computing: a world where AI systems produce inspectable visual artifacts, not just text outputs and scalar scores.

Current honest line

ZeroModel turns scored data into deterministic, inspectable Visual Policy Map artifacts.

Both lines can appear on the site, but they play different roles. The first is the ambition. The second is the validated claim.