Capvise Labs
VetoBench

Systems

Systems are where claims get expensive.

Capvise Labs currently has one public research system in development and one applied venture. Nothing else is being presented as a released program.

Public evaluation systemProtocol development

VetoBench

VetoBench will test how often generative-video models refuse safe work, fail without a useful explanation or behave differently across repeated attempts.

Benign completionRepeated-run stabilityVideo extensionRefusal calibration

No scores have been published. The pilot protocol, access covenant and founding-model program are being prepared.

Open VetoBench
Applied venturePilot stage

Marka

Marka is being built to run human data annotation and model-evaluation work for AI teams. It has a separate commercial site and operating model.

Data annotationHuman evaluationAgent reviewQuality operations

Paid work through Marka will not buy placement, suppress results or alter a VetoBench score.

Visit Marka ↗

Project labels

Research system and applied venture mean different things here.

Research system

Public methods, inspectable results and no paid influence over the outcome.

Applied venture

A separate product or service with customers, delivery work and commercial goals.

Shared origin

Both may use ideas and engineering developed inside Capvise Labs.

Hard boundary

Buying work from Marka will not buy or alter a public VetoBench score.