DotCheck

Organization for DotCheck AI-likeness people. Public model repos ship Apache-2.0 open weights (.npz heads on frozen upstream backbones), architecture notes, and measured holdout gates. Product video assembly Covenant (Muybridge frames + Helmholtz audio) stays with DotCheck. Everyday scoring still runs through DotCheck.

Hub id	Wire	Backbone / stack	Artifact
vermeer-image-v12	inhouse@12	Adapted SigLIP2-base + linear head	siglip2_base_patch16_224_linear_head_v12lora.npz
valla-text-v9	inhouse-text@9 (+ lang heads)	TMR + Fakespot + logistic heads	text_stack_head_v9.npz + lang
muybridge-video-v2	inhouse-video@2	SigLIP2-base + video-frame head (Muybridge)	siglip2_base_patch16_224_linear_head_video_v2.npz
helmholtz-audio-v2	inhouse-audio@2	Dasheng-Base + logistic head	dasheng_base_linear_head_audio_v2.npz

• Collection: DotCheck Engines
• About: DotCheck/dotcheck
• Full inventory: see REGISTRY.md in this about repo / hub/REGISTRY.md in source

Runtime

Serve: DotCheck CPU FastAPI. Public HTTP: Express → inference. Guests: https://dotcheck.ai/check. Programmatic: https://dotcheck.ai/api (Authorization: Bearer dc_…, Pro). Gates: https://dotcheck.ai/docs · https://dotcheck.ai/how

Public repos include Apache-2.0 open weights. Everyday product scoring still runs through DotCheck.

Metrics SSOT

Measured holdout floats live in product Data.json / MODEL-CHOICE.md (image bal 0.9950, EN text bal 0.980, video bag-p90 bal 0.985, audio bal 0.986). Do not cite stale bake-off YAML.