docs: org about + brand icon

README.md

@@ -6,34 +6,25 @@ tags:
 
 # DotCheck
 
-**See what's real.**
+Organization for DotCheck **closed-weight** AI-likeness engines. Public model repos are **documentation cards** (architecture, gates, license). Inference heads (`.npz`) are not published on public Hub repos.
 
-DotCheck scores how AI-like **images**, **video frames**, and **text** look — so people can tell synthetic media from human-made content. Scoring runs on DotCheck infrastructure (not on-device). Closed weights: public Hub pages describe engines and measured gates; inference heads are not redistributed.
+| Hub id | Wire | Backbone / stack | Artifact |
+|--------|------|------------------|----------|
+| [`vermeer-image-v5`](https://huggingface.co/DotCheck/vermeer-image-v5) | `inhouse@5` | SigLIP2-base + linear head | `…_linear_head_v5.npz` |
+| [`valla-text-v9`](https://huggingface.co/DotCheck/valla-text-v9) | `inhouse-text@9` (+ lang heads) | TMR + Fakespot + logistic heads | `text_stack_head_v9.npz` + lang |
+| [`muybridge-video-v2`](https://huggingface.co/DotCheck/muybridge-video-v2) | `inhouse-video@2` | SigLIP2-base + video head | `…_video_v2.npz` |
 
-**Leviathan** is DotCheck's product scoring system: in-house engines (Vermeer · Valla · Muybridge) + generator-surface prior + shared content-hash cache. Same path for Free, Premium, and Pro — not a tier upsell.
+- Collection: [dotcheck-engines](https://huggingface.co/collections/DotCheck/dotcheck-engines-6a659e57f7934cb510677785)
+- About: [`DotCheck/dotcheck`](https://huggingface.co/DotCheck/dotcheck)
+- Private vault (owner): [`heads-live-private`](https://huggingface.co/DotCheck/heads-live-private)
 
-## Engines (this org)
+## Runtime
 
-| Label | Modality | Wire | Hub model (docs only) |
-|-------|----------|------|------------------------|
-| **Vermeer** | Image | `inhouse@5` | [`DotCheck/vermeer-image-v5`](https://huggingface.co/DotCheck/vermeer-image-v5) |
-| **Valla** | Text | `inhouse-text@9` (+ lang heads) | [`DotCheck/valla-text-v9`](https://huggingface.co/DotCheck/valla-text-v9) |
-| **Muybridge** | Video frames | `inhouse-video@2` | [`DotCheck/muybridge-video-v2`](https://huggingface.co/DotCheck/muybridge-video-v2) |
+Serve: DotCheck CPU FastAPI. Public HTTP: Express (Heroku) → inference. Guests: https://dotcheck.ai/check. Programmatic: https://dotcheck.ai/api (`Authorization: Bearer dc_…`, Pro). Gate tables: https://dotcheck.ai/docs · https://dotcheck.ai/how
 
-Collection: [DotCheck Engines](https://huggingface.co/collections/DotCheck/dotcheck-engines-6a659e57f7934cb510677785).
+Do not `from_pretrained` DotCheck head repos for runnable weights.
 
-Public claim floats are lockstep with product facts (`Data.json` / MODEL-CHOICE). Do not invent accuracy numbers from bake-off shortlists.
+## Metrics SSOT
 
-## Product links
-
-- How it works: [https://dotcheck.ai/how](https://dotcheck.ai/how)
-- Docs (gates & method): [https://dotcheck.ai/docs](https://dotcheck.ai/docs)
-- Check (try it): [https://dotcheck.ai/check](https://dotcheck.ai/check)
-- Pro API: [https://dotcheck.ai/api](https://dotcheck.ai/api) (keys in Dashboard)
-- Site: [https://dotcheck.ai](https://dotcheck.ai)
-- Org about repo: [`DotCheck/dotcheck`](https://huggingface.co/DotCheck/dotcheck)
-
-## Weights policy
-
-DotCheck heads are **closed**. Hub model repos in this org publish **cards and license text only** — no `.npz` / checkpoint files on public repos. Use Check or the Pro API; do **not** call `from_pretrained` on DotCheck head repos expecting runnable weights.
+Measured holdout floats live in product `Data.json` / `MODEL-CHOICE.md` (image bal **0.9875**, EN text bal **0.980**, video bag-p90 bal **0.985**). Do not cite stale bake-off YAML.