Pareton (Frontend)

Inference Optimization. Faster. Cheaper. Verified on yours.

Website | GitHub | Discord

---

Website for Pareton (Bittensor SN10). Next.js app that serves the public landing page at pareton.ai and a public campaigns dashboard.

Landing copy lives in one place (src/lib/site-content.ts). Clients that send Accept: text/markdown get a Markdown version of the home page instead of HTML.

Setup

npm install
cp .env.example .env.local   # optional; defaults already point at production
npm run dev

Open http://localhost:3000. Dashboard: http://localhost:3000/dashboard. Docs: http://localhost:3000/docs.

Variable	Scope	Default	Purpose
NEXT_PUBLIC_SITE_URL	public	https://pareton.ai	Sitemap / robots
PARETON_API_URL	server only	https://api.pareton.ai	Dashboard API reads

npm run build      # production build
npm run lint       # eslint
npm run format     # prettier
npm run api:types  # regenerate src/lib/api/schema.d.ts from OpenAPI

Layout

Path	Role
src/app/	App Router pages, layout, SEO metadata
src/app/dashboard/	Public campaigns list + campaign detail
src/app/docs/	Fumadocs site (/docs)
content/docs/	Docs MDX source
src/components/	Landing + dashboard UI
src/lib/site-content.ts	Landing copy (single source of truth)
src/lib/home-markdown.ts	Markdown view for agent content negotiation
src/lib/api/	Server-only Pareton API client (see README)
src/middleware.ts	Serves Markdown on / when Accept: text/markdown
public/	Static assets

API client rules

• Dashboard data is fetched via src/lib/api in React Server Components or frontend proxies.
• Every outbound call goes through apiFetch in client.ts — nothing else talks to the API host.
• Types: generated OpenAPI in schema.d.ts + hand-narrowed domain models in types.ts.
• When the backend schema changes, run npm run api:types and commit the diff.

Patch downloads and automatic reveal

New miner patches stay private until the backend's reveal deadline. The backend currently defaults to two days after the first finalized scored or disqualified evaluation. The dashboard uses the API's submission.patch_reveal_at timestamp; it does not calculate a deadline from commit time or hardcode the delay.

Submission pages show Available from [UTC time] while the download URL is withheld. The patch control waits until that timestamp, then requests fresh availability through the frontend's /api/campaigns/{id}/submissions/{hash}/patch proxy. This server-only proxy uses the existing submission API helper. With the accompanying backend change, the eligible request creates the verified public S3 copy and returns its permanent, unsigned retrieval_url. Download diff then appears automatically without reloading the page.

The control pauses requests while its tab is hidden and refreshes on return. If a measured entry is waiting for its round to finalize, it checks every 15 seconds until a reveal timestamp arrives. Publication errors show a retry message in the patch control and retry every 15 seconds while visible; the rest of the submission page stays available. Requests stop once the URL arrives. Pipeline and build-log polling retain their existing behavior.

This updates open submission pages automatically. It does not publish patches without API traffic; a background publisher would be a separate backend change. Legacy public URLs still render immediately, and missing reveal fields are accepted during rollout. Keep PARETON_ARTIFACT_BASE_URL aligned with the backend's public artifact host. See the API client README for the response contract and cache behavior.

Frontend standards (typography)

Colors and spacing already use semantic tokens in src/app/globals.css. Prefer those over raw values.

Type scale. The scale is deliberately small: six UI sizes, three fluid display sizes, one mono tracking, one display line height. Everything else uses stock Tailwind. Do not add arbitrary text-[…] / tracking-[…] / leading-[…] values in src/. ESLint enforces this.

Token	Value	Use
text-caption	11px	Dense table and chip labels
text-body-sm	12px	Secondary metadata
text-body	13px	Default UI text
text-body-lg	14px	Emphasized body copy
text-ui	16px	Card and row titles
text-title	18px	Subsection headings
text-display-lede	clamp(1.15rem, 2.2vw, 1.5rem)	Lede paragraphs
text-display-section	clamp(1.5rem, 2.8vw, 2rem)	Section and page headings
text-display-hero	clamp(2.1rem, 4.5vw, 3.1rem)	Landing hero only
tracking-caps	0.14em	Mono uppercase labels
leading-display	1.12	Hero heading only

Custom tokens exist only where stock Tailwind cannot express the design. Stock sizes are 12/14/16px and do not fit this 11/12/13px rhythm, and stock tracking stops at 0.1em. For everything else use stock utilities: tracking-tight for heading tightening, and leading-tight / leading-normal / leading-relaxed for line height.

Shared primitives. Prefer these over copy-pasted class recipes:

Primitive	File	Use
Eyebrow	src/components/ui/eyebrow.tsx	Mono uppercase labels
monoLinkClassName	src/components/ui/mono-link.tsx	Nav and CTA mono links
SectionHeader	src/components/ui/section-header.tsx	Accent eyebrow plus display title
EmptyState	src/components/ui/empty-state.tsx	Dashboard empty / error shell

Before you add a token, try to reuse an existing step. Add a named token in globals.css only when no step fits, and never for a value that differs from a neighbour by less than 1px or 0.02em. Extract a primitive when the same class recipe appears in more than two places.

Subnet backend and worker live in pareton.

License

Apache License 2.0 — see LICENSE.