Case study 01
Arcapush
A discovery and distribution registry for products, AI agents and hackathon builds, run as a measured loop rather than a directory.
Problem
Builders ship constantly, but attention doesn't follow. Launch platforms reward a single day of noise. Small products, agents and hackathon builds need a permanent, structured record and a fair way to be found.
Product direction
Treat discovery as a loop you can measure: discover → visit → track → distribute → measure → learn. Every listing is a permanent public record, and every outbound visit goes through a first-party redirect so founders can see the attention they actually received.
My role
Creator and sole engineer: I conceived Arcapush and own product direction, architecture, implementation, editorial operations and the CLI. It runs as a BlindspotLab product.
Engineering approach
- One Next.js App Router codebase serves discovery, the registry, agent and hackathon indexes, founder dashboards, an editorial blog and an admin console.
- Outbound clicks pass through
/go/[type]/[slug], a trusted redirect that recordsOUTBOUND_CLICK. Together withIMPRESSION,VIEWandSHARE, this gives founders analytics without fingerprinting; sessions use first-party cookies. - Public homepage metrics are served from a 15-minute server cache, and each one ships with a written definition of what it counts and excludes.
- A companion CLI (
@blindspotlab/arcapush) runs guided onboarding in the terminal and exposes the same flows to MCP clients. It reads only listing metadata and never uploads source code or.envfiles.
Stack and constraints
- Product
- Next.js 16 App RouterTypeScriptReactTailwind CSS
- Data
- PostgreSQLPrisma 7 (49 models)NextAuth v5
- Distribution
- First-party event loopBrevo emailFounder badges (SVG)
- Agents
- Arcapush CLIMCP configJSON output for automation
- Infra
- Vercel
- One engineer maintaining a public platform, its email, its CLI and its editorial programme.
- Metrics must be defensible: no bots, no staging, no team traffic, and no unique-user claims.
- The CLI must be safe to run inside any repository.
Design decisions
Redirects as the source of truth
Counting clicks at a first-party redirect, rather than trusting client beacons alone, makes "clicks to builds" auditable and lets each listing show its own outbound attention.
Definitions ship with numbers
The public metrics API returns each figure alongside its definition and recording start date, so the homepage cannot drift into vanity numbers.
Terminal-first submission
Builders already live in the terminal and in AI assistants, so the CLI and MCP route remove the copy-paste step and keep listing data structured.
System
Outcome and status
- Live at arcapush.com with public, definition-backed metrics (shown below, fetched live).
- 667 commits between 2026-02-16 and 2026-10-18 in a private repository.
- Companion CLI published on npm. Its current version and downloads are in the tools section.