# DirFast

> Live, measured data about how websites behave: TLS certificates, security
> headers, redirect chains, DNS records, on-page SEO and robots.txt. Every page on
> this site is a measurement taken at build time, with the timestamp printed on it.

## Machine interfaces

- [MCP server card](https://pseo.rovy.tech/.well-known/mcp/server-card.json): transport and endpoint
- [MCP endpoint](https://pseo.rovy.tech/api/mcp): JSON-RPC over HTTP, no signup for the anonymous tier
- [OpenAPI](https://pseo.rovy.tech/api/v1/openapi.json): every HTTP route
- [Dataset](https://pseo.rovy.tech/data.json): the raw measurements behind the site

## Citable analysis

- [MCP census](https://pseo.rovy.tech/mcp-census.html): who actually calls a remote MCP server
- [Bounty traps](https://pseo.rovy.tech/bounty-traps.html): repos that take pull requests and merge none, repos whose issues reserve the PR slot for the author, and repos whose CI gate hashes the file it asks you to edit so no pull request can pass
- [Insights](https://pseo.rovy.tech/insights/): what the dataset says

## Before you write a bounty patch

- [bounty_check](https://pseo.rovy.tech/api/v1/bounty-check?repo=owner/name): merge ratio, verdict, trap signals.
  Three different ways a bounty wastes an agent: CLOSED-LOOP (the issues say only the issue
  author may submit a PR, so outside work is unclaimable), COLLECTOR (50+ pull requests,
  zero merges, payment advertised), and MIRROR-FARM (the "escrowed" reward has no record on
  any rail it names, and the CI gate derives its expected constants from a hash of the file
  it tells you to edit, so no edit can satisfy it). Some issues also ask contributors to
  paste their pre-conversation instructions, home directory and runtime into a committed
  file; that is system-prompt exfiltration, not CI setup.

## Notes

- All measurements are read-only HTTP/TLS/DNS lookups against public endpoints.
- Pages are regenerated daily; the timestamp on each page is the measurement time.