Sockrates 0.5.0 — UDP relay test

SOCKS5 proxies can now be tested for real UDP relay (--udp), and reliability is recency-weighted so a proxy that died today scores low.

Sockrates 0.4.0 — anonymity classification

Each proxy is now graded transparent / anonymous / elite (--anonymity, --only-elite). Plus Type and Anonymity columns in the app.

Sockrates 0.3.0 — all proxy types

SOCKS5, SOCKS4 and HTTP/HTTPS proxies are now all supported (--type). More sources, curated by measured hit rate.

Sockrates 0.2.3

Clearing the scan range now falls back to the public lists instead of demanding a range.

Sockrates 0.2.2 — typeable scan/custom fields

Fixed

• The scan Range/Ports and custom host:port fields could not be typed into. They were
  disabled until you first picked the matching radio button, and a disabled field looks exactly
  like an active one — a dead black box. The fields are now always typeable; clicking or tabbing
  into one selects its mode for you.

Sockrates 0.2.1 — settings focus fix

Fixed

• Could not type in the settings fields on some Linux window managers. A click did not hand
  keyboard focus to the input, so the fields looked dead. Any click on a field now claims focus,
  and the window claims keyboard focus on launch.
• The numeric settings (workers, timeout, latency, refresh interval) now reject non-numeric
  input instead of silently accepting text that would break the next run.

Sockrates 0.2.0 — scan mode

Added

• Scan mode (--scan): discover proxies by scanning a CIDR / range / host you are
  authorised to test, instead of reading public lists. Every open port is then verified like
  any other candidate. Capped at a /16 per call, and the desktop app confirms before it starts.
• The desktop app grew a header with the wordmark, zebra-striped result rows, and the scan
  option in the Sources tab.

Fixed

• The scan port-knock now reads the SOCKS5 greeting with an exact-length read, so a reply
  split across packets is no longer mistaken for a refusal.

Sockrates 0.1.0

First release.

The idea: finding proxies is trivial — public lists hand you thousands in under a second.
Keeping the ones that work is the entire problem, so nothing here is taken on the proxy's word.

• TLS targets are proven end to end, certificate checked through the tunnel
• Telegram datacenters get a real MTProto handshake (req_pq_multi → resPQ with our
  nonce). This cut a naive TCP-connect result of 194 down to the 115 that were true — 41%
  were lying
• every candidate is also asked to connect somewhere unroutable; anything claiming success
  there is discarded

Telegram is two targets, not one. Of 230 working proxies, only 42 served both the Bot API
and MTProto. Test the one you will actually use.

What it hands you is alive: results are re-verified before every save or copy, --watch
keeps a file permanently true, and a local record tracks how long each proxy has been known to
work and how often it held up.

Runs in the terminal and as a Tkinter desktop app. No third-party dependencies.