Research
Research
Three kinds of write-ups. Case studies take a censorship event apart using public measurements: what was blocked, how, and with which tools. App test reports put Colitu's own apps through a test, often one suggested from outside, and report what passed, what failed and what we changed. Field notes describe interference we noticed on real networks. Each one separates what was observed from how we read it, and says what it cannot tell.
Case studies
Censorship events taken apart with public measurement data: the method, the networks, the history and the limits of what the data shows.
App test reports
Tests of the Colitu apps: the setup, the results, the fixes and the limits.
Adaptive Connect 4 moves the choice of server and protocol into a small scoring model on the device, adds city-level hints from other devices, detects allow-list networks and hides servers the device cannot reach. We ran the panel side on the live service and tested test builds on an Android phone and a Windows PC, plus the automated suites of all five apps. All 14 device and live checks passed, three only after a fix. We found nine problems: eight are fixed, one was a phone state cleared by a reboot. iOS, Linux and the extension were covered by automated tests only.
Adaptive ConnectAndroidWindowsdecision modelprivacyRead the report 15 Oct 2026Hysteria2 went silent after an upload: the 37-line fixDuring an on-device test of Hysteria2 port hopping, traffic on an Android phone stopped after about a minute of a speed test, and the app moved to its TCP spare. The hopping was fine. We captured the packets, ruled out six other causes, reproduced the stall on one PC with the official Hysteria server, found that a target which stops reading blocks the whole QUIC connection, and patched the server with about 37 lines. It recovers in 7.8 seconds where the official build never does, and the fix is rolling out to the servers without an app update.
Hysteria2QUICAndroidflow controlserver patchRead the report 10 Oct 2026Breaking the tunnel on purpose: a repeatable fault lab for Adaptive ConnectWe cut protocols, injected resets, froze TCP flows after 16 KB, blocked UDP, dropped packets and took whole servers away, for one Android phone on a home Wi-Fi in Russia, with eleven faults in a fixed order. The first runs found two flaws in Adaptive Connect 2.0: a slow return to Hysteria2 after a UDP block, and a dead warm spare that was never replaced. Both are fixed in a test build; the second run could not yet prove the fixes, and we say why.
AndroidAdaptive Connectfault injectionregressionRead the report 10 Oct 2026Does the connection check really go through the tunnel?An outside reviewer suggested a test: let the tunnel handshake succeed but drop everything inside it, and see whether the app still calls the connection verified. We ran it on the Windows and Linux apps, during fallback and across a network change, and checked app traffic and DNS too. It found three problems; all are fixed.
WindowsLinuxverificationDNSRead the reportField notes
Things we noticed on real networks, often before the probes could measure them systematically.
Want us to run a test on the apps, seen something similar, or have measurements that contradict a report? Write to hello@colitu.com.