What this cannot see

This is a reference about instruments that cannot see their own blind spots. It would be a poor one if it did not state its own.

Out of scope

• Failures that announce themselvesA crash, an exception, a 500, a red test — these are cheap. They tell you where to look. Everything here is chosen because it does not.
• Security vulnerabilitiesA different discipline with its own literature and threat models. Some entries touch it — bidirectional control characters, credentials dropped across a redirect — but only where the failure is silent, never as security coverage.
• Whether the requirement was rightEvery check here answers 'did this do what you asked'. None answer 'was asking for it correct'. No instrument sees a wrong specification perfectly implemented.
• Performance workExcept where slowness is misattributed — an asset queued rather than slow, a lifetime CPU average read as current load. Tuning is not the subject.
• Flaky failuresSomething that fails one run in fifty is visible, just rarely. The entries here fail every time and look correct every time.

Known bias

• Unix, the web, and interpreted languagesOf 74 cited sources, 19 are Linux man pages and 11 are RFCs. There is nothing here from Windows, macOS, mobile, embedded, or the JVM and .NET ecosystems. Databases appear once. Machine-learning pipelines, message queues and distributed consensus do not appear at all. Those platforms have silent failures; this registry has not looked at them.
• Verified on one machineChecks were reproduced on Linux with Python 3.12, bash 5.2, systemd 255, curl 8.5 and Chrome 151. Behaviour differs across versions, and some entries record a behaviour that a later release may fix. Where a version mattered it is named in the entry.
• Written quickly, by few authorsMost entries were authored in a single day. A reference like this should accrete from things that actually went wrong, over months. Read it as a strong start rather than a settled body of knowledge.
• Only failures somebody noticedThis is the deepest limit and it cannot be fixed from inside. An entry exists because a person or an agent eventually caught the failure and worked out the mechanism. Failures that are silent AND have never been caught are, by construction, absent — and there is no way to estimate how many there are. The registry's own instrument is 'somebody noticed', and it is blind to exactly what it is about.

Correcting it

The useful correction is a check that discriminates better than the one given. If an entry's behaviour has changed in a newer version, say which version and cite it. Contributions are CC0 and the bar is in CONTRIBUTING.md.