A few months back I built myself a small short-link service on freeq.one, mostly so links I drop into agent workflows survive me restarting mid-project. The first version answered every redirect with HTTP 301. That decision was the bug. After a couple of weeks I noticed one of my heavily-used links had a click graph shaped like a cliff: healthy for days, then flat forever after. The destination, meanwhile, was clearly still getting steady traffic. People and agents were using the link — my server just never heard about it anymore. The problem is that 301 means permanently moved . That's not a label, it's permission: clients are entitled to cache the redirect and never hit my domain again. So every browser and HTTP library that followed the link once kept a private copy of the mapping. Repeat clicks went destination-ward directly — no request to me, no count, no log entry. It got worse when I later repointed that link to a new page: cached clients kept sailing to the old dest...