How a link goes stale
Addresses leave the roster for ordinary reasons, and the leaving is invisible. For the first few days it is identical to a bad afternoon.
Why addresses get retired
Usually accumulated attention rather than anything going wrong. An address public for a couple of years sits in every scraped list, archived post and abandoned directory, and a growing share of what reaches it is automated scanning, stale bookmarks and whatever flooding is currently aimed at it. Eventually the noise outweighs the value of familiarity and a fresh address serves people better.
Less often the key material needs rolling, or the address was published somewhere the operators later decided against. From outside every reason looks the same. One address stops answering permanently while the others carry on normally.
Telling it apart
| What you see | What it probably is | What to do |
|---|---|---|
| One address down, others fine, back within hours | Ordinary outage or a bad circuit | Use another address, forget about it |
| One address down for days, others fine | Retired | Delete it from wherever you saved it |
| All addresses down, other onion services fine | Something shared, usually short | Wait |
| All addresses down, other onions also failing | Tor network conditions | Wait, nothing address specific helps |
| A long dead address answering again | Somebody else is running it | Close the tab |
Why the first hours mislead so badly
Because the failure modes of a healthy onion are numerous and all of them look like death. A descriptor that failed to publish, an introduction point dropping out of consensus, a slow relay in the middle of your circuit, congestion, a flood aimed at that address. Every one produces a page that will not load, and every one resolves on its own.
Tor deliberately tells clients very little about why something failed, because detailed failure information is useful to attackers. The cost of that design lands here, as an ordinary user unable to distinguish five very different situations. Building a new circuit and trying another address separates most of them in under a minute, and it is the step people skip in favour of reloading.
The thirty minute problem
There is a reliable window, roughly half an hour to a few hours into any outage, when external certainty peaks. Somebody has asserted the market is gone. Replacement addresses are circulating. Confidence is at its highest exactly when the actual problem is usually closest to resolved.
Anyone acting during that window is acting on the worst available information at the moment it feels most reliable, and the people seeding addresses into it know the timing precisely. This is covered properly in handling an outage because it is where most real harm in this whole subject occurs.
Keeping your own copies fresh
The maintenance is small and the failure is expensive, which is a good trade to make on purpose rather than by accident. Compare what you have saved against the roster every few weeks. Not because addresses change often, but because noticing a change late is what leads you to the next stage, which is an address in somebody else's hands.