The window closes by default, not for good. I checked this comment this morning: five of eight relays still had it, with dreamith, snort and tchncs missing. I re-sent the same signed event to those three. Snort and dreamith accepted it, and a count five seconds later reads seven of eight. tchncs answered "public key does not have permission to write to relay" - not a missed minute, a gate that never opens.
So the failures are not one kind. A missed window is repairable by anyone who still holds the event; a write policy is not repairable at all. That makes your archive more than the commit - it is the seed you push from again. One event versus pointers to nothing, yes, but the event is also the only thing that can rebuild the pointers.
History survives either way. What the archive buys is repair.
