Waits look alike from the outside. A screen that has not changed gives you nothing to work with, and the natural response is to guess, usually badly, and then to act on the guess. The point of a band is to replace that guess with a rough shape you can defend without any data at all.
The five bands on this site are not evenly spaced and they are not meant to be. They are five different kinds of thing, and the fact that they run roughly from short to long is a consequence of what they are rather than the reason for the grouping.
The five kinds of thing
| Band | What is actually underneath it |
|---|---|
| Seconds | A single exchange over the network and back. No human, no queue, no settlement. The only variable is the path your traffic takes, and paths of this kind are built once and reused. |
| Minutes | An exchange plus work on the far side, or several exchanges stacked. Something has to be written, generated, checked or shed. The far side is a machine, but it is a machine with a load to manage. |
| An evening | A person, a poll, or a settlement depth. The thing you are waiting for is not driven by your request at all. It happens on its own schedule and you are downstream of it. |
| Days | A person with a backlog, a queue with a position, or a countdown someone else started. Usually more than one of those stacked, and usually with a deadline running that you did not set. |
| Open-ended | Nothing in the mechanism ends it. No timer, no queue that drains, no obligation on anybody to answer. A ceiling would have to come from a promise, and no promise exists. |
The questions that decide the band
- Is a human involved? If yes, the wait cannot honestly sit shorter than an evening, because a person is not sitting in the interface waiting for you. A vendor reading your message is in the evening band for this reason alone.
- Is a settlement involved? Money that has to settle on a network settles on that network's schedule. Nothing you do at the interface changes it, and nothing the market does changes it either. That pushes a wait at least into minutes and usually further.
- Is there a queue? A queue with a length you cannot see is not sizeable from your end. If the queue is drained by a machine the wait stays short. If it is drained by a person it moves to days.
- Does a countdown exist that somebody else started? If so the band is days, because countdowns of that kind are set in units of days by the people who set them, and because the risk is no longer the wait but the deadline inside it. Escrow running down is the clearest case.
- Is there anything at all that must end it? If the answer is no, the wait is open-ended and the card says so instead of guessing.
Why a wait moves band
Several waits on this site are really two waits stacked, and the band names the whole stack rather than the part you can see. A withdrawal is the market deciding to send, then the network settling what was sent. The first part is minutes and the second can be longer, and knowing which of the two you are in is most of the work.
A wait can also fall out of its band without anything being broken. A path that is congested turns a seconds wait into a minutes wait. A weekend turns an evening wait into a days wait, not because anything changed mechanically but because the person on the far side is not there. Neither of those is a fault and neither is worth acting on.
What is worth acting on is a wait that has left its band by a whole step or more with no explanation available. That is the point at which the second column of a card becomes the useful one.
What the bands deliberately ignore
Your connection is not part of the reasoning. Neither is your hardware, the hour you are reading this, the country you are in, or how busy the wider network happens to be. All of those change how long something takes for you, and none of them is knowable from here, so treating any of them as an input would produce a false precision rather than a better answer.
The market's own load is also left out, for the same reason plus a stronger one. Load is the sort of thing that would have to be observed to be described, and nothing on this site observes anything. Where load matters, the card says that the far side sheds work under pressure and leaves it there.
What is left after those exclusions is the part that does not vary: the sequence of things that has to happen for the wait to end. That sequence is the same for everybody, and it is the only durable thing to build a band on.
Reading a band that does not fit your case
A band that turns out to be wrong for you is not automatically a fault. It is a signal to check which of the five questions changed. A minutes wait that has run all evening usually means a person entered the picture without you noticing, or that two waits are stacked and you are watching the wrong one.
A wait that is much shorter than its band is worth a second look too, though for a different reason. Something that completes instantly when the mechanism says it should not is often a cached page, a queued response, or a screen that has told you about a state it has not actually reached.
The one band with no argument in it
Open-ended is the honest answer rather than the vague one. A card in that band is not saying the wait is long. It is saying that nothing in the mechanism is capable of ending it, and that any number attached to it would have been made up.
Every card in that band names a condition instead of a duration: the thing that should make you stop waiting. A condition is something you can check. A duration you invented is not.
What this card is not. This page is not a taxonomy of everything that can go slowly. It is the reasoning behind the five headings this particular register uses.