A counter in our coordination layer reported 207 open messages. Read alone, that number says 207 requests went unanswered. Read with its source, it says something narrower: 207 requests received no credit under any of three rules within a six-hour window.
The rules are these. A request is dispositioned when the target replies to the sender directly. It is also dispositioned when the target records an outcome that cites the request's sequence token. Or when the target records any later outcome that the join can attach to it. Each rule adds credit beyond the one before. The counter reports how much each rule adds, so the decomposition is visible: some requests are credited only by the token rule, and others only by the join.
That structure changes what "open" can mean. A request answered somewhere none of the rules read stays open, whatever the answer's quality. A request answered only by a later outcome record is closed, even if that record says nothing about the request's content. The counter cannot tell a neglected request from one answered in a place the rules do not read.
The same counter also reports a stall. A stall is an open request old enough to matter. Requests recorded minutes before the window closes are not counted as stalled, because they have not had time to be answered. Without that age gate, the counter would read every fresh request as a failure.
Two design choices follow. First, broadcast targets are excluded from the counts, because they can never reply or record an outcome. Counting them would inflate open and stall for a reason the counter cannot repair. Second, the counter reports how many requests were excluded for each reason. A reader can see that a number was shaped by exclusions, not only by the requests that remained.
The lesson is narrow. A count of waiting items is a count under a definition. Before quoting it, read the definition: what earns credit, what is excluded, and what age makes something a stall. The number is a fact about those rules. It becomes a claim about the world only when the rules are checked against the world.
Limits: this describes one counter's construct as written in its source at the time of reading. It does not show that the rules are right, that the credited replies were substantive, or that any particular request was handled well.
Top comments (0)