Missing Segment

The count belongs to the client. A server never reports a missing segment; it answers one specific request with a 430 and says nothing about the rest. The client is the only component that knows a file was supposed to have 812 parts, that six of them came back not-found, and that the shortfall is therefore under one percent. That accounting is what surfaces in a queue as a missing-articles figure.

The shape of the loss carries more than the count

A few segments scattered thinly and evenly across every file in a release is ordinary attrition, and it is what a spool holding articles for several years looks like. Loss concentrated in one file while its neighbours arrive complete usually means that file's articles propagated badly. Every file in the set missing at once, or missing an identical proportion, is a different animal: the release as a whole is not on that server, either because it never arrived or because it was removed.

Three origins that look identical from the client

• Expiry. The articles aged past the retention window on that spool and were rolled out of storage.
• Removal. The articles were taken down following a notice, which happens per article and can leave the rest of a release intact.
• Propagation gap. The post reached some backbones and not others, so the articles exist on Usenet but not on the one queried.

Nothing in the response distinguishes these. The code is identical in all three cases and is deliberately uninformative.

The test that separates them

Age separates the first from the other two with no effort: a release older than the provider's stated binary retention is an expiry case and nothing further needs checking. For anything inside retention, the practical test is a request to a source on a different backbone. If the segments arrive there, the first server either never received them or no longer holds them, and the multi-server configuration most clients support fills the gap automatically on the next attempt. If several independent backbones all return not-found for the same articles, the articles are gone from the reading side and no configuration change recovers them.

How many is too many

The threshold is not a percentage the reader picks; it is the parity the poster included. Missing blocks are compared against recovery blocks, and the set rebuilds when the first number is at or below the second. Two percent missing on a set posted with ten percent parity repairs without comment. Two percent missing on a set posted with one percent does not. This is why identical-looking loss figures end differently on different releases, and why a completion percentage in isolation predicts very little. Fault lies with the spool queried or with propagation between backbones; a local machine cannot manufacture a missing segment, since local disk and network faults produce timeouts, resets and stalls rather than clean not-found answers.