When requests are sent

There is no submit action, so OpenCues decides when to analyse your text and which words to include. Most of the rules exist to send less.

When requests are sent

Cues send automatically; blanks send when you invoke them

Two mechanisms, two triggers.

Cue analysis is automatic: it runs after a short pause in typing, not on every keystroke, so a half-typed word is never analysed. There is no submit action.

A blank is user-invoked: typing _ is itself the request, and it dispatches immediately. There is no pause to wait through, because the underscore already states what you want.

Words with alternatives are not re-sent

Each word is analysed once.

A word that already has alternatives is excluded from later requests. This keeps request sizes small, and it prevents a loop: without it, a word you just cycled to would be analysed as fresh input and offered alternatives to the alternative.

Local answers skip the model

Only the remainder is sent.

Alternatives that can be computed on your machine are computed there and appear without network latency. The request that goes out contains the words no local source could answer, not your whole text.

When two files match the same word

Higher priority: wins.

Each cue file carries a priority: number. When two files match the same word, the higher value claims it; load order plays no part.

Shipped filepriority:Effect
calendar90claims time-mentioning sentences first
more-formal85claims casual sentences next
example (template)65the default value to copy for a new file
spelling10claims whatever no other file wanted

Spelling is lowest so that it acts as a catch-all: a term your own subject file claims is handled by that file, not marked as a typo.