You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit bd04924
Browse filesBrowse the repository at this point in the historyBrowse files
test(query): pin the text page budget and cursor contracts
The token-efficiency pass changed behaviour that tests did not pin, so this
adds four regression tests beside the existing agent-query tests:
* a page renders at most the Agent View profile (12 primary results) even when
the token budget would allow far more, reports the ledger's true total and
continues with `next=`;
* a resolved answer prints no entity identifiers while a `search` pick list
prints them, and the JSON projection carries identifiers for both;
* the page opens with one `RESULT` line, its pagination line has no version or
budget echo, and a continuation cursor stays within its compact size bound;
* the previous release's long-form cursor is rejected instead of being
reinterpreted as the compact wire form, whose decode failure now names the
version explicitly.
The shared paged-text footer (`path`, `node`, `explore`) also stops repeating
the caller's own budget and the previous page number, which no caller needs:
`Pagination: page=A/B <item>=X-Y/T next=...`.
Final measured state on the 50-question suite: Compass answers 50/50 with a
paired median of 362 tokens against the pre-pass 560 (total output 40,082 ->
21,280), and the first suite stays at 47/47.
0 commit comments