Skip to content

Offer the queries the answer ran when suppressing them leaves nothing to click - #72

Merged
Robbie1977 merged 1 commit into
mainfrom
fix/followon-chips-when-every-query-ran
Sep 11, 2026
Merged

Offer the queries the answer ran when suppressing them leaves nothing to click#72
Robbie1977 merged 1 commit into
mainfrom
fix/followon-chips-when-every-query-ran

Conversation

@Robbie1977

Copy link
Copy Markdown
Contributor

A follow-on chip is an offer of something new, so a query this turn already ran is suppressed. That is right until it is all of them. The olfactory system offers exactly two — PartsOf and NeuronsPartHere — and when the absence check runs both, which it does to about one welcome-screen answer in three, every chip is suppressed and the reader is left an answer with nothing at all to click. This is battery WS3, the single failure on main's first run after #71; a WS3-only re-run reproduced it 1 in 3, and it is unrelated to that PR — it has been latent since the absence escalation and the chip suppression have coexisted.

The chip is still worth offering in that case: it opens the rows themselves, where the prose only summarised them. Only offering it alongside an unasked query is worse, so the unrun queries still win wherever there are any; the run ones are the fallback, not the default.

Suite 1,391/1,393 (two skips), lint clean. The new test fails without the change.

… to click

A follow-on chip is an offer of something new, so a query this turn already
ran is suppressed. That is right until it is all of them. The olfactory
system offers exactly two — PartsOf and NeuronsPartHere — and when the
absence check runs both, which it does to about one welcome-screen answer
in three, every chip was suppressed and the reader was left an answer with
nothing at all to click (battery WS3, reproduced 1 in 3 in CI on main and
in all three prior Qwen runs' absence-escalation path).

The chip is still worth offering in that case: it opens the rows
themselves, where the prose only summarised them. Only offering it
alongside an unasked query is worse, so the unrun queries still win
wherever there are any; the run ones are the fallback, not the default.

Suite 1,391/1,393. The new test fails without the change.
@Robbie1977
Robbie1977 merged commit 14e898b into main Sep 11, 2026
10 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant