Pin the last research subject onto follow-ups - #53
Merged
Conversation
Research already extracts a name or figure onto the first line of the digest. The 0.8B model still invents a different one, or quotes the office instead of the incumbent. reviewAnswer now treats that line the way it already treats the calculator: if the draft drops it, one correction asks for it back. The check stays shy. A surname standing in for the full name, a German decimal, and a figure rounded the way extractFigure already treats as one reading all pass. A biography digest with no Answer line is left alone. When the wind-down round comes back empty, budgetFallback now hands over that same one-liner instead of an apology. Co-authored-by: Sebastian <devbadya@users.noreply.github.com>
After *Wer ist der Bundeskanzler?*, a follow-up like *und der von Frankreich?* was searched as a fragment. topic.ts already pinned the last weather city; it now pins the last office or question the same way, and recency decides which one, so an older Frankfurt does not leak onto a chancellor follow-up. A complete new question, or a follow-up that names a different person (*und Elon Musk?*), stays unpinned. The research skill has a follow-up exemplar so the model is shown to keep the office and add the place. Co-authored-by: Sebastian <devbadya@users.noreply.github.com>
devbadya
marked this pull request as ready for review
September 10, 2026 10:37
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
After Wer ist der Bundeskanzler?, a follow-up like und der von Frankreich? was searched as a fragment.
topic.tsalready pinned the last weather city into the prompt; it now pins the last research office or question the same way.When the line is injected
researchcall, or the last researchable user question if the eval history has no tool cardWhen it stays off
The
research-questionskill has a follow-up exemplar so the model is shown to keep the office and add the place.This is working memory in the current chat, not the model training itself. Auto-saving every answer into IndexedDB would double the GPU cost of a turn and fill recall with stale world facts. Durable memory stays explicit: the Memory panel, or remember that….
pnpm checkis green (1,335 tests). Generation was not run: this VM has no GPU adapter.