Repository navigation
Kid Mode: joke asks get a joke, not a story (#191) - #205
Merged
Merged
Conversation
This was referenced Nov 1, 2026
…191) HELD: do not merge or deploy until robot-ops' 18-prompt Kid Mode rerun shows the drift is still present. About one K01/V06 attempt in three became a 50-65 word story; the length rule allows six sentences for open-ended asks and the 4B treats a joke as one. Append a one-line reminder beside the words for joke asks only (the long rules stopped tool calls when appended). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
BrettKinny
force-pushed
the
agent/overnight-1101-kid-prep
branch
from
November 1, 2026 20:18
d0c66a3 to
c4cc333
Compare
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Stacked on the PR for
agent/overnight-1101(base branch is that one); merge after it.Summary
Kid Mode: asked for a short joke, the 4B sometimes tells a 50-125 word story. The length rule allows 6 sentences for open-ended asks and the model reads a joke as one. For joke asks only (not when "story" is also asked), append a one-line reminder beside the user's words: "Reply with just the joke: one or two sentences, not a story." (The long rules are not appended per turn because that stopped tool calls.)
Evidence (overnight A/V baseline, kid mode on, fw 1.3.5-dev.1101.2,
qwen3.5:4b, log.md 03:15 and 03:48)response_too_long(124-word story) of 6, plus one 32-word story-form pass; K01 3/3; K04 5/6 (one 50-word safe refusal too long).brain offline, unrelated: container restart); K04 3/3.What is NOT proven
The sample is small: 1 hard failure in 6 before versus 0 in 3 after. That is consistent with a fix but does not show one, since the drift was about 1 in 3 to 1 in 6 to begin with. K02 ("12 plus 10 room", #178) is unchanged and still answers 22 (accepted as a pass on 2026-10-24).
Refs #191
AI-assisted: drafted and implemented by Claude (Opus 5.5); a human reviews and is accountable for what lands.
🤖 Generated with Claude Code