Trim tool copy; land gating-column drops; rebuild recipe_save spec - #546
Merged
Merged
Conversation
…tokens conversation_search: 2460 -> ~1990 chars. Drops the 'summary is auto-generated' restatement (implied by the return shape), the passage-absence clause (a negation the model can infer), and tightens the within_days/prefer_recent param prose. The RANKING-IS-BY-TOPIC guard - the observed-failure guard that made this tool big - stays verbatim. recipe_save rebuilt: 2181 -> 1718 chars. The accreted prose rules collapse into a one-line grammar + inline-reference example + terse wrong-vs-right guards. Every observed-failure guard intact; the full cooklang spec stays in docs/user/cookbook.md for humans. toolboxes_enabled / tools_enabled drops go live: two releases have deployed since the deferral, so no cached pre-gating-retirement bundle can still be in service. Verified dropped against prod. Token recheck (30h post-merge window, small n): declare-everything did NOT raise cold prompts by the predicted +10k (24,865 vs 31,393 gated era); continuation marginal growth is 17.3% of full prompt - the cache rides as designed; zero empty completions since #543 (both 900-1000 ct turns delivered full content). Full writeup in scratch/.
QA surfaced the miss on first attempt: the model wrote prep hints inside the ingredient name (@finely chopped basil) - syntactically legal cooklang, so the validator accepts it and the hint renders as part of the name instead of a note. The rebuilt spec's guards now name it explicitly with the wrong-vs-right pair, matching the other observed-failure guards.
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
1 participant
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.
SYNOPSIS
Rebuilds the recipe_save description examples-first, trims conversation_search, lands the deferred toolboxes_enabled/tools_enabled drops, and rechecks token costs against prod.
PURPOSE
Continues the tool-consolidation session (#543/#544/#545). The photo quartet and record pairs stay separate by design (disjoint param sets), so their token cost comes from copy trim; the recipe spec had accreted prose rules that needed collapsing into intention; the gating columns were waiting their one-release deferral.
DESCRIPTION
NOTES