docs(benchmarks): note ponytail's cost is task-shaped in agentic runs (#121)#122
Open
Lakshya77089 wants to merge 1 commit into
Open
docs(benchmarks): note ponytail's cost is task-shaped in agentic runs (#121)#122Lakshya77089 wants to merge 1 commit into
Lakshya77089 wants to merge 1 commit into
Conversation
The headline benchmark is single-shot completions; a Notes bullet generalized 'the cost gap widens further in ponytail's favor' to real sessions. Independent multi-turn Cursor-SDK A/B (DietrichGebert#121) shows that on large completion-forced tasks ponytail ON can raise tool calls and tokens (more reading before writing) while shrinking output — net cost either way; the win is clearest on snowball-prone/blocked tasks. Scope the single-shot claim and add a 'saves vs. costs' section, plus the skillCount-is-availability-not-usage clarification.
Owner
|
This is great and I want it merged with your name on it, the credit to Ricardo's benchmark repo is exactly right. One snag: it conflicts now. Main already rewrote that "Cost reflects single-shot calls..." bullet under Easy fix on your end:
Ping me when it's pushed and I'll merge it. Thanks for the rigor on this one. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Addresses the documentation gap raised in #121.
The headline benchmark is single-shot completions, but a Notes bullet generalized "the cost gap widens further in ponytail's favor" to real sessions. The independent Cursor-SDK A/B in #121 shows that's not universal: on large completion-forced tasks, ponytail ON can raise tool calls and tokens (more reading before writing) while shrinking the written output — net cost can go either way. The cost win is clearest on snowball-prone / blocked tasks.
Changes (docs only)
skillCountis skills available in the workspace, not skills the model read — only areadof aSKILL.mdis usage.No code change, so
check-rule-copies.jsandnpm testare unaffected.