Skip to content

Clamp packet ranges and make handoff metadata optional - #27

Merged
TheCryptoDonkey merged 1 commit into
mainfrom
fix/packet-friction
Sep 23, 2026
Merged

TheCryptoDonkey merged 1 commit into
mainfrom
fix/packet-friction

Conversation

@TheCryptoDonkey

Copy link
Copy Markdown
Member

Model-free retrieval analysis of the 98 recorded experiment cells (docs/experiments/retrieval-recall-20260923/):

  • Every arm usually retrieves all required evidence (mean recall 0.90 plain, 0.96 Graphify, 0.92 Context); three quarters of full-recall answers are still rejected, so acceptance on these tasks measures synthesis, not retrieval.
  • The Context arm's measurable friction is tool errors: 71 across its cells, one call in five on DeepSeek.

repository_packet through MCP (CLI spec unchanged):

  • an endLine past the end of a file reads to its last line; originalSpec records the range read so verifyPacket rebuilds it exactly; a range starting past the end is still rejected;
  • only sources is required; blank or omitted handoff metadata gets a neutral task and acceptance check;
  • a plan anchor outside every block points at mode: "build".

Checks: npm run check (182 context-tools tests), navigation smoke, both benchmark gates with recall 1.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KzXbuBSq88i3Re82qt6c4L

…tory_packet

Retrieval measured from 98 recorded cells without model calls shows every arm
usually retrieves the required evidence; the Context arm's friction is tool
errors (one call in five on DeepSeek). Most were whole-file reads with a guessed
endLine and placeholder packet metadata.
@TheCryptoDonkey
TheCryptoDonkey merged commit 3e1cd6a into main Sep 23, 2026
1 check 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