docs(measurements): 라운드 결산 -- main 전체 벤치 수치와 남은 지렛대

[choiceoh]

장부만. main 0995825 의 전체 벤치(GEMM/MHC/KDA, VERDICT PASS) 한 번의 수치와, 단독 커널이 하한 근처라는 판단, 다음 지렛대(서빙 무장, 체인 융합).

🤖 Generated with Claude Code

---

Note

Low Risk
Documentation-only update to MEASUREMENTS.md; no runtime, serving, or kernel code changes.

Overview
Adds a ★★★메가커널 결산 ledger block (2026-09-02 evening, main 0995825, srv2 single full bench) immediately after the megakernel 2차 verdict, before the existing MHC 9차 chapter.

The new section records MK vs stock µs for GEMM W4 (four shapes, x2), MHC T=8/T=32, and KDA acc=1/3/8 (including replication), with accuracy notes and VERDICT PASS. It contrasts the round baseline after #219 (GEMM flat at ceiling; MHC/KDA improved from 30.7/59.4 and 478) with merges #222 (KDA phase attribution/rewrite, kernel span 199 µs) and #224 (MHC restructure).

It also updates the forward plan: standalone kernels are treated as near their floors, so the documented next levers are serving enablement (MK-GEMM W4 bracket, MK-KDA shadow→bracket) and chain fusion via a_ready from KDA p4→p5 into MHC tail → in_proj/qkv, with serving glue tensor identity called out.

^{Reviewed by Cursor Bugbot for commit c3416d4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-02T10:14:38.533795Z	c3416d4	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[cursor]

Approved. Cursor Bugbot completed successfully with no findings that need human review, and no applicable approval policy requires extra review. No reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[Copilot]

🟡 Changes recommended

The newly added summary table has an internal inconsistency with adjacent documented MHC numbers, which should be clarified or aligned to avoid confusing the recorded benchmark story.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds a new “★★★메가커널 결산” ledger entry to document a single full-benchmark snapshot on main (0995825), summarizing MK vs stock performance and outlining the next optimization levers.

Changes:

• Add a new round-summary section with MK vs stock µs for GEMM/MHC/KDA and a VERDICT PASS note
• Record round baseline/merged PRs and update the forward plan (serving enablement + chain fusion via a_ready)

File summaries

File	Description
MEASUREMENTS.md	Adds a new megakernel round “settlement” block summarizing benchmark results and next steps

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.