Run finalization cleanup targeted profile closeout

[jununfly]

This was generated by AI during triage.

Parent

Plan: docs/plans/2026-06-20-rust-hybrid-finalization-cleanup-diagnostics-and-batching.md

What to build

Run targeted evidence for the finalization cleanup diagnostics and bounded resolved-cleanup optimization, then write a closeout decision. The closeout should decide whether to keep the cleanup batching optimization, mark it no-go, or identify a prerequisite before further work.

This issue is evidence and decision work only. Do not broaden into edge-write ownership, Rust subprocess cleanup, schema changes, README metrics, full scoreboard, or agent A/B.

Acceptance criteria

• Run a current-repo targeted rust-hybrid profile.
• Run a VS Code sparse targeted rust-hybrid profile using /private/tmp/codegraph-corpus/vscode-sparse if it exists, is a Git checkout, and contains src/vs/workbench, src/vs/platform, and src/vs/base.
• If VS Code sparse is missing or incomplete, mark the evidence as needs-human-setup and do not clone a replacement corpus.
• Record RSS or unavailable reason.
• Record graph stats, fallback taxonomy, finalization write/cleanup sub-buckets, row/edge counts, and timing context.
• Write a closeout decision under docs/benchmarks/ with keep, no-go, or prerequisite.
• Update PRD: Rust-hybrid architecture and performance optimization #295 and Post-PRD optimization tracker for Rust indexing performance #165 with the closeout link and conclusion.
• Do not update README metrics.

Blocked by

• Split finalization write and cleanup profile diagnostics #326
• Add deterministic cleanup contract tests #327
• Implement bounded resolved cleanup batching optimization #328

[jununfly]

This was generated by AI during triage.

Completed targeted closeout. Evidence artifacts:

• docs/benchmarks/2026-06-20-finalization-cleanup-current.profile.json
• docs/benchmarks/2026-06-20-finalization-cleanup-vscode-sparse.profile.json
• docs/benchmarks/2026-06-20-rust-hybrid-finalization-cleanup-closeout-decision.md

Conclusion: keep diagnostics; treat resolved cleanup batching as no-go as a standalone performance lever. VS Code sparse was present and hydrated; RSS was recorded from /usr/bin/time -l.

[jununfly]

This was generated by AI during triage.

Closing with targeted profile evidence and closeout decision recorded in docs/benchmarks/2026-06-20-rust-hybrid-finalization-cleanup-closeout-decision.md.