[AdaWorldAPI]

This adds W8 and the focus-churn item from .claude/plans/d-dr64-focus-cache-falsifiers-v1.md to drf-focus-probe.

Change

• FocusManager::evict_to_budget(budget_bytes) -> BudgetReport removes unpinned built foci, least recently loaded first, until the resident bytes fit the budget. Every load, hit or miss, refreshes a focus's recency tick.
• A focus that a reader still holds is never evicted, so a pass can end over_budget. A slot that is still being built counts as pinned.
• Also added: resident_bytes() and FOCUS_BYTES. One focus counts as its gathered i32 column, 16 KiB. The 8 KiB sealed map is not counted, because the caller owns it and shares it through an Arc.

Tests (tests/integration/budget.rs; 27 pass with --features handoff on native and on x86-64-v3; clippy -D warnings is clean)

• W8: a held focus survives evict_to_budget(0) while the idle ones go, and the pass reports over_budget. The held focus still folds to the direct-read result.
• Rebuild: an evicted focus re-gathers on its next load with an identical column and fold.
• Order: the least recently loaded focus is evicted first, and an exact fit evicts nothing.
• Churn, measured: cyclic switching over 3 foci, 30 loads, one eviction pass after each load. A budget of 3 foci gives 3 builds; a budget of 2 or 1 gives 30.

Disable runs (each turns at least one test red)

• Evict pinned foci.
• Skip the recency update on a hit.
• Evict newest first.
• Treat an exact fit as over budget.

Open: under a cyclic working set one focus larger than the budget, LRU is the worst policy; every load becomes a full gather. Whether real thought workloads switch cyclically, and whether to fall back to direct reads instead of rebuilding, is not decided here.

Merge note: this PR and #1769 both edit FocusManager::load near the key construction, and both add a row to the entries index. Whichever merges second may need a small rebase.

Board: .claude/board/entries/2026-10-25-drf-w8-budget.md.

---

Generated by Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 05c923ea-be91-481f-bd6f-019613a81644

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[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	⚠️ Failed 2026-10-25T07:33:29.456446Z We couldn’t complete this review. Comment @codex review to try again.	984c2db	Draft marked ready

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.