Recover live updates and validate integrated performance, visuals and network attribution

[ProtonDev-sys]

Why

A transient sampling exception could permanently stop live updates, and an earlier successful snapshot could still let an unhealthy UI benchmark pass. Failed service refreshes need an accessible stale-data indication, and Performance selection needs consistent behavior when a device disappears.

Changes

• Retry failed sampling with bounded, interruptible backoff while retaining the sampler and last good snapshot; balance COM lifetime and reliably interrupt paused shutdown.
• Show live-update recovery in the title and service inventory errors in the Services footer; retry dropped status notifications.
• Preserve Performance summary behavior and choose CPU consistently after a selected resource disappears.
• Reject UI benchmark results if sampling fails during measurement, including failures followed by recovery.
• Add a paired performance workflow with immutable source and executable identities, strict evidence validation, native light/dark captures, and an owned UDP attribution diagnostic.
• Keep smoke instrumentation out of normal callbacks and all diagnostic tooling out of production.

Verified evidence

Windows verification passed all 12 CTests, 354 named native self-test checks, all 21 suite cases, Python checks on Windows/Linux, production exclusions, and packaging. All 18 retained UI reports meet the existing responsiveness budgets; the worst queue p95 was 32.2983 ms against 100 ms. These reports were checked offline without rerunning benchmarks.

The paired study completed 18 serial AB/BA/AB measurements. An independent audit matched 486 metric values to raw reports and recomputed all 81 summary rows. Live Windows UDP passed with 32 datagrams, 32,768 bytes, and 64 own-process events; no skip was used.

All 38 native light/dark views were inspected. Missing GPU-engine feedback on the VM is being corrected in a focused follow-up. These images do not establish physical GPU or complete Windows 11 coverage.

The main timing ranges overlap. Favorable sampler/idle CPU medians do not establish a general speedup. Production grew by 22,016 bytes; diagnostic private memory and working set were somewhat higher. Full results and limits will be preserved in the release performance report.

Integration

After #37 merged, overlapping ancestry required a merge-resolution commit. Its tree remains exactly 394fdb43ac86e09a2d30eba5eb4e399d4ef3d50e, identical to the fully verified candidate. No source file changed. The one-off performance-study label was removed after reviewing the completed evidence to avoid measuring unchanged contents again. Normal CI still runs for the integration commit.

[coderabbitai]

Important

• 🔍 Trigger review

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

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 25273bf5-aa22-4477-9ed1-a665ae656b89

---

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

---

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.}