/ultrareview returns empty findings array on 10K-line diff (silent failure or false negative)

Summary

/ultrareview completed with status completed but returned an empty findings array ([]) on a substantial diff. The free-quota counter appears to have been consumed despite no review output being produced.

Expected behavior

Ultrareview launched on a non-trivial diff (10K+ insertions across 34 files, including SVG animation engine changes, vendored library adoption, new e2e tests, and host-page rewrites) should return at least some findings — bugs, regressions, code-quality notes, principle violations, or even just style nits. Returning zero findings on a diff of this size and complexity is statistically improbable in real code review, and indistinguishable from a silent pipeline failure from the user's perspective.

Actual behavior

<task-notification>
<task-id>rrxk0pleb</task-id>
<task-type>remote_agent</task-type>
<status>completed</status>
<summary>Remote review completed</summary>
</task-notification>
The remote review produced the following findings:

[]

The launch dialog reported Free ultrareview 2 of 3, suggesting a quota slot was consumed.

Repro context

• Branch: main
• Diff scope: 34 files changed, 10282 insertions, 2644 deletions
• Diff content includes:
• Animation engine changes (sfs/static/js/scrolly-v2.js — legend rendering, mark cascade timing, axis transitions)
• New host page logic (sfs/static/scrolly-v9-clay.html — Motion library integration, async play loop, render-t clamp, deferred setup, legend choreography)
• Vendored library (sfs/static/js/vendor/motion@12.38.0/)
• 9+ new Playwright e2e probes
• New skeleton module directory (sfs/static/js/chart-muse/)
• Memory file regenerations
• Session URL: https://claude.ai/code/session_01N37PoFBCSDEasogRzJz8kt
• Task ID: rrxk0pleb
• OS: macOS (Darwin 25.4.0)
• Triggered via: /ultrareview in Claude Code CLI

Impact

Two concerns:

1. The session URL is the only artifact for diagnosing whether the review pipeline actually ran or errored silently before reporting findings.
2. Free quota is limited (3/user), and an empty-findings result feels indistinguishable from a failed run — should this case decrement the counter, or be treated as a non-result and refunded?

What would help

• A way to differentiate "review ran, found nothing" from "review pipeline silently failed" in the user-visible output (e.g., show how many sub-agents ran, what their reports were, how many files each touched).
• If this turns out to be a pipeline failure, a way to retry without consuming a quota slot.