Run benchmark review for post-fit and downstream-task behavior

Run a second focused manual review against the benchmark suite to validate
post-fit usage, posterior reweighting, and downstream-task execution behavior.

This issue is also review-oriented and should stay narrow.

Scope:

- run benchmark prompts through the current skill set
- focus only on post-fit usage and downstream-task behavior
- collect and categorize failures

Acceptance criteria:

- a review artifact exists
- post-fit and downstream-task failures are categorized clearly
- follow-up fixes can be opened as separate small issues if needed

Start by running the benchmark prompts through the current skill set, focusing only on post-fit usage and downstream-task execution. Record and categorize failures, then produce a review artifact showing the categorized results and any follow-up fixes that should become separate issues.

Written by the indexing model from the issue text.