Fix vf-eval concurrent rollouts label to show effective capped value

[willccbb]

Motivation

• The eval UI could display a concurrent rollouts value larger than the actual work or runtime concurrency because it showed config.max_concurrent directly instead of the effective concurrency used at runtime and did not cap by total rollouts.
• The intent is to make the display reflect the same effective concurrency calculation used by evaluate(...) (respecting rollouts_per_example and independent_scoring) and avoid inflated labels.

Description

• Add a static helper EvalDisplay._display_max_concurrent(config, total_rollouts) that computes the effective concurrency by applying rollout-per-example scaling (when independent_scoring is False), using math.ceil for division, and capping the result to total_rollouts when appropriate.
• Use the new helper in _make_env_panel to render the concurrent rollouts label instead of showing config.max_concurrent directly, and import math for the calculation.
• Add focused unit tests in tests/test_eval_display.py that assert capping to total_rollouts, correct scaling by rollouts_per_example, and that independent_scoring=True preserves the unscaled value.

Testing

• Ran uv run pytest tests/test_eval_display.py and the tests passed (3 passed).
• Ran uv run ruff check verifiers/utils/eval_display.py tests/test_eval_display.py and style checks passed.

---

Codex Task

---

Note

Low Risk
UI-only display logic change with small, well-covered arithmetic adjustments; no runtime evaluation behavior is modified.

Overview
Fixes the eval UI’s concurrent rollouts label to reflect effective runtime concurrency rather than raw config.max_concurrent.

Adds EvalDisplay._display_max_concurrent() to scale concurrency by rollouts_per_example when independent_scoring is disabled (using ceil) and to cap the displayed value to total_rollouts, and wires this value into the panel rendering. Includes unit tests covering capping, scaling, and the independent_scoring=True exception.

^{Written by Cursor Bugbot for commit c31ec0d. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Capping uses wrong unit when independent_scoring is false

Medium Severity

When independent_scoring=False and rollouts_per_example > 1, the function divides max_concurrent by rollouts_per_example to get concurrent example groups, but then caps against total_rollouts (measured in rollouts). This unit mismatch causes incorrect display values. For example, with max_concurrent=100, rollouts_per_example=4, and num_examples=5, the display shows 20 instead of 5. The cap should use total_rollouts // config.rollouts_per_example when scaling has been applied.