[JIN-strong]

What

The PAI-Bench-C runner (run_paibench_c.sh) wrote its shard JSONLs without a trailing newline and counted pending tasks with wc -l, which counts newline characters — a shard holding exactly one pending row counted as zero. The documented 4-task quick start therefore launched no workers and produced no videos, and a resumed 600-task run silently dropped the final task on every GPU left with a single-row shard.

This PR fixes the counting and closes the adjacent coverage gap:

• input.jsonl and the per-GPU shard JSONLs are now written with a trailing newline, and pending rows are counted with awk 'END{print NR}', which is correct for unterminated and empty files alike (and, unlike wc -l, exits 0 under set -o pipefail).
• The worker-pid wait loop uses the ${var+...} expansion so a fully-complete resume cannot abort under set -u on bash < 4.4.
• The per-modality evaluation step is wrapped in evaluate_modality() and gated on the requested task set: the collected video count must match the row count of that modality's input.jsonl (falling back to PAIBENCH_C_NUM_SAMPLES for pre-existing trees), and any mismatch aborts with both numbers and the recovery paths instead of scoring a subset as the benchmark result.

Why

A run interrupted near the end could report metrics computed over fewer tasks than requested with no signal, and the documented default invocation could not produce any videos at all.

Fixes #376.

Validation

• bash test_run_paibench_c.sh → 16 assertions pass. The new test extracts the real build_jsonl_and_run and evaluate_modality functions from the script and drives them with a stub inference entrypoint (no GPU, network, or venv needed): before the fix it reproduces the zero-video quick start, and it now covers single-row shard counting, fully-complete resume under set -euo pipefail, full-coverage evaluation, and rejection of a tree with one video missing (asserted to fail before any evaluation work starts).
• bash -n on both scripts → clean.
• shellcheck reports only warnings on pre-existing lines unrelated to this change.

[yaoxu-crypto]

Thanks @JIN-strong for the careful fix and tests. Closing here because the evaluation suites moved to Cosmos Framework (.agents/skills/evaluation/cosmos3/generator/paibench_c/ in NVIDIA/cosmos-framework) and were removed from this repo in #384, so this PR can no longer merge.

The single-row shard counting bug is being fixed there in NVIDIA/cosmos-framework#317, which writes a trailing newline per row. That PR does not include the rest of your change: the per-modality coverage check before evaluation, the set -u fix in the worker-pid wait loop, and test_run_paibench_c.sh. Those are still worth having. Please open a PR against NVIDIA/cosmos-framework with them (the file paths are the same under .agents/skills/), and reference #376 and cosmos-framework#317.