school-research-ops

name	school-research-ops
description	Use when operating the school research queue for kaoyan program_variant work: submit school batches, inspect queue status, interpret gap reasons, and decide whether to resume, lower rpm, or close out.

School Research Ops

Use this skill for repo-local operations around school_research_api.

What This Skill Covers

• Submit CSV/JSONL batches into a school research queue
• Read queue status and explain gap_reason distributions
• Decide whether to resume, lower rpm, lower max_workers, or stop and close out
• Read school-level JSON/Markdown reports after a queue run

Commands

Submit a batch:

python3 scripts/school_research_api.py submit \
  --queue-id <queue_id> \
  --input-file <schools.csv|schools.jsonl>

Run a queue:

python3 scripts/school_research_api.py run \
  --queue-id <queue_id> \
  --rpm <rpm> \
  --max-workers <workers> \
  --llm-transport gate|direct|off \
  --resume

Inspect status:

python3 scripts/school_research_api.py status --queue-id <queue_id>

Write batch report:

python3 scripts/school_research_api.py report --queue-id <queue_id>

Cancel a queue:

python3 scripts/school_research_api.py cancel --queue-id <queue_id>

Interpreting Gap Reasons

• rate_limited: Prefer lower rpm or stop and resume later.
• login_gated_pagination: Rerun only when candidate growth still matters.
• query_split_exhausted: Prefer report-first closeout or explicit review.
• resolver_fetch_failure: Mark for review before more reruns.

Operating Rules

• Default to --resume once a queue has started.
• Do not promote or export from this workflow.
• Treat rpm as the hard upstream request budget; raising workers does not bypass it.
• If rate_limited keeps growing while login_gated stops falling, stop and report instead of brute-force rerunning.