Call for independent reviewers: try the agent on YOUR data (~30 min, no experience needed)

[Jackxiaozhiren]

We are looking for a few independent reviewers to try Data Science Agent on their own CSV and tell us what broke. No project knowledge needed — everything is in one kit.

What is this? An evidence-grounded data science agent (natural language in, analysis + evidence + reproducible report out). We just finished honest external benchmarking (DataSciBench GT lane: 0/44 passed, mean CR 0.026 — published as-is in research/external/DATASCIBENCH_REPORT.md §9), and now we need the human half: does it actually help on real questions?

What we ask (≈30 min, blind — please don't read our internal reports first):

1. Install: pip install jack-data-science-agent, run dsa doctor + dsa demo
2. Ask it one question you already know the answer to, on your own CSV
3. Score it 1–5 on 7 dimensions (correctness, statistical validity, evidence, clarity, uncertainty, usefulness, trust)
4. Optional: try to break it (empty file, weird question — failures are welcome)

Kit (self-contained): external-validation/ in the repo — README, INSTALL, TASKS, RUBRIC, FEEDBACK_TEMPLATE, RESULTS_SCHEMA.json.

Ground rules: raw feedback archived verbatim, never edited or cherry-picked; anonymous/pseudonymous OK; partial feedback welcome; environment failures are results, not noise. Current status: no independent review conducted yet — you would be the first, and that will be recorded honestly either way.

Questions welcome below. Thank you!