Draft known-answer checks for data transformations

Draft checks for grouping, joins, missing values and duplicates using small fictional inputs and human-known answers.

適用情境 / When to use

Mike Woodward’s personal 2025 Cursor assessment describes plausible-looking dataframe code producing incorrect results. This exercise drafts a test checklist only.

步驟 / Steps

1. Open an ordinary text chat at gemini.google.com with an eligible account. Prepare shareable or fictional plain text using the prompt parameters; transcribe it yourself and mark ambiguities. Do not upload source files or connect accounts.
2. Paste the text and prompt; request only a draft, comparison or plan, not tool execution.
3. Check every output against supplied sources, references and unknowns. A human decides next steps; no automatic publishing, payment or deployment.

提示詞 / Prompt

From [small fictional tables with column definitions and row IDs], [intended transformation rules] and [my hand-calculated answers and workings], draft a test checklist without executing it. Cover empty inputs, missing values, duplicate keys, ordering and grouping boundaries. For joins, describe key uniqueness, possible row-count changes and unmatched rows. List inputs, expected results, evidence, unknowns and manual verification steps. My answers may be wrong too: flag conflicts. Do not use output from the same generated implementation as an independent answer. With no execution record, label every test unrun. Do not install packages, read/write files, delete data or connect databases.

留意事項 / Cautions

• Historical subjective scores are not a current-version benchmark. No code or tests were executed.
• Checked 2026-10-30: official Gemini web lists Hong Kong, Macau, Taiwan, Traditional/Simplified Chinese and English; account and age eligibility apply. Limited use is available without a paid AI plan and varies by account and workload. Taiwan’s public subscription page lists Free at NT$0/month; no paid plan is required for this exercise. HK/Macau paid prices and local taxes, offers and final quotes are unverified. Output quality in all three languages is untested. Web eligibility does not establish App/API eligibility; neither is needed here. Original-tool regional features, devices and paid eligibility are unconfirmed; do not bypass restrictions. Android’s official region list also includes HK/Macau/Taiwan. The iOS and AI Studio/Gemini API lists include Taiwan, not Hong Kong or Macau; feature and device restrictions still apply.

來源限制 / Evidence limits

Historical subjective scores are not a current-version benchmark. No code or tests were executed.

流程說明 / Workflow note

Original, untested text-only exercise by this site; it does not reproduce source devices, agents or integrations. Checked 2026-10-30: official Gemini web lists Hong Kong, Macau, Taiwan, Traditional/Simplified Chinese and English; account and age eligibility apply. Limited use is available without a paid AI plan and varies by account and workload. Taiwan’s public subscription page lists Free at NT$0/month; no paid plan is required for this exercise. HK/Macau paid prices and local taxes, offers and final quotes are unverified. Output quality in all three languages is untested. Web eligibility does not establish App/API eligibility; neither is needed here. Original-tool regional features, devices and paid eligibility are unconfirmed; do not bypass restrictions. Android’s official region list also includes HK/Macau/Taiwan. The iOS and AI Studio/Gemini API lists include Taiwan, not Hong Kong or Macau; feature and device restrictions still apply.

原始工具組合 / Original tool stack

Cursor; pandas; project data

Source publication date: 2025-06-05

Source checked: 2026-10-30

Cursor for data science - a scorecard