GEMINI 4 ARGON — CSV CODING EXERCISE
Prepared for SimplifyAITools / Harpal Singh, 19 October 2026.

WHAT WAS ACTUALLY RUN
Only the locally authored, deliberately incomplete baseline was run. With Node v24.19.0, the supplied suite had 18 tests: 9 passed and 9 failed. baseline-tests.tap contains the actual output. No Argon request, generated diff, final model test, model timing or API charge is claimed. The starter is intentionally broken and is not production code.

QUICK START
1. Use Node.js 20 or newer; a currently supported LTS release is preferable.
2. Extract this folder and open a terminal where package.json is located.
3. Run npm test. Failures are expected in the untouched baseline. There are no package dependencies to install.
4. Keep an untouched copy or a clean version-control checkpoint.
5. Run npm run prompt. Copy the resulting prompt-with-files.txt into your selected Argon interface.
6. Confirm the actual model is Gemini 4 Argon, record the route and plan, and use a recorded thinking setting. Do not silently use another model or an automatic selection.
7. Review the proposed diff, apply it to your working copy and run npm test again locally. A plain chat interface may not have run Node.js at all.
8. Keep every prompt and code revision. Record all manual edits. Preserve the original tests and fixtures for the comparison; do not weaken them to get a passing score.
9. Apply only the final proposed change to a clean baseline copy and run the complete suite again. Save the output, diff and original file checksums.
10. Use result-log.csv to record observed model information, time and usage. Mark unavailable information as not available; do not estimate a version or claim zero API cost when no cost information is exposed.

TASK SCOPE
See task-prompt.txt for the full contract. The task covers these three columns in a fixed order: name,url,category; quotes and escaped quotes; a newline in a quoted field; LF/CRLF record endings; empty rows/fields; invalid field counts and unclosed quotes; and a UTF-8 BOM before the header.
It does not validate URLs, write to a database, cover every CSV dialect, support arbitrary delimiters, handle huge streaming uploads or certify production safety. A maintained CSV library may be preferable in a real application. No new runtime dependency is allowed in this exercise so the comparison stays small and reproducible.

RESULT STATUS
- Requested model: no call made.
- Verified returned model/version: not available; no call made.
- Model-generated code or failures: none observed.
- Final model tests: not run.
- Model request timing, tokens and charge: not observed.
- Local baseline fixture: checked; 9 pass / 9 fail.

baseline-sha256.json records the original source, fixtures, tests and prompt. baseline-status.json records the actual local runtime and distinguishes unobserved model results using null fields.

SHARING
All example domains use .test. Do not insert credentials, private customer information or confidential source into a model prompt.
