Nativ translation-review exercise, 30 October 2026

Fictional Field Notes examples, manually authored for this guide. They are not Nativ-generated output, a translation-quality benchmark, or reviewed/approved French copy. The first two rows intentionally change a percentage or placeholder. The qualifier row demonstrates why structural checks do not establish linguistic accuracy.

Run locally with Python 3:
  python3 preflight.py exercise.csv

Expected exit code: 1. Two rows are flagged. A nonzero exit is deliberate for this broken-input exercise. No packages or network are used. The checker only compares ASCII integer percentages and double-braced named tokens; it does not parse ICU, HTML, dates, decimal percentages, other numeral systems or general message formats. A pass approves nothing.

For your own handoff, remove exercise_note, keep stable IDs and source/target locale, and add reviewer, corrected_translation, reason, reviewed_version, decision and decision_date. Return Pending rather than treating empty cells as approval. Inspect the final asset again after edits.