| """Short-prompt (q3s) copies of q3 Clef-labelled JSONLs for the Glint student. | |
| Keeps id, images, state, gold, teacher, teacher_logits and meta (incl. box_pairs, so box_primary_button stays | |
| paired with primary_button) exactly as they are -- no re-labelling. Only the question texts change: | |
| * choice / noul: one-line question (per record language), choice criteria reduced to the option ids | |
| (descriptions null, the compact encoder renders ids only); noul criteria dropped. | |
| * box: a short element description (box_primary_button = "the primary action button", same pairing). | |
| Sets meta.question_version = "q3s" and record["prompt_style"] = "compact" (glint/schema_compact.py). | |
| usage: python scripts/make_short.py in.q3.clef.jsonl [more.jsonl ...] -> writes <name>.short.jsonl next to each | |
| (refuses to overwrite an existing output) | |
| """ | |
| import json, sys | |
| from pathlib import Path | |
| SHORT = { | |
| "screen": {"en": "What is the current screen state?", "ja": "\u73fe\u5728\u306e\u753b\u9762\u306e\u72b6\u614b\u306f\uff1f"}, | |
| "has_error": {"en": "Is an error message visible?", "ja": "\u30a8\u30e9\u30fc\u8868\u793a\u304c\u3042\u308b\u304b\uff1f"}, | |
| "primary_button": {"en": "Which is the primary action button?", "ja": "\u4e3b\u8981\u306a\u30dc\u30bf\u30f3\u306f\u3069\u308c\u304b\uff1f"}, | |
| "is_loading": {"en": "Is something loading?", "ja": "\u8aad\u307f\u8fbc\u307f\u4e2d\u304b\uff1f"}, | |
| "has_modal": {"en": "Is a modal or sheet presented?", "ja": "\u30e2\u30fc\u30c0\u30eb\u3084\u30b7\u30fc\u30c8\u304c\u8868\u793a\u4e2d\u304b\uff1f"}, | |
| "has_destructive_action": {"en": "Is a destructive action visible?", "ja": "\u7834\u58ca\u7684\u306a\u64cd\u4f5c\u304c\u3042\u308b\u304b\uff1f"}, | |
| "is_keyboard_visible": {"en": "Is the keyboard visible?", "ja": "\u30ad\u30fc\u30dc\u30fc\u30c9\u304c\u8868\u793a\u4e2d\u304b\uff1f"}, | |
| "is_empty": {"en": "Is it an empty state?", "ja": "\u7a7a\u306e\u72b6\u614b\u304b\uff1f"}, | |
| "has_tab_bar": {"en": "Is a tab bar visible?", "ja": "\u30bf\u30d6\u30d0\u30fc\u304c\u3042\u308b\u304b\uff1f"}, | |
| } | |
| BOX_SHORT = { | |
| "box_tab_bar": "the tab bar", | |
| "box_selected_tab": "the selected tab", | |
| "box_primary_button": "the primary action button", | |
| "box_modal": "the modal or sheet", | |
| "box_focused_text_field": "the focused text field", | |
| "box_keyboard": "the keyboard", | |
| "box_loading_indicator": "the loading indicator", | |
| "box_error_message": "the error message", | |
| "box_empty_state": "the empty-state message", | |
| "box_destructive_button": "the destructive button", | |
| } | |
| def shorten(r): | |
| lang = "ja" if (r.get("meta") or {}).get("lang") == "ja" else "en" | |
| qs = {} | |
| for qid, q in r["questions"].items(): | |
| t = q["type"] | |
| if t == "box": | |
| if qid not in BOX_SHORT: | |
| raise SystemExit(f"{r['id']}: no short text for box question {qid}") | |
| qs[qid] = {"type": "box", "instructions": BOX_SHORT[qid]} | |
| elif t == "choice": | |
| qs[qid] = {"type": "choice", "instructions": SHORT[qid][lang], "criteria": {k: None for k in q["criteria"]}} | |
| elif t == "noul": | |
| qs[qid] = {"type": "noul", "instructions": SHORT[qid][lang]} | |
| else: | |
| raise SystemExit(f"{r['id']}: unsupported type {t} for {qid}") | |
| out = dict(r) | |
| out["questions"] = qs | |
| out["meta"] = {**r["meta"], "question_version": "q3s", "question_version_src": r["meta"].get("question_version")} | |
| out["prompt_style"] = "compact" | |
| return out | |
| def main(paths): | |
| for src in map(Path, paths): | |
| dst = src.with_name(src.name[: -len(".jsonl")] + ".short.jsonl") | |
| if dst.exists(): | |
| raise SystemExit(f"refusing to overwrite {dst}") | |
| n = 0 | |
| with open(src, encoding="utf-8") as f, open(dst, "x", encoding="utf-8", newline="\n") as out: | |
| for line in f: | |
| if line.strip(): | |
| out.write(json.dumps(shorten(json.loads(line)), ensure_ascii=False) + "\n"); n += 1 | |
| print(f"{n} records {src} -> {dst}") | |
| if __name__ == "__main__": | |
| main(sys.argv[1:]) | |