SimpleQA Verified — 15-item selection rule, frozen 12 Nov 2026, 1:11:38 AM IST (Asia/Calcutta) This file records the rule and the IDs it produced. It was written before any model was prompted. No replies have been collected. Source file - Hugging Face dataset: google/simpleqa-verified - File: https://huggingface.co/datasets/google/simpleqa-verified/resolve/main/simpleqa_verified.csv - Downloaded 12 Nov 2026, 1:11:38 AM IST (Asia/Calcutta), HTTP 200, 349184 bytes - SHA-256: b5db21155444763543fe31b67e7cf28ce2bb225742a5b889421c5f182e2f92f5 - Rows: 1000. Fields used for selection: topic and original_index only. The CSV was loaded as a whole (it is one file), but no item was chosen by reading the problem or answer text, a difficulty label, or any model output. After the 15 original_index values were fixed by the algorithm below, those rows' problem and answer fields were written to items.csv. Algorithm (deterministic; no random seed) 1. Load the CSV at the SHA-256 above. Keep all 1000 rows. Do not drop any topic or row. 2. Group rows by the topic field as a Unicode string, exactly as written in the file. 3. Within each topic, sort rows by original_index parsed as a base-10 integer, ascending. 4. Let N = 1000 and K = 15. For each topic, compute quota = (row count of that topic) / N * K. 5. Give each topic floor(quota) seats (Hamilton / largest-remainder method). 6. Let L = K minus the sum of those floors. Award L leftover seats, one each, to the topics with the largest leftover (quota - floor(quota)). Ties are broken by topic name as a Unicode string, ascending (A-Z, case-sensitive as in the file). 7. For each topic, take the first (seats) rows in the sorted order from step 3. Topics with zero seats are unused. 8. Number the 15 items 01 to 15 by topic name A-Z, then original_index ascending. Freeze those 15 original_index values. This is the pre-registered set. Application of the rule to the pinned file (computed 12 Nov 2026, 1:11:38 AM IST (Asia/Calcutta)) Topic quotas: - Art: count=145, quota=145/1000*15=2.175000, floor=2, remainder=0.175000, leftover_seat=no, seats=2 - Geography: count=111, quota=111/1000*15=1.665000, floor=1, remainder=0.665000, leftover_seat=yes, seats=2 - History: count=52, quota=52/1000*15=0.780000, floor=0, remainder=0.780000, leftover_seat=yes, seats=1 - Music: count=102, quota=102/1000*15=1.530000, floor=1, remainder=0.530000, leftover_seat=yes, seats=2 - Other: count=102, quota=102/1000*15=1.530000, floor=1, remainder=0.530000, leftover_seat=no, seats=1 - Politics: count=176, quota=176/1000*15=2.640000, floor=2, remainder=0.640000, leftover_seat=yes, seats=3 - Science and technology: count=160, quota=160/1000*15=2.400000, floor=2, remainder=0.400000, leftover_seat=no, seats=2 - Sports: count=117, quota=117/1000*15=1.755000, floor=1, remainder=0.755000, leftover_seat=yes, seats=2 - TV shows: count=20, quota=20/1000*15=0.300000, floor=0, remainder=0.300000, leftover_seat=no, seats=0 - Video games: count=15, quota=15/1000*15=0.225000, floor=0, remainder=0.225000, leftover_seat=no, seats=0 Leftover seats (L=5) awarded in this order: History, Sports, Geography, Politics, Music. Frozen 15 items: - item 01: original_index=13, topic=Art, answer_type=Date - item 02: original_index=15, topic=Art, answer_type=Other - item 03: original_index=55, topic=Geography, answer_type=Number - item 04: original_index=73, topic=Geography, answer_type=Number - item 05: original_index=29, topic=History, answer_type=Other - item 06: original_index=32, topic=Music, answer_type=Date - item 07: original_index=39, topic=Music, answer_type=Place - item 08: original_index=37, topic=Other, answer_type=Person - item 09: original_index=5, topic=Politics, answer_type=Number - item 10: original_index=8, topic=Politics, answer_type=Person - item 11: original_index=9, topic=Politics, answer_type=Person - item 12: original_index=97, topic=Science and technology, answer_type=Date - item 13: original_index=179, topic=Science and technology, answer_type=Place - item 14: original_index=59, topic=Sports, answer_type=Person - item 15: original_index=66, topic=Sports, answer_type=Person Do not change this set after the first prompt. If a later download of the CSV has a different SHA-256, stop and report; do not silently swap items.