Name one hundred real women, then satisfy three intersecting quotas without losing count: exactly fifteen must be controversial or widely disliked, exactly twenty must have had a non-English first language, and exactly twenty-five must qualify as historical. The model has to know the people, their biographies, and the bookkeeping at the same time.
This is a knowledge challenge disguised as a counting challenge. The mechanical checker settles the envelope: 100 distinct rows, numbered in order, and exactly 15 / 20 / 25 true flags. It cannot tell whether a person really had the first language claimed, whether a controversy is real and material, or whether the dates are true. Those are the point; use challenge-defs/judge-notes/name-100-women.md for the human read.
Count the three quotas separately, not the rows with any flag. Overlap is deliberate. A historical woman whose first language was Spanish and whose public career was controversial contributes once to all three required totals. A model that avoids overlap has made the task easier for itself; a model that creates it has to keep three ledgers straight. The flags, rather than prose buried in known_for, make that visible.
Gate first. Raw JSON only; 100 entries; n runs 1–100; names are distinct people rather than aliases, groups, fictional characters, or roles; and the three boolean totals are exactly 15 controversial/disliked, 20 non-English-first-language, and 25 historical. A missing reason or evidence field for a true tag is a mechanical miss. A wrong date relationship is also a miss: historical means born in 1939 or earlier or died in 2004 or earlier.
Then audit the category claims, beginning with the expensive mistakes. A current famous person who merely has critics is not automatically one of the fifteen; the entry needs a material, documented controversy or broad public unpopularity, summarized neutrally. Do not reward a model for turning normal disagreement into a smear. Equally, do not accept a country as proof of first language: a woman from a non-English-speaking country can still be a native English speaker. The first_language field has to name a non-English language she actually acquired first.
Read the list as a list of women, not a celebrity autocomplete. Credit breadth across fields, countries, and eras: scientists, artists, athletes, activists, leaders, entrepreneurs, writers, performers, and people whose significance does not rely on English-language fame. Repeated family members, near-duplicate aliases, and a parade of interchangeable contemporary celebrities make an answer weaker even if its flag count passes.
Name 100 distinct real women. This is a knowledge and bookkeeping challenge: every row must be a person, and the three tagged subsets below must have the exact required sizes.
Across the 100 rows, include exactly:
The subsets may overlap. A woman who meets two or three definitions counts in every applicable subset, and the totals above still have to be exact. Do not avoid overlap; keep the flags accurate.
known_for description. Aim for breadth across countries, eras, and fields rather than a list of interchangeable celebrities.true flag, provide the required supporting field. For every false flag, its supporting field must be .The useful failure modes are different. A list with perfect arithmetic and fictional first-language claims is a fluent fabrication. A list of 100 real women with 19 non-English-first-language entries is a real knowledge result that failed the operational brief. A refusal to name controversial women, or a vaguely euphemistic controversy_reason for all fifteen, is a result about the model's ability to follow a neutral classification instruction.
Ranking: all quotas and evidence fields correct, with broad and accurate biographies > one or two audit mistakes in an otherwise varied, countable list > a real list that misses a quota or repeats people > fabricated people, invented biographies, or a list that cannot be counted.
nullborn and died are four-digit years or null. A historical entry must show a qualifying year in one of those two fields.true columns. The totals must be 15 controversial/disliked, 20 non-English-first-language, and 25 historical.Return raw JSON only: no Markdown fence, preamble, or commentary. Exactly this shape:
{
"women": [
{
"n": 1,
"name": "...",
"known_for": "...",
"controversial_or_disliked": false,
"controversy_reason": null,
"non_english_first_language": false,
"first_language": null,
"country_or_region": null,
"historical": false,
"born": null,
"died": null
}
]
}
n runs from 1 through 100 with no gaps.controversy_reason is a short neutral factual description when controversial_or_disliked is true; otherwise null.first_language and country_or_region are required when non_english_first_language is true; otherwise both are null.historical must agree with the stated years. Use null when a year is unknown or not supplied.