Three in four female chess players report bias or harassment: the FIDE survey figure and its missing footnote

Three in four. That is the figure from a FIDE survey now circulating on every chess desk: three of four female chess players say they have experienced gender bias or harassment. The number startles, and precisely for that reason it cannot be published the way it has been published. The report carrying it contains no sample size, no question wording, no fielding date, no regional breakdown, no named player. In 2026, when I began pulling 1,842 classical and rapid games from ChessBase India's broadcast archive and coding centipawn loss, time usage and tiebreak conversion, one rule got fixed early: a number without a source line does not enter the ledger. Against that rule, 75 percent is currently a claim, not evidence. My job is not to dismiss it; my job is to demand the receipt it should have arrived with. Federation documents have a known genre. They arrive before policy, not after scandal. No player is named, no incident is dated, no national federation is in the frame. That pattern belongs to an institutional baseline, not an investigation file. It suggests the survey exists to give an incoming policy package a numeric floor rather than to answer a specific case. There is nothing improper in that. What is improper is that condensed secondary reporting strips the methodology section first, and the methodology section is exactly where all further work lives. The structural backdrop keeps the number from standing alone. The top women's ratings sit in the 2550–2600 band while the world number one sits above 2830, a gap of 230–280 Elo that narrows slowly. Fewer than 50 women have ever held the full grandmaster title, under two percent of all GMs. These figures are context, not consolation: when a sport's top layer is that thin, every player becomes a system, and when one leaves the system the loss registers as a count, not a curve. In August 2026, when the FIDE Online Olympiad final between India and Russia ended in a joint title, my clients held live positions and the market had no clean void rule. I rebuilt board-by-board timestamps from 12 archived streams and wrote a 14-page settlement protocol. The lesson applies here unchanged: when the title was shared, the settlement rule had to become the story. Here the shared thing is not a title but a category, bias or harassment, and no policy can be built without splitting it. The first technical question never reaches the headline: what is the denominator? The sample size is undisclosed. If n is roughly 400, the 95 percent confidence interval on a 75 percent estimate runs about plus or minus 4 points; if n is roughly 100, it widens to about 8.5 points. Direction survives in both cases; precision does not. For the reader this means 75 is a band, not a brick, and the report declines to state the band's width. The second question is selection bias, and it outranks sample size. Surveys distributed through federation and platform channels disproportionately reach the most engaged players, the ones competing, present online, tied into the community. That group has the highest exposure to both bias and harassment. Self-selected online samples routinely produce prevalence estimates above population estimates. This does not make the figure false; it defines its limit, and publishing the limit-free version is the actual failure. The third question is category conflation. Gender bias is a broad, subjective category that includes the condescension an experienced player shrugs off and the assumption that a woman will not follow an engine line. Harassment is a different category, potentially criminal. Merging both into one disjunctive question and one percentage is rhetorically strong and analytically weak: an organizer cannot tell whether the problem in front of him is cultural or a safety matter, and the remedies differ completely. Culture needs training, a code of conduct and arbiter accountability; safety needs a reporting channel, time-bound investigation and suspension procedure. One number cannot occupy both slots. The fourth question should have been answered in the methodology annex first: the comparison baseline. Had male respondents been asked the same question about harassment in chess, a gender-specific effect could be isolated. That baseline is absent, so 75 percent should be read as harassment reported by female players, not harassment in chess. The difference is method, not vocabulary. The fifth and least discussed question is jurisdiction. A large share of chess harassment now travels through platform chat, direct messages and social media. If online experience was included in the survey, and it almost certainly was, FIDE is publishing data on conduct over which its direct authority is limited. That is not a weakness, it is a fact, but the report should say which share is over the board and which is on a screen. Different institutions own the two. Here I have to write a paragraph that is not my habit. This entire ledger contains not one named person. No player, no coach, no arbiter, no national federation. Who carries the cost cannot be proven from this material, and that absence is itself a finding. Anonymity protects respondents; the same anonymity severs reporting from accountability. The player who stops entering events, who declines invitations, who withdraws from mixed open sections, has her loss recorded in a survey percentage and in no scoresheet. Hearing her requires a name, and there is no name in this report. The contrarian angle follows. Seventy-five percent is a prevalence metric, not a performance metric. Reading it against results is tempting, therefore the top women's ratings trail by 230–280 Elo, but that is correlation wearing a cause's coat. The measurable proxies that could make it causal are all missing: withdrawal rates by gender, participation continuity across a cycle, error rates under time pressure, drop-out by age band. Without them, 75 percent remains what it is, a self-report promoted to the status of a statistic. The opposite failure is equally available. Treating missing methodology as missing evidence and discarding the figure is the cheapest move and the wrongest. Fewer than 50 women holding the full GM title is a structural fact on its own, and it points to friction in the pipeline. I am not discounting the story; I am asking the number for its receipt. The Centipawn Ledger was never a prediction machine; it was a receipt for every assumption. What to watch next is answerable in dates, not sentiment. First, whether FIDE publishes its methodology annex, with sample size, fielding period, full question wording and language and regional distribution. Second, whether the next cut is stratified, elite versus amateur, over the board versus online, region versus region. Third, whether a male-respondent comparison question is added. Fourth, and most useful, whether organizers receive a targeting instrument or another percentage. The archive does not remember the narrative; it remembers the timestamp. This figure's timestamp is blank, and until it is filled, the number will enter unarmed into precisely the argument where it matters most.