Do not update only the displayed decimal. Preserve raw results, contested-bout state, defaults, absences, denominator formula, interval and extraction version together. Tournament progression, official display updates, transcription repairs, interpretation corrections and redefinitions are distinct.
During a tournament, one day of results arrives at a time and withdrawal or return can change exposure. Silently replacing yesterday’s provisional percentage prevents readers from reproducing what an earlier article knew.
Three linked layers
The raw bout table stores tournament, day, wrestler IDs, opponent, result, default, kimarite, retrieval time and official URL. The state table stores withdrawal, return and no scheduled bout. The aggregate table stores formula, win and loss counts, denominator, interval, raw-table hash and code version. Freeze the population by rank, tournament and extraction cutoff; count fields absent from the official record as missing. Treat an active or in-progress record as right-censored rather than replacing unplayed days with losses or zeroes.
The atomic unit of a win-rate update is one bout row. Validation stops publication when wins plus losses do not reconcile with the declared denominator, when the scorecard and daily results have an unresolved difference, or when raw evidence changes without a new aggregate release. Quarantine that bout and every affected denominator, then recalculate official-decision, contested-bout and opportunity rates separately from the same raw hash and name the formula that changed.
Reasons and affected outputs
Use separate codes for tournament progression, following an official-page update, transcription correction, result-state interpretation correction and win-rate definition change. Removing a fusen decision accidentally counted as a contested bout is interpretation correction. Adding absences to a denominator is a definition change.
One repaired row can alter rankings, averages, the ARTICLE_002 reproduction, article prose and structured data. Rebuild the impact list. If the conclusion changes, preserve the former conclusion and explain why.
Current public baseline
- 3 September 2026, v1.0: competed-bout, official-decision and ranked-exposure rates are separate formulas.
- 3 September 2026, v1.0: raw results, defaults, absence states and aggregate formulas occupy separate layers.
A monitoring run that finds no difference creates an audit log but does not move the public modified date. Copyediting and a numerical release remain separate so the same formula always reproduces the same value.
Worked example: which percentage changes when one default win is corrected?
This is a fictional example of a correction. It is not a real wrestler’s record, a correction made to actual tournament data, or a statement of the Japan Sumo Association’s win-rate definition.
The record contains seven contested wins, one default win, five contested losses and two days absent: eight wins, five losses and two absences in total. There are no default losses. The same record answers three different questions.
| Question | Calculation | Result |
|---|---|---|
| What share of the thirteen recorded decisions were wins? | 8 / (8 + 5) | 61.5% |
| What share of the twelve bouts actually contested were wins? | (8 − 1) / (8 + 5 − 1) | 58.3% |
| On what share of the fifteen days was a win recorded? | 8 / 15 | 53.3% |
Now suppose that the default win was mistakenly transcribed as a contested win. The proportion won in bouts actually contested would have been reported as 8 / 13, or 61.5%. Correcting that entry changes it to 7 / 12, or 58.3%. The total of eight wins and thirteen recorded decisions has not changed, so the first percentage remains 61.5%. The share of the fifteen days with a win also remains 53.3%.
The example correction note can therefore say: “One default win was transcribed incorrectly; contested bouts changed from 13 to 12 and contested wins from 8 to 7; the other two percentages did not change.” A correction to a real record would also identify the tournament, day, wrestler and checked source record.
The last percentage is the share of days with a win; it does not turn absences into losses. None of these percentages accounts for opponent strength or the action within a bout. Use the same formula and period when comparing records, and mark a zero denominator as not applicable rather than 0%. See the win-rate calculation guide for the basic method.
Return the decimal to a bout
When a win rate moves by one digit, find the bout responsible in the official results and watch the first step. Even a single unit in the numerator contains an opponent and kimarite. The correction log protects more than a decimal: it is a signpost back to the wrestler behind the aggregate.
A Comparison Does Not Prove a Cause
A visible group difference may coincide with era, rank, opponent strength, absences and recording detail. This comparison alone cannot conclude a causal explanation. Try matched conditions, another defensible denominator and a missing-data sensitivity run as prespecified falsification checks. If the pattern disappears, retract the first story; if no specification is comparable, the alternative conclusion is that the cause remains unknown.
Sources
- Japan Sumo Association tournament scorecard, daily bouts/results and absence information (checked 3 September 2026).
Source links used for verification
Each link was used for the corresponding claim in this article; inclusion is not a blanket endorsement of the whole page.
Article verified:
- Japan Sumo Association tournament scorecardPrimary source (as checked for this article)
- Japan Sumo Association daily bouts and resultsPrimary source (as checked for this article)
- Japan Sumo Association absence informationPrimary source (as checked for this article)