UFC data and prediction methodology
Coverage and identity
Fighter identity, professional records, physical attributes and recorded bout histories are compiled from published MMA records. UFC performance statistics describe the available UFC bouts, not the athlete's entire professional career. Profiles use stable identity routes; similar names do not establish that two records belong to the same person.
A full career record and the wins, losses and draws counted from the available bout history can differ when that history is incomplete. We label those history totals as known bout outcomes. Missing records or metrics remain unavailable rather than being turned into zeros.
Reading the statistics
Significant strikes landed per minute measures landed significant strikes relative to recorded fight time in minutes. Absorbed strikes per minute uses the opponent's landed significant strikes. Accuracy describes landed attempts as a percentage of attempted strikes; defense describes the share of the opponent's attempts that did not land.
Takedown and submission-attempt rates are expressed per 15 minutes. Target breakdowns describe the share of recorded significant strikes aimed at the head, body or legs. UFC finish counts separate knockout or TKO, submission and decision wins. Total octagon time is recorded UFC bout time, not scheduled time.
Rates and percentages depend on their denominator. A zero can be a measured result, while an unavailable field means the observations required to publish that value are missing. Population percentiles compare the available numeric values in the database, rather than treating missing values as zero; division-specific comparisons are labelled separately.
Predictions and confidence
Verdikt AI publishes model outputs for moneyline, over/under rounds, over/under significant strikes, go the distance and winning method. Each prediction concerns its stated outcome and, where relevant, its stated line. A prediction for one line must not be compared with the result or odds for a different line.
Displayed confidence is the model's estimated probability for the selected outcome, rounded for display. It is not a promise, the historical win rate or evidence of profit. For example, a displayed 65% probability still leaves an estimated 35% chance of another outcome. Probabilities, odds and results must be assessed together.
Evaluation and limitations
The documented model evaluation protocol separates earlier training events from later evaluation events and keeps bouts from the same event together. Model selection, probability calibration and final evaluation have distinct roles. A historical holdout used repeatedly during research is not a fresh independent test of the selected model.
Probability quality is assessed with measures such as log loss, Brier score and calibration, alongside outcome accuracy. A sportsbook comparison requires the same fights, market definitions, decision timestamps and evaluation period. This page does not claim an independently established sportsbook advantage or guaranteed returns.
The public archive records published predictions and their available grades. It is an inspectable publication history, not a substitute for an independently held-out model evaluation. Model versions and data availability can change across the archive.
Archive results and odds
Each archive event exposes the available picks, confidence, grading and recorded public odds for all five markets. Correct and incorrect labels follow the recorded market grade. Draws, no contests, cancellations and unavailable grades remain distinct. Predictions for events that have not been fought stay pending on public archive pages.
Historical American odds are snapshots, not live prices. Positive odds indicate the profit on a 100-unit stake; negative odds indicate the stake required for a 100-unit profit. Line movement depends on the sportsbook, market, line and capture time. Missing odds are not reconstructed from model confidence.
Read event prediction records · Open the interactive archive
Medical and regulatory sources
Medical suspensions, injury withdrawals, anti-doping sanctions and testing activity are different record types. Each public entry preserves its date and attribution and links to the recorded source when available. Routine post-fight clearance requirements do not imply wrongdoing. Testing-session counts are activity counts, not positive tests or pass/fail results.
These records reproduce published information and may later be amended by the issuing body. They do not establish an athlete's current health or eligibility and are not medical advice.
UFC medical and compliance records · UFC Anti-Doping · USADA · California State Athletic Commission · Texas combative sports regulation
Freshness and corrections
Fighter profiles receive an individual modification date when their underlying content changes. That date describes the page's update, not necessarily the date of every observation on it. We do not assign a new date to every unrelated page because another fighter was updated.
A correction should identify the page, field, proposed value and supporting source. Promotion, event and regulator records help resolve factual disputes. Unverified values are not added solely to make a profile look complete.
