PropBetEdge Fight DNA, for developers.

What Fight DNA is

Fight DNA is the proprietary analytics layer above the normalized UFC record. It is a versioned, reproducible, as-of feature system that transforms event-dated bout and round history into fighter-specific and matchup-specific intelligence. It is deterministic: the same source rows always produce the same output. It is not a model, and it is not a prediction.

It answers questions a fighter page does not: how does a fighter do against southpaws, does pace survive into round three, is offense head-heavy or leg-heavy, does a takedown produce control, where do finishes happen, and which of those variables are unusually different between two fighters about to meet.

Source facts versus derived features

The API labels origin on every payload so you never confuse the two.

The feature pipeline

1. Source rows. Only canonical tables: fighters, events, bouts, results, round stats. Career snapshot fields are never inputs.
2. Per-bout features. Two rows per completed bout, one from each fighter's perspective: stance context, outcome, method, scheduled rounds, title and main-event flags, stat coverage, observed seconds, raw totals, derived per-bout values and provenance.
3. Observed time. 300 seconds per full standard round; a partial final round uses the recorded stoppage time; unrecognised time formats set observed seconds to null with a warning.
4. As-of snapshots. One snapshot per fighter per as-of date: metrics, stance splits, round profile, finish profile, context splits, provenance.
5. Matchup composition. Two snapshots compared against each other's actual profile with threshold-gated insights.

As-of reconstruction

as_of_date is exclusive: the snapshot dated D contains bouts with event_date < D, so a bout fought on D is excluded by the snapshot dated D. ?as_of=D resolves to the latest stored snapshot with as_of_date <= D, which therefore never contains a bout fought on or after D.

Use ?as_of=YYYY-MM-DD on any fighter DNA route (Pro and above) to reconstruct what was knowable before a given date. If no snapshot exists at or before that date the API answers 404 dna_not_available with the reason, never an empty profile. Example, captured live:

{
  "ok": false,
  "data": null,
  "error": {
    "code": "dna_not_available",
    "message": "No Fight DNA snapshot exists for Sean Strickland at or before 2024-01-01. Nothing is synthesized.",
    "detail": {
      "fighter_id": "ec94d296-2db3-4e0d-be6a-46de4f480672",
      "requested_as_of": "2024-01-01",
      "definition_version": 1,
      "reason": "no_snapshot_at_or_before_as_of"
    }
  },
  "meta": {
    "api": "PropSports UFC",
    "version": "2026-09-06.3",
    "request_id": "475fa213-d202-4353-a79d-d31519c3b4a3"
  }
}

Why sample size matters

Round-stat coverage is a backfill in progress, and many fighters have few covered bouts. A rate built on ten minutes of observed time is not the same as one built on ninety. Fight DNA therefore never returns a bare number: sample_bouts, sample_rounds and sample_seconds travel with every value, and the UI contract forbids turning a low-sample split into a categorical claim.

Example: Sean Strickland's sig_landed_per_min is 6.2327 per minute from 2 stat-covered bouts, 8 rounds and 37.4 observed minutes, confidence medium. That is exactly how it is reported.

Confidence

coverage_status uses the same scale and describes how much of the fighter's bout history has round-stat coverage. A zero denominator produces value: null with confidence: "insufficient".

Provenance

Every MetricObject carries: metric_key, value, unit, numerator, denominator, sample_bouts, sample_rounds, sample_seconds, confidence, coverage_status, definition_version, origin, source_families and as_of_date. Snapshots additionally carry the list of bout ids in scope, the source watermark and the build run id. The commercial gateway never strips any of it; it is part of what you pay for.

{
  "unit": "per_min",
  "value": 6.2327,
  "origin": "pbe_derived",
  "numerator": 233,
  "as_of_date": "2026-09-06",
  "confidence": "medium",
  "metric_key": "sig_landed_per_min",
  "denominator": 2243,
  "sample_bouts": 2,
  "sample_rounds": 8,
  "sample_seconds": 2243,
  "coverage_status": "low",
  "source_families": [
    "ufcstats"
  ],
  "definition_version": 1
}

Definition versioning

Every metric is defined in a registry row with definition_version, formula, unit, source families and minimum sample. A change to a definition is a new version, never a silent recalculation; snapshots carry the version they were built with, and GET /v1/ufc/dna/metrics?version= serves the registry for a version. The commercial contract pins the definition version it targets (currently v1) and a drift guard fails our build if it changes upstream.

What Fight DNA cannot tell you

• It does not predict outcomes, price a market or output a probability. Nothing in the API is a pick.
• It does not know what is not in the record. Bouts without round stats contribute results only; position data does not exist until a licensed source is added; injuries, weigh-ins and odds are not part of Fight DNA.
• It does not adjust for opponent quality yet. Opponent-adjusted metrics and archetypes are reserved for a later definition version.
• It cannot make a small sample large. Low-sample values are reported as low-sample values.
• It is not an official UFC statistic and must not be presented as one.

Unlock Fight DNA on Pro See Fight DNA in action →