Skip to content
Scouting & Strategy·Lesson 25 of 32

When OPR and EPA Mislead You

Public analytics are powerful but have well-known failure modes; learn to recognize defense distortion, small samples, blowouts, and shared-task confounds.

Sign in to track progress, earn XP, and save lessons.

Powerful, not infallible

OPR (The Blue Alliance) and EPA (Statbotics) let you rank teams with zero scouting effort, which is exactly why people over-trust them. Knowing where each breaks keeps you from a bad pick.

OPR's blind spots

  • Defense is invisible (or negative). OPR only models offensive point contribution. A robot whose job is to shut down the opponent's L4 scorer adds nothing to its own alliance's offense, so its OPR can be low or even negative, while its real value is huge. If you pick purely on OPR you will pass over the best defender in the event.
  • Small samples are noisy. Early in an event, after only 2-3 matches, OPR swings wildly because the linear system barely has enough equations. Do not lock a picklist on Friday-morning OPR.
  • Strong/weak partners contaminate it. OPR tries to separate contributions, but a team that always plays with powerhouses can inherit inflated numbers until the math has enough mixed pairings to untangle them.
  • Shared/cooperative tasks confound it. When two robots feed one scoring action, OPR cannot cleanly attribute credit, so component OPR for those tasks is shakier than it looks.

EPA's blind spots

EPA is designed to be more stable and predictive than OPR and includes auto/teleop/endgame and ranking-point components, but it is still a model of expected points, not a scouting report:

  • It does not know a robot's L4 mechanism jammed in its last two matches; it sees the lower scores but not the why or whether it is fixed.
  • It cannot capture driver skill under pressure, reliability of a climb, or whether a team plays defense well.
  • Like OPR, early-event EPA is less certain; Statbotics weights recent matches more heavily so a robot's rating can move as it improves.

A cross-check workflow

Use the public numbers as a triage tool, not a verdict:

  1. Agreement = confidence. When your scouted average, TBA OPR, and Statbotics EPA all agree, you can trust the ranking and spend attention elsewhere.
  2. Disagreement = investigate. When they diverge, that team needs a super-scout. Common causes: a defender (low OPR, high real value), a robot that got better mid-event (rising EPA), or a scouting error (your number is the outlier).
  3. Never pick a defender from OPR. Defensive value comes only from watching matches and from opponent-score impact, so keep a separate, scouting-driven defense ranking.

Blowout and bonus distortion

In lopsided REEFSCAPE matches a dominant alliance may stop scoring once the win and RPs are secure, depressing their measured output, while a team grinding in garbage time looks better than it is. Always read a robot's capability (best observed performance, mechanism reliability) alongside its averages, because averages flatten exactly the situational nuance that wins elims.

Key takeaways

  • OPR cannot see defense (low/negative OPR for great defenders), is noisy on small samples, and is confounded by partner strength and shared tasks.
  • EPA is more stable and component-split but is still expected points, blind to mechanism failures, driver skill, and climb reliability.
  • Treat agreement among scouting/OPR/EPA as confidence and disagreement as a super-scout trigger; never rank defenders by OPR.

Go deeper

Lesson quiz

Required

Answer all 3 questions correctly to complete this lesson.

1.A core assumption of OPR (Offensive Power Rating) that makes it mislead in some games is that:

2.Why can OPR be especially unreliable early in an event?

3.Statbotics' EPA improves on OPR primarily because it:

Answer every question to submit.