Your Scouting Data Lies to You

Symptom: Your pick list says a team is great, you pick them, and they underperform in elims. Or two scouts hand you wildly different numbers for the same robot. The data betrayed you.

Failure 1 — inconsistent scouts. If two people score the same robot's FUEL count 30% apart, your averages are noise. Diagnose: double-scout one robot in one match from video and compare. Fix: tighten field definitions until any two scouts agree within about 10–15%. Ambiguity ("was that a score or a miss?") is the enemy; a countable, unambiguous sheet beats a detailed one nobody fills the same way.

Failure 2 — survivorship/sample bias. A team looks mediocre because they played matches with a partner who hogged the active HUB, or great because they always played weak opponents. Diagnose: look at who they played with, not just their averages. Fix: this is exactly why OPR exists — it mathematically separates a team's contribution from its partners. Cross-check your averages against TBA OPR; big disagreements flag a sample-bias problem worth a closer look.

Failure 3 — over-trusting a single number. OPR is a least-squares prediction, not measured truth; it can be inflated by alliance context and it can't see reliability. EPA is more interpretable (point units, split into auto/teleop/endgame components) but still can't see that a robot died twice. Diagnose: any team whose public rating is high but whose died-count in your own data is also high. Fix: rank by analytics, then have a human read the reliability column and the pit notes before finalizing the pick list. The number narrows the field; the human catches the landmine.

Failure 4 — stale or wrong-event data. Pulling last week's numbers, or the wrong event key, silently poisons everything. Fix: always confirm the event key (year + code, e.g. 2026wabon) and re-pull before alliance selection, not the night before.

The discipline: treat scouting like any data pipeline — validate inputs (scout agreement), check for bias (who they played with), triangulate sources (your data + OPR + EPA), and keep a human in the loop for reliability. Public analytics and hand scouting each see what the other can't; trusting only one is how good teams make bad picks.