Bad Data at the Source: Mislabeled, Missing, and Miscounted

The garbage-in problem

No analysis survives bad inputs. Three errors account for most corrupted scouting datasets, and all three are preventable at the form and process level.

1. Wrong team (mislabeling)

A scout watches robot 1678 but records it under 1768. Now both teams’ averages are wrong and you may never notice. Fixes:

• Make team selection a pick-from-schedule field, not free text. In QRScout use the TBA-team-and-robot field type; in ScoutingPASS (PWNAGE Robotics, FRC 2451) pre-load the event schedule. The scout taps “Red 2” and the app knows the team number.
• Assign each scout a fixed station (e.g., “always Red 2”) for a whole match block, so attention is on one robot, not on remembering numbers.

2. Missing matches

A scout misses a match (bathroom, dead tablet, confusion in the stands) and you get a hole. One missing match in a 3-match sample swings an average wildly. Fixes:

• Run a completeness check after every few matches: count rows per match; you should have exactly 6 (one per robot). A pivot table COUNTIF(match) that highlights any match without 6 entries catches gaps in seconds.
• Keep paper backups at every station so a scout with a dead device keeps recording.

3. Miscounting

During a fast REEFSCAPE cycle a scout records 4 L4 coral when the robot scored 5, or attributes a partner’s coral to the wrong robot. Fixes:

• Tap-as-it-happens counters (QRScout counter/multi-counter) beat tallying from memory after the buzzer.
• Reduce cognitive load: one scout tracks fuel, another tracks Tower/endgame if you have the people, rather than one scout tracking everything.

A debugging workflow when numbers look wrong

When a team’s stats look implausible, do not delete data; trace it:

1. Reproduce: pull that team’s raw rows and find the outlier match.
2. Cross-check against TBA: compare the alliance total your scouts recorded for that match against the TBA score_breakdown. If the alliance sum is off by far more than match-to-match noise, a scout miscounted or mislabeled.
3. Isolate the scout: which station’s row is the outlier? Look at that scout’s other matches; a systematic bias (always low on L4) points to a training gap, not a one-off.
4. Fix the cause, not the cell: retrain or reassign the scout, add a validation rule, then re-run aggregation. Editing one number without finding the cause guarantees the error returns.

Build validation into the sheet

Add a column that flags impossible values: more L4 coral than the 12 branches a reef level physically has, a negative count, or an alliance scouted-sum wildly above the real match max. Conditional formatting that turns these red means errors announce themselves instead of hiding in averages.

Key takeaways

• The three killers are wrong-team labeling, missing matches, and miscounting; each has a specific form/process fix.
• A per-match completeness check (exactly 6 rows) and a TBA score_breakdown cross-check surface holes and outliers fast.
• Debug data by tracing the outlier to a scout/station and fixing the root cause, never by silently editing the cell.

This lesson was adapted from learnfrc.com.