From 112 Firstbeat Metrics to the 38 That Actually Matter (and the 10 That Matter Most) — Ice Hockey Edition

Wearables and GPS platforms like Firstbeat are incredible at capturing everything — and that's the problem.

If you've ever opened a Firstbeat export and stared at the ocean of columns, you already know: more data doesn't automatically mean more insight. In fact, it usually means more noise, more false confidence, and more arguments about "what matters."

So we did the hard part.

The Core Idea

Of the 112 metrics available in the export, 38 are meaningfully tied to on-ice performance capacity (shift intensity + repeatability), fitness adaptation, and readiness/recovery — so we include the ones that matter and discard the ones that don't (metadata, redundant fields, or signals that don't reliably track performance).

This isn't about making a prettier spreadsheet. It's about turning wearable data into something that can actually help coaches and programs predict performance and reduce guesswork.

Why Most Wearable Exports Fail in Practice

Most teams fall into one of two traps:

"Include everything."
You end up with a dashboard that looks impressive but can't answer simple questions like: Who is trending up? Who is ready? Who can sustain shift intensity?

"Pick a few favorite metrics."
Often based on tradition, not evidence. Different staff members choose different stats, and the scorecard becomes subjective.

The truth is: a large share of metrics in a wearable export are just different ways of describing the same thing, or they're administrative fields that have nothing to do with performance.

The 38 Metrics We Keep in the StatLink Upload (and Why)

We selected 38 because they map cleanly into three buckets that actually drive hockey outcomes:

1) Shift Intensity + Repeatability (the hockey-specific engine)

These capture the ability to hit high gear in short windows and sustain it over repeated efforts — closest proxy to "hockey fitness."

Examples:

• Maximal Intensity Periods (MI 30s / 45s / 1min / 2min)
• Maximal Intensity Period TRIMP/min windows
• High intensity time
• Movement intensity

2) Fitness Adaptation (how training changes the athlete)

These measure stimulus/strain in ways that correlate with changes in capacity over time.

Examples:

• TRIMP, TRIMP/min
• EPOC, EPOC Peak
• Aerobic & anaerobic training effect
• Threshold / zone time

3) Readiness & Recovery (availability + next-day output)

These are the "can they perform today/tomorrow" indicators — often the difference between strong and flat performances.

Examples:

• Heart rate recovery (30s/60s/120s)
• HRV markers (RMSSD, SDNN)
• Training load (acute/chronic) and workload ratio (ACWR)

We discarded the rest because they're typically: metadata (start time, end time, notes, analysis period), redundant fields, or noisy signals that don't consistently track performance.

The "10 That Really Matter" (the Core Performance Capacity Set)

After we built the 38-metric foundation, we ran an evidence-based reduction to isolate the smallest set that still captures most of the signal.

These are the 10 metrics that consistently carried the most predictive weight:

1. Maximal Intensity Period MI 30s
2. Maximal Intensity Period MI 1min
3. TRIMP/min (Index)
4. Average movement intensity
5. Movement load
6. EPOC (ml/kg)
7. Heart Rate Recovery (%) 60s
8. RMSSD (ms)
9. Chronic Training Load
10. ACWR (scored as a target band, not "higher is better")

If you want a clean headline from this entire project, it's this:

Out of 112 available Firstbeat metrics, only ~38 are worth ranking, and the top ~10 drive most of the performance signal.

That's why wearable data feels "messy" for everyone right now: the noise-to-signal ratio is brutal unless you reduce it deliberately.

How We Validated Correlation to On-Ice Performance

We didn't just pick metrics that "sound right." We tested them.

We compared Firstbeat physical performance to on-ice performance rankings and used multiple methods so we weren't relying on one fragile approach.

Models / methods used:

• Metric-by-metric correlation testing
• Spearman rank correlation (ideal when you're comparing rankings)
• Pearson correlation (helpful but less robust in sports data)
• Regularized regression models (to identify what actually matters when everything overlaps)
• Ridge regression (stabilizes coefficients when metrics are correlated)
• Lasso regression (forces feature selection by shrinking weak metrics to zero)

This allowed us to do two things:

• Confirm the 38-metric set is directionally aligned with on-ice impact
• Isolate the "top 10" that carry most of the predictive signal

What This Means for StatLink + Firstbeat Users

If you're a Firstbeat program, this is the unlock:

You can now upload your wearable export into StatLink and get:

• A clean, ranked scorecard that emphasizes performance capacity (not noise)
• A smaller core score driven by the "10 that matter"
• The ability to compare athletes across time windows with consistent scoring
• An advantage in connecting physical output to on-ice performance trends

In other words:

Firstbeat users can now bring their training data into StatLink to gain an edge in predicting on-ice performance — without drowning in 112 columns.

The Bigger Point: The Industry's Problem Isn't Data — It's Signal

Everyone has wearables now. Everyone has dashboards. Everyone has exports.

But if you're ranking athletes on fields that don't correlate, your scorecard becomes:

• Inconsistent across coaches
• Hard to defend
• Impossible to benchmark
• And worse: it can push training decisions in the wrong direction

That's why we built this the way we did:

• Keep 38 for a robust upload and reporting layer
• Isolate the 10 to drive the core "Performance Capacity" signal

Closing

The future isn't "more metrics." It's fewer, better metrics, validated against outcomes.

112 available doesn't mean 112 useful.
38 correlated is already a huge reduction.
And the 10 that matter most are what turns wearable noise into a competitive advantage.

If you're using Firstbeat and want to connect training data to on-ice performance, StatLink now gives you a way to do it — cleanly, consistently, and defensibly.