Training vs. Sleep in Professional Cycling

Repo: tercasaskova311/training-vs-sleep

I used to race professionally, which meant living at the edge of what my body could recover from. That’s why this question keeps interesting me :

How tightly does sleep relate to next-day training load for a World Cup MTB rider?

If sleep truly supports training, the training upward after good sleep and sags after poor sleep.

Data & pipeline (what I joined)

• WHOOP: sleep performance (% of need), recovery, strain, SWS (deep sleep).
• Garmin: ride metrics and TSS (Training Stress Score).
• ETL: Excel → SQL standardization → Python (pandas/matplotlib) for analysis and plots.

Key terms

• TSS: combined stress from duration × intensity (via NP, IF).
• Sleep Performance: % of sleep obtained vs. estimated need.
• SWS (Deep): restorative stage tied to physical recovery.

• Sleep quantity & quality
  Sleep duration ↔︎ Sleep Performance r ≈ 0.92; Sleep ↔︎ Deep (SWS) r ≈ 0.74; Sleep Performance ↔︎ SWS r ≈ 0.72.
• TSS shows weak links to sleep metrics.
  In this sample, TSS has very weak correlations with sleep duration, Sleep Performance, and SWS.
• Practical read: Sleep metrics track each other well (good), but don’t expect a simple “slept more → trained harder” relation day-to-day.

What the scatter actually says

Picture a scatter of Sleep (x) vs TSS (y). If points form a clean rising stripe, sleep is a strong next-day driver. In my data, the stripe just isn’t there—TSS looks almost orthogonal to sleep.
However, sleep metrics agree with each other: more total sleep → higher Sleep Performance → more SWS. That’s consistent with physiology and makes the signals trustworthy.

Takeaway: Sleep is coherent; TSS is influenced by plan, terrain, goals, and race calendar, not just last night’s hours.

Real-life coaching cues (actionable)

• Green-light intensity after a sleep window, not a single night.
  Use a 3–7 day average of Sleep Performance or SWS as your “go/no-go” for VO₂/neuromuscular days.
• Let sleep gate progression, not dictate daily TSS.
  Keep the plan; adjust set count or recovery between sets if the rolling sleep window is low.
• Watch sleep debt.
  If 7-day Sleep Performance < ~85%, expect stubborn legs and flatter power targets—even if yesterday’s sleep looked fine.

Methods (how I tested the hunch)

• Correlations: Pairwise Pearson among Sleep, Sleep Performance, SWS, TSS.
• Visuals: Scatterplots with LOWESS lines; correlation heatmap.
• Sanity checks: outliers, missing days, and device export quirks (WHOOP vs Garmin timestamps).

Why this sequence? First see the geometry (scatter), then the summary statistic (r). The geometry guards against being fooled by a single number.

What this doesn’t claim

• No causality. Training plans, blocks, and races set TSS targets; athletes don’t pick intervals because they slept 7h vs 8h.
• Single-athlete scope. Effects are often athlete-specific; don’t generalize across the peloton without replication.
• Acute vs chronic. One-night sleep blips ≠ chronic sleep patterns. The body responds to patterns.

Conclusion

Sleep metrics cohere; TSS doesn’t cleanly track them day-to-day. For athletes and coaches, that’s good news: you can trust your sleep dashboard for readiness trends, but keep prescribing training from the plan, only trimming or swapping sessions when rolling sleep flags accumulating debt.

Next step I’m exploring: does rolling SWS predict session quality (e.g., % of target power achieved) better than it predicts TSS? That’s closer to what athletes actually care about.