Comparing This Month to Last Month in Performance Trends — Without Cherry-Picking
5 min read
Learn how to compare this month to last month without cherry-picking: lock metrics, smooth noise, tag confounders, and choose one clear next move.
On this page
Photo by Ricardo IV Tamayo on Unsplash.
Compare months objectively by predefining metrics, smoothing noise, and avoiding post-hoc window choices. That keeps one bad ride from steering the plan.
Simple month-to-month comparisons can mislead because you may choose the best ride, the worst week, or the neatest story after seeing the chart. A fair check uses the same metric, the same window, and clear notes on travel, sickness, sleep loss, equipment changes, or heat.
Photo by Markus Spiske on Unsplash.
A month is a short window, so random day-to-day swings can look like a trend. Your goal is not to prove a story; it is to tell signal from noise before you change the plan.
Regression to the mean is one common trap. If last month held an unusually strong test, this month may look worse even when your base state has not changed.
Repeated measures help because they lower the pull of any single ride. For broader context, compare this month against twelve weeks of trend data, not just one bright or bad day.
In N+One terms: your training system drift, not a single bad ride, usually explains month-to-month shifts.
Separate signal from noise before changing training.
Treat extreme months as suspect until repeated.
Use several matched rides, not one peak file.
Check the wider trend before judging the month.
This keeps the month-to-month check tied to a stable decision, not a sharp chart move.
Lock the comparison rule before you open the data. Choose one main metric, one backup metric, and the window you will use for both months.
Use rolling windows when calendar months do not match your training rhythm. A rolling view also helps after a rest block, especially when two easy weeks change the trend line.
Write down confounders instead of hiding them. Travel, illness, poor sleep, heat, indoor setup changes, and equipment swaps can bend the chart without meaning your fitness changed.
If a race or group ride skews the month, tag it rather than deleting it. A cleaner workflow is tagging outliers without distorting the trend.
Pick the main metric before looking.
Use the same window for both months.
Flag illness, travel, heat, and sleep loss.
Tag outliers instead of deleting them.
Keep the rule written for next month.
The rule protects the plan from being shaped by the ride you remember most.
In N+One terms: set the decision rules first, then let the data answer.
One tactical email with training ideas and product updates. No spam — unsubscribe anytime.
Define 1–3 primary metrics before looking at results (e.g., mean power, normalized power, TSS).
A difference matters when it is large enough, persistent enough, and linked to the training context. If it only appears in one file, treat it as a clue, not a verdict.
Look at load and freshness beside the performance metric. A drop after a heavy block means something different from a drop while form and freshness both look steady.
Separate volume from intensity before you act. A lower monthly mean may reflect more long endurance work, while the top end still looks sound on your power-duration curve.
Your next move should be small enough to test. Keep intensity, cut volume modestly for seven days, then reassess with the same locked rule.
Check whether the change repeats.
View load beside the performance metric.
Separate volume drift from intensity drift.
Make one small training change.
Recheck with the same rule.
Keep reading
- Comparing Two Workouts Side‑by‑Side in N+One: A Cyclist Walkthrough — Step through a focused side-by-side comparison of two rides in N+One, compare the same metrics, check context, and choose one clear next move.
- Mid-Season FTP Check‑In in N+One: How the App Detects a New Threshold Without a Test — How N+One flags a mid-season FTP shift from recent ride power, heart-rate, and recovery context, plus a short micro-check for mixed signals.
- N+One Power Curve vs Strava Power Curve: Why the Numbers Differ — Why N+One and Strava power curves can differ, what processing choices change the numbers, and a 7-day protocol to compare matched ride files.
Field data is useful, but it is not perfect. Power meters, route choice, wind, indoor settings, and ride type can all change what the numbers show.
If the stakes are high, simple chart checks may not be enough. Formal analysis, coach review, or repeat testing can help when small changes carry real planning weight.
Be careful with physiology claims from noisy field data. You can say the trend changed; you may not know why it changed without better controls and stronger evidence.
When one workout looks odd, inspect the file before changing the month. Start with single-workout pacing and variability, then return to the wider trend.
Check device and setup changes.
Review odd files before judging trends.
Ask for review when stakes are high.
Avoid strong physiology claims from weak data.
Day 0 — Define and lock metrics: Pick one primary metric, commit to a 28-day comparison window, and flag non-training confounders before judging the chart.
Days 1–7 — Reduce noise, preserve intent: Maintain the planned intensity mix, reduce total volume modestly, and keep the core training stimulus simple.
Days 8–14 — Reassess with rolling windows: Compute the new rolling average and compare the change against your own recent month-to-month spread.
Decision point: If the change persists beyond your normal spread, make a targeted block or recovery adjustment. If not, resume the planned progression.
Compare months objectively by locking the metric and window first, smoothing the noise, tagging confounders, and making one small adjustment only when the change persists.
Use rolling windows when your training weeks do not line up neatly with the calendar. Calendar months are easy to explain, but rolling windows often give a fairer view.
No. Best rides are useful, but they are also prone to cherry-picking. Use repeated measures and keep peak efforts as supporting context.
Keep the data, but tag the confounder. Silent deletion makes the comparison harder to trust, while notes help you judge whether the change reflects training.
Keep the same comparison rule, reduce training strain briefly, and reassess. Your threshold did not disappear; your recovery inputs may have shifted, so the output dropped.