The same interval,
eight weeks apart.
You ride the same session every week, and one block in it is the one you care about: the second set of sweet spot, the last threshold rep. You want to know whether that block is getting easier. The whole-ride summary is too coarse, because a long warm-up and a slow ride home move it more than the block does. A chart of every interval in every ride puts that block in the same cloud as the cool-down.
What answers the question is one block, followed across the weeks, compared only where the comparison is fair. This article is about doing that, and about how much change is too small to mean anything.
Choose the block, then the sessions
If the session is a structured workout in Intervals.icu, each interval in the ride carries the label the workout gave it, and that label is how the same block is found again next week. A ride without that structure has no labels to match, and a trainer ride that arrived as one long interval has nothing to choose from at all; the steady efforts in it have to be found from the power trace first.
Then decide which weeks count. A repeat belongs in the series when it is:
- the same block, by its label, not by its position in the ride;
- within about 10% of the same duration, because a longer block reads differently on anything heart rate touches;
- at a similar intensity, because the same label at a different target is a different effort;
- on sound data, with no flagged power fault inside it.
Weeks that fail are left out and listed, with the reason. Six sessions with two left out is a better series than six with a mismatched pair quietly included.
The numbers worth following
For each interval, IntervalSync returns its label, duration, average and weighted average power, minimum and maximum power, intensity, training load, zone, and average, minimum and maximum heart rate and cadence, with a warning on any interval that spans a flagged power fault. From those, three comparisons hold up across weeks:
- Heart rate at the same power. For a block held at a target power, the heart rate it took. Lower, over several weeks, is the usual sign the effort is costing less.
- Power at the same heart rate. For a block ridden by feel or to a heart-rate cap, the power held.
- The settled end of the block. Heart rate lags a change in power by 30 to 60 seconds, so the first minute of a block mostly measures the lag. Comparing the last minute of each repeat, read from the raw samples between the interval's start and end, removes most of it.
Aerobic decoupling is the number people usually reach for here, and it is the one most easily fooled across sessions: a block that runs a minute longer reads better whatever your fitness did. Decoupling and block length shows the arithmetic.
Compare the "Set 2 sweet spot" interval across my last six sessions. Before reporting anything, check the blocks are comparable: same label, duration within 10%, similar intensity. If they are not, say so and do not report decoupling. List every block you excluded and why.
How much change is noise
Heart rate on the same block, on the same rider, moves by a few beats from one day to the next for reasons that have nothing to do with fitness: the room's temperature, the night's sleep, caffeine, what you ate and when. A four-beat drop between two sessions is inside that. The same drop holding across five or six sessions, at the same power, is a pattern; one pair of sessions is not.
So ask for the series rather than the pair, and for the spread as well as the trend. An answer that says "the last three weeks are lower than the first three, by about as much as they vary week to week" is more honest than a percentage, and more useful.
What it cannot tell you
- Why it changed. Training data does not record fatigue, illness, heat or a bad night. A trend can be adaptation, or a cooler month.
- A block with no label. Without the workout's structure there is no label to match, and blocks have to be found from the trace and matched by duration and power instead.
- What to do next. This is about what the data can support, not about what to ride on Tuesday.
Each session's intervals are a request to Intervals.icu, and checking their power for faults is one more, so six sessions are about twelve. On IntervalSync's Free plan, after the trial, each tool reads from Intervals.icu once a day, so a series like this needs the trial or a paid plan; pricing sets out each plan's limit.
More questions the data can answer · Decoupling and block length · All articles · Connect your training