Editorial analysis · Manufacturer documentation checked November 17, 2026
You did not train today. Your training-load number moved anyway. Before treating that as a failed sync, a sudden loss of fitness or an instruction to squeeze in a workout, ask a less dramatic question: what aged out of the calculation?
A rolling training-load number is partly a description of the calendar. It depends on which sessions still count, how much weight they receive and what history the app uses for comparison. Time can change all three without a new workout arriving.
That does not make training load useless. It makes the definition indispensable. The trouble begins when an app compresses a history-dependent estimate into a number that looks as direct as a stopwatch reading.
We checked current Apple, Garmin and Polar documentation and built two transparent numerical examples. We did not conduct a device experiment, reproduce a proprietary algorithm or test whether any load target prevents injury. The examples explain arithmetic; they do not prescribe training.
“Seven days” does not identify one calculation
| System and field | Documented basis | Time context | Limit to keep visible |
|---|---|---|---|
| Apple Watch training load, watchOS 27 guide | Workout intensity and duration | Last seven days compared with previous 28 days | Do not infer the complete weighting formula from this guide |
| Garmin acute load, current technology description | Estimated EPOC-based exercise load | Activity influence fades over ten days; total normalized to a seven-day window | Older devices used a different seven-day total |
| Polar Strain and Tolerance | Daily Cardio Load, expressed as TRIMP | Seven-day and 28-day daily averages | Current status feedback uses more than their ratio alone |
Sources: Apple's watchOS 27 guide, Garmin's training-load description, and Polar's Training Load Pro documentation. Availability and behavior depend on the device and software; this is not a compatibility list.
Garmin's distinction is particularly useful. Its current description says a new activity contributes fully, then its influence expires gradually over ten days. It contrasts this with older devices that simply combined the past seven days. Reading “weekly load” as “add up seven visible rows” can therefore produce a discrepancy before a single sensor error enters the picture.
A 40% drop with no deleted workout
Here is a deliberately simple model, not an Apple, Garmin or Polar output. We invent eight consecutive days with arbitrary load units: 100, 0, 50, 0, 50, 0, 50, 0. Each value is the total assigned to that day. The zeros mean known zero activity in this illustration, not missing observations.
| Displayed day | Seven-day window | Sum | Daily average |
|---|---|---|---|
| Day 7 | Days 1 through 7 | 250 units | 35.71 units/day |
| Day 8 | Days 2 through 8 | 150 units | 21.43 units/day |
The rolling sum fell by 100, or 40%, because day 1 left the window and a zero entered. Every workout is still in the underlying history. The apparent collapse says nothing by itself about an overnight change in the person's body. It is exactly what this model was built to do.
A weighted system can make the same general phenomenon smoother: an older activity contributes less as it ages instead of disappearing at one boundary. That is a mathematical possibility, not our reconstruction of a vendor's undisclosed weights. We publish the simple model precisely because every step can be inspected.
The comparison can rise while today's load is zero
A relative value has a second moving part: its denominator. In a second hypothetical 29-day history, day 1 has 100 units, days 2 through 21 have zero, day 22 has zero, days 23 through 28 each have 50, and day 29 has zero.
On day 28, the seven-day total is 300 and the 28-day total is 400. Their daily averages are 42.857 and 14.286; the short-to-long average ratio is 3. On day 29, the short total stays 300, but the long total falls to 300 as the original 100-unit day leaves. The ratio rises to 4.
No new load was added. The recent window stayed constant. The comparison changed because the longer reference window lost a large day. These numbers are intentionally exaggerated to make the mechanism visible. They are not real training records, risk thresholds or a reason to increase or decrease exercise.
In both examples, interpreting the headline alone hides the cause. Keep the numerator, denominator and date range beside the ratio. If the longer average is zero, the ratio is undefined; a spreadsheet should not quietly replace it with zero or a comforting category.
Sometimes the label changes because the software did
Polar documents another complication: updated Cardio Load Status feedback considers absolute Strain and Tolerance and the preceding 14 days of status, rather than only their ratio. Some older watches still use the previous calculation, so their feedback can differ from current Flow feedback. Polar lists Flow 6.25.0 or later for iOS/Android and 7.25.0 or later for Huawei for the update.
That is a documented reason to record the app surface and version before comparing screenshots. A disagreement between a watch and a phone does not establish missing data. Conversely, a software explanation should not become an excuse to ignore an actually absent workout. Check the underlying record first.
Apple's guide also lets people filter training load by workout type and edit a workout's effort rating. Before comparing two views, confirm that they cover the same activities and that the inputs have not been edited. A comparison is only useful when its scope is stable.
An audit that starts with the history
Our proposed audit has four questions. Which field is displayed? Which dates and activity filters does it include? Did a workout arrive late or change? Did the app or device calculation change? Record the answers before deciding the chart is broken.
The distinction between a true zero and a missing session matters here. Our synthetic examples have complete histories. A real log may not. If a device was unworn or an activity failed to sync, you cannot assume its empty day means no exercise. Equally, finding the activity in the history does not prove it contributes to every derived metric.
A load chart should help you understand your record. It should not turn a declining line into a daily obligation to earn points. Whether to train depends on more than the arithmetic shown here; this article makes no individualized recommendation.
Inspect the definitions and reproduce the examples
Download the field dictionary, sourced JSON reference and complete synthetic daily records. The records include rolling totals, averages and ratios wherever a complete window exists. Blank means the full window is unavailable or the ratio is undefined, not zero.
For another metric whose definition travels badly, see why HRV numbers are not interchangeable. Vora's existing proactive AI brings wearable information together with nutrition and workouts for ongoing context. That does not make differently defined training-load scores convertible. Our product guides explain supported workflows without changing the definitions above.