You open the export of a run and find a one-second gap after the first point, four seconds after the next and then two seconds again. Every row in the table has the same height, yet the graph draws a smooth curve. It is easy to infer that the points were created at regular intervals too. Recording, storage and display are three different operations, however.

Smart recording may mean precisely that records are not written to the file at equal intervals. The series can still be useful, but needs reading differently from regular every-second samples. Original timelines show why timestamps matter, where an average can go wrong and what graph smoothing does not restore. The aim is to understand the export, not prescribe one mandatory setting for every watch.

A row number is not a unit of time

The Garmin FIT Activity description explains that Record messages store instantaneous data with timestamps. Timestamp resolution may be one second while the recorder writes records less frequently; smart recording can produce irregular intervals. The file format therefore does not guarantee exactly one second between successive rows.

In our original example, the first five records occur at zero, one, five, seven and eight seconds from the start. There are five rows but a timeline spanning eight seconds. If a spreadsheet treats each row as a second, the event at the seventh second is already misplaced. The series remains in order, but the time axis assigned to it becomes distorted.

This matters particularly when placing two devices’ curves side by side. The tenth record does not necessarily represent the same moment in both. First align actual times rather than row numbers. Otherwise, a sensor can appear to respond later simply because its data was saved less often. A recording difference can easily lead to a mistaken judgement about a device.

What does smart recording mean in the manufacturer’s description?

The official Garmin recording information describes smart mode as saving points in response to changes in direction, speed, heart rate or elevation. Every-second mode can provide a denser track log. Available settings vary by model, so your watch’s own manual is needed for the precise choice. The name “smart” does not mean the same algorithm across manufacturers.

The interval between saved points and the sensor’s internal operating frequency are not necessarily the same either. Sparse records in a file alone cannot reveal every internal sensor measurement. Equally, a display updating every second does not prove every displayed value appears in the export as a separate record saved at that same moment. Always identify where the observation was made.

An original digital-note analogy: you can observe an event continuously while writing about it only occasionally. A later reader sees the written points, not the entire observation. This analogy is not a claim about the specific hardware operation of a watch. It helps separate what may have been measured from what the transferred file actually retained.

While reviewing training-log data, a runner adds a separate note to an uncertain measurement.Enlarge image
An editorial illustration created using AI showing traceable marking of suspect sensor data; not a documentary photograph.

A simple row average can mislead with unevenly timed samples

Our original artificial dataset has three points: two metres per second at second zero, four at second nine and four again at second ten. The simple average of the three table values is 3.33 metres per second. This is the arithmetic mean of the samples, weighting every row equally. It does not automatically give the time average over all ten seconds.

If, as a separate illustrative assumption, we hold the previous value until the next sample, we count nine seconds at two and one second at four. The time-weighted average is then 2.2 metres per second. Assuming uniform, linear change between points gives another result, 3.1. Both values arise from a chosen intermediate model, not newly measured reality.

The lesson has two parts. With irregular intervals, the simple row average often is not the quantity sought. For time weighting, however, you must also state what you assume between two measurement points. The missing intermediate history does not become known by interpolating it onto a regular grid. A sound analysis documents both weighting and interpolation assumptions.

A short section or turn can fall between two saved points

As an original example, imagine a short, six-second pace change. If the saved series contains points just before and after it, the two endpoints cannot reveal the exact intervening acceleration. If the programme draws a straight line between them, the brief peak may disappear or take a different shape. A tidy final curve does not mean complete temporal detail.

The spatial story of a sharp turn can be simplified in the same way. In an original geometry example, a runner covers two perpendicular twenty-metre sections. If only the start and end are saved, the straight line between them is about 28.3 metres, while the actual broken line totals forty. This is not a typical error claim for a particular watch, but a demonstration of what an endpoint-only description loses.

Denser records can retain more detail but do not eliminate every measurement error. If coordinates themselves are inaccurate, the file may contain more faulty points. Recording density is therefore one important property among many. “More rows” and “a more accurate account of reality” are not automatically equivalent in every circumstance.

A pause and sparse records are different phenomena

The Garmin Activity encoding guide also distinguishes start and stop events from the frequency at which records are written. A paused timer is not the same as a longer gap between stored points. Appropriate event information can help establish when the timer was running and when actual pausing occurred.

In an original example, twenty seconds pass between two records. This could be steady movement with sparse saving, or a stop with different events and different summary time. The time gap between the two points alone does not distinguish the stories. Treating every gap as a zero-speed pause can create false stopped time. Filling every gap with movement can conceal a genuine stop.

Keep the concepts of time explicitly named. Total elapsed time, time counted by the device’s timer and a service’s retrospectively estimated moving time can differ. Before analysing an export, check which appears in the summary and which the average uses. Many apparent “sampling errors” actually arise from comparing different definitions of time.

Smoothing and every-second resampling are not new measurements

A graph renderer may connect points, average within a window or adapt detail to the screen space. That is processing for display. The file’s recording frequency does not change as a result. Sparse data can produce a smooth curve, while dense measurements can be displayed so compressed that little detail is visible.

In an original screen example, an hour-long run’s every-second series would contain 3600 samples, while the graph is only six hundred pixels wide. Every point cannot have its own vertical pixel column. The renderer combines or selects in some way. A short peak disappearing from a small graph therefore does not establish that it was absent from the original series. Zooming and examining actual records can help distinguish the two.

Conversely, if a programme inserts estimated values every second between sparse points, row count increases but there are not that many original measurements. Treat the new values as interpolated. Otherwise, later analysis may count them as independent observed points. A regular table is convenient, but must retain the original data’s provenance.

A woman runner reviews a notebook beside a café window, looking for patterns in her running log.Enlarge image
An editorial illustration created using AI showing a running-log review; not a documentary photograph.

Check power-saving settings separately

The official Apple Low Power Mode guide distinguishes general low-power workout mode from the additional setting for fewer GPS and heart-rate readings. The documentation describes the latter on specified models and from a particular watchOS version for outdoor walking, running and hiking. Enabling it also restricts other workout functions. Not every power-saving setting reduces readings in the same way.

This specific example matters because the single word “power-saving” can cover several behaviours. When comparing two exports, note the full workout settings and model alongside the recording mode. Software version may matter too. Check your own device’s manual for which data is reduced and under what circumstances, rather than another brand’s or an older watch’s description.

On a long run, battery life and detail may be a real trade-off. The right choice depends on the task: analysing a short interval workout needs different detail from logging an all-day route. This article does not claim identical energy savings for every device. A checked sample file and documented setting together provide a usable comparison.

How can you check your export’s time axis?

First preserve the original, then use a copy to examine the actual times of several successive records. Calculate the differences rather than treating row numbers as seconds. Find a familiar section and a point where you changed pace or stopped. The aim is to identify which temporal detail the file preserved, not redraw the run more attractively afterwards.

If data goes into a spreadsheet, keep the timestamp as a genuine date-time field. Do not automatically replace an empty sensor value with zero: missing data and an actual zero mean different things. When calculating averages, name the time basis and assumptions about intermediate values. These brief notes make later analysis much more reliable than an unexplained rounded total.

Smart and every-second recording can therefore differ significantly in the temporal detail retained by an export. File rows, sensor operation and the drawn graph are separate layers. Following timestamps, preserving the meaning of pauses and labelling estimated intermediate data lets even a sparse series be used meaningfully. The first step towards accuracy is knowing which point is an actual saved record and which is a display inference.

Sources

Helpful? Pass it on.