Two reps from the same set hit identical peak force. On a count log they’re indistinguishable: rep 7 and rep 9, both 380 N. On the curve, they’re not the same lift. One ramped clean. The other stalled mid-range, then snapped to peak in the last 200 ms with the help of a hip kick.
Force came out the same. Form did not.
What form-shape variance actually is
Every rep is a curve — force on the y-axis, time on the x. The shape of that curve is a fingerprint of the rep. A clean rep at a given movement pattern has a recognizable shape: where force builds, where it plateaus, where it descends.
When the shape changes mid-set, what changed isn’t the load. It’s how the lifter is producing the load — recruitment patterns, joint angles, compensations.
tensr surfaces this through a small family of related numbers:
- Force variability — standard deviation of force across the rep.
- Force range — peak force minus minimum force during the rep.
- Min force during rep — does the curve dip in the middle (stall) or stay above a floor (smooth grind)?
- Form-shape variance — how much each rep’s curve deviates from the set’s average curve shape (cross-correlation).
- Curve-shape classification — eventually: labels like “ramp,” “plateau,” “dip-recovery,” “step.” Research-grade today; v1 ships variance only.
Why it matters
Two arguments for why this isn’t curve-fitting nostalgia.
Form breaks before force does. A lifter at the edge of failure usually finds a way to maintain peak force for one or two more reps — but the way they get there changes. The eccentric shortens. There’s a mid-range stall. The top snaps instead of locking. Peak force lies; curve shape doesn’t.
Comparable shape across weeks is technique consolidation. When the same movement, same load, same tempo produces the same curve shape week after week, the lifter has stable mechanics. When the shape drifts, the technique is changing — sometimes intentionally (working on positions), sometimes not. Either way: data the lifter can act on.
The rep where the fingerprint broke is the right place to set your “this is my real working range” boundary. Past that, you’re not training the same movement anymore.
What to track together
Treat the curve-shape family as one consistency view, not four separate readings.
| Metric | What it tells you |
|---|---|
| Force variability | Was the rep smooth or jagged? |
| Force range | Did the curve dip mid-rep, or stay high throughout? |
| Min force during rep | Below what threshold did the load briefly fall? |
| Form-shape variance | Did this rep look like its neighbors, or like an outlier? |
The reading worth attending to is form-shape variance: it answers “is this set actually one set, or is it three reps of one movement and three reps of a degraded one.” Variability and range are the components that explain why the variance is high.
What gear it needs
Wants 80 Hz. Curve-shape comparison needs sub-second resolution to be meaningful — at 8 Hz you have only ~80 samples for a typical 10-second set; that’s barely enough to draw the curve, let alone fingerprint variations.
The honesty matrix marks form-shape variance as ⚠️ on 8 Hz scales (computable but resolution may mislead). On Tindeq or a C100 retrofit, it’s ✓.
A specific note: curve-shape classification — the named labels (ramp, plateau, dip-recovery, step) — needs labeled training data to be honest. We don’t have a corpus yet. Initial app versions surface variance as a number; the named labels arrive when there’s enough recorded data to validate them.
What to do tomorrow
After your next working set, scrub through three reps: the first, the middle, the last. Don’t compare peaks. Compare shapes.
Find the one that doesn’t match. That’s your real working rep range — the part of the set where the lifter you started as is still the one doing the work.
Two reps with the same peak force can look completely different. The numbers said the same thing. The curves told the truth.