Most of cycling analytics treats the rider as a scalar. One FTP number, one set of zones, one training prescription built on the pretense that a single threshold can carry the weight. The Power Threshold Array is the claim that this is no longer sufficient, and that the resolution to replace it is now achievable.
The problem
A rider's power-duration curve is the surface projection of several distinct energy systems operating on different timescales, with different substrates, different fatigue profiles, different trainabilities. The transitions between them are not evenly spaced. They are not proportional to FTP. They move independently in response to training.
Two riders with identical FTPs can have radically different shapes everywhere else on the curve. Those shape differences are what determine how they actually race. A sprinter and a time trialist at the same FTP are not the same rider. FTP cannot express the difference. PTA can.
The framework
PTA names seven inflection points on the power-duration curve that correspond to specific bioenergetic transitions. Each letter is a mnemonic for the energy system the threshold separates. These are not zones. They are named boundaries between physiological regimes, found on the rider's actual curve rather than calculated from a single number.
| Threshold | Energy System | |
|---|---|---|
| PT-N | Neuromuscular | Peak phosphocreatine sprint capacity |
| PT-S | Sprint | Phosphocreatine depletion onset |
| PT-G | Glycolytic | Anaerobic glycolytic activation |
| PT-X | Maximal | Glycolytic ceiling, maximal anaerobic output |
| PT-V | VO2 | Aerobic transition, VO2max onset |
| PT-A | Aerobic | Sustained aerobic capacity |
| PT-T | Threshold | Functional threshold, lactate equilibrium |
Each threshold carries two independent measurements. The first is power: the watts at which the rider crosses it. The second, and the one that most platforms ignore entirely, is offset: the duration at which the crossing occurs.
Seven thresholds times two dimensions gives fourteen independent development signals where FTP offers one. That is the resolution at which a rider's shape can actually be described.
Methodology
The framework's central methodological move is that thresholds are found, not computed. PT-V is not derived from FTP by multiplying by some ratio. PT-S does not sit at a fixed fraction of PT-G. The algorithm does not assume where a threshold lives. It looks at the rider's actual power-duration envelope and locates the inflection.
This matters because the assumption that thresholds sit at stable ratios to each other is false, and it is precisely the falsity that makes one rider a sprinter and another a time trialist at the same FTP. Calculated thresholds describe an average rider who does not exist. Detected thresholds describe the rider in front of you.
An envelope gets built from the best efforts across a training window. Analysis of that curve reveals where the curvature changes, because a threshold between two energy systems is exactly a place where the rate of power decline with duration shifts. The algorithm finds those shifts inside physiologically plausible search windows and assigns each one a severity, a confidence, and a contextual interpretation.
Precision
A single envelope and a detector is the toy version. The serious version is that detection gets sharper when it knows the rider.
An athlete's PT-V has lived in a specific duration band for three years. When the algorithm runs today, it does not need to search the full population window. It can search the rider's own historical range, with a tight prior. Detections tighten. Confidence rises. False positives from noisy envelopes decline. Change becomes meaningful precisely because the baseline is personal.
The first pass builds the rider's historical profile using population priors, because that is all that is available for a new athlete. The second pass, once the profile exists, uses personal priors derived from the first. The profile is a regression target, not a ground truth. It says here is where this rider's thresholds have historically lived, and the current detection gets to use that as a lens, not a cage.
Output
What the algorithm produces for a given training window is called a PTA fingerprint. Seven thresholds, each with power and offset. Each flagged as detected, balanced, or insufficient_data depending on whether the rider's data expressed that threshold well enough in the window.
Alongside each threshold sits an elevation reading: is the rider's power at this threshold rising, stable, or declining relative to a baseline? The baseline is selectable, because "rising relative to last quarter" and "rising relative to last year" are different claims about the same curve.
Application
The macro fingerprint tells a rider who they are. The session tells them what they just did. You need both for the framework to matter.
Most sessions do not express most thresholds. A steady endurance ride expresses PT-A and maybe PT-V and says nothing about PT-N or PT-S. A criterium expresses PT-N, PT-S, PT-G vividly and leaves PT-T silent. Honest session analysis reports only what the session actually expressed, and reports it as a delta against the current fingerprint.
This splits sessions into two modes. A probing session expressed at least one threshold with enough signal to read as a detection or delta. A loading session expressed no thresholds but still did physiological work. The ninety-minute endurance ride didn't probe any boundary. It spent its time below PT-T, in the sub-threshold aerobic domain that PTA does not yet partition with named thresholds.
The rider learns that "nothing detected" does not mean "nothing happened." It means the session was developing capacity below PTA's current detection boundary, not probing a threshold above it. Sub-threshold detection is acknowledged as open work. Both modes are PTA-native. Over a training block, session reports accumulate, and the rider begins to read their own training through the framework's lens.
Coexistence
PTA does not replace the Performance Management Chart. It completes it.
The PMC, built on Banister's impulse-response model and translated into TSS, CTL, ATL, and TSB, measures load and readiness. It describes the tank and its state. Nothing in PTA is in competition with it.
Distinction
Zones are prescriptive. They slice the continuous curve into bands for the purpose of telling a rider what to do. PTA is diagnostic. It describes the curve's actual structure so the rider knows what they are, how they are changing, and where their development is accruing.
Zones can be built on top of PTA. A derived layer, PTA Zones, numbered one through seven, will give coaches and training platforms a prescriptive surface to work with. But the layers are distinct on purpose. Detection comes first. Prescription comes after, and comes from something real.
Ambition
The long arc is that a rider's PTA fingerprint, tracked over time, resolves questions that FTP cannot.
Where is my training actually landing? Which energy system responded to the block I just finished? Is my anaerobic capacity declining as my aerobic base grows, or holding? Am I the same rider I was last March, or has my shape changed? What does my shape say about the races I should target?
FTP answers none of these. PTA is designed to answer all of them, in the rider's own terms, with the rider's own history as the reference frame. The rider who uses PTA seriously does not need to be told what zone to ride in next week. They can read their own shape and know.
That is the framework's ambition. Everything RaceProof builds on top of PTA, from the macro fingerprint to the session delta to the interpretation matrix to Loom, is in service of delivering that ambition to riders and to the sport.