Field notes · Methods
Borrow the durable math. Grade it honestly.
Running watches now record heart rate, pace, elevation, ground-contact time, stride length, and overnight physiology. Cycling had a head start: power meters gave coaches decades to develop ways of describing training load without mistaking a chart for a verdict. Plover brings across the durable parts of that work and adapts them to running.
The useful question is what changed, not what will happen. Plover is readiness and load intelligence: it observes the training record, ranks findings, and states how much evidence supports each one.
The first port is Banister TRIMP. Each run becomes a heart-rate-reserve training impulse, summed into a session load; daily load is then smoothed into acute 7-day and chronic 42-day exponentially weighted series. The plain-English comparison is recent training stress beside the base built over roughly the prior six weeks. It is a way to see a ramp, not a diagnosis.
We also adapt Friel-lineage aerobic decoupling, usually called Pa:HR. In cycling it compares output per heartbeat across a steady effort. In Plover, it compares grade-adjusted pace per heartbeat between the moving-time halves of a steady run after the first 5 minutes. If the back half covers less ground for the same cardiovascular cost, the finding says that the runner faded; heat, fatigue, terrain, and fueling can all contribute. A warning begins above 5% decoupling and a serious signal above 10%, but only when the run supplies the qualifying steady data.
Illustrative example — not your data. The band above 5% is a warning; above 10% is a serious signal.
Efficiency factor is the companion measure. It divides grade-adjusted speed by heart rate on qualifying steady runs after the first 5 minutes. The runner-facing question is simple: how much ground did you cover per heartbeat? A trend can be useful when conditions are comparable; a six-observation trend needs both a meaningful 4% change and sufficient statistical support before Plover surfaces it.
Illustrative example — not your data. The shaded band is your usual range; a trend needs a 4% change and statistical support before it surfaces.
The mean-maximal pace curve is another careful translation. Cycling uses a mean-maximal power curve to show the best sustainable output at different durations. Plover builds the same rolling best-effort envelope from running speed and expresses it as pace. It is a chart of capability across supported durations, not a claim about a runner's future race.
Some methods do not survive the crossing intact. Plover retains the vendor-provided acute:chronic load ratio as a descriptive cross-check, looking at the most recent 6 days and flagging a ratio of at least 1.5. It does not treat the ratio's popular "sweet spot" as a personal law. The acute:chronic workload ratio has important methodological critiques, so a load ratio can add context but cannot earn a verdict by itself.
Running keeps its own evidence
A running signal should respect what a runner's body actually does. Plover therefore treats mechanics, repetitions, distance, and pace as first-class evidence rather than forcing them into a cycling model.
Matched-pace ground-contact-time drift asks whether the feet stayed on the ground longer late in a run at the same pace. It compares the first and last thirds in a shared 0.25 m/s pace bin, on a steady run of at least 30 minutes with grade held within ±2%; a finding begins at 3% drift. Time on the ground each step can creep up as fatigue arrives, but it is an observation of mechanics, not proof of cause.
Stride fade across reps asks whether stride length shortened through a selected interval set. It compares front-half and back-half mean stride length, with a finding when the back half drops by at least 2% or the slope is negative with the required statistical support. At least 4 reps are needed before the comparison is made.
Weekly mileage ramp is deliberately running-specific. It measures the week-over-week change in weekly mileage and follows the Nielsen running-cohort line: it surfaces above a 30% increase, becomes critical above a 100% increase, and ignores a prior week below 3.0 miles. The point is not that every increase is harmful. It is that a large weekly jump is a well-supported reason to look closer.
Single-session distance spike is more immediate. It compares a run with the longest run in the preceding 30 days, then marks an elevated change above 10%, a moderate change above 30%, and a critical change above 100%. This is the strongest population-level load observation in the system, drawn from Garmin-RUNSAFE work; it still cannot say what will happen to one runner after one long run.
Effective VO2max keeps the running model where it belongs. Plover uses a Runalyze-style, heart-rate-regressed grade-adjusted velocity and the Daniels/Gilbert cost curve to estimate aerobic engine size from pace and heart rate. Heat can drag a single estimate down, so the six-observation trend matters more than any one day.
Recovery physiology, capped
Recovery data can help, but it should not speak alone. Plover uses nightly median rMSSD transformed to ln(rMSSD), comparing a 7-day mean with a 21-day baseline and a 0.5-SD smallest worthwhile change. A lower-than-usual value can describe under-recovery; it can also describe normal adaptation, travel, measurement noise, or a changed routine.
A recovery signal never turns red without performance confirmation. This cap is enforced in the signal logic: physiology can strengthen a performance finding, but it cannot independently produce the strongest warning.
Grade the finding, show the gap
Evidence is part of the result. Every Plover finding belongs to one of three signal families, and each family carries a different claim.
Load signals have strong prospective evidence. They include the single-session distance spike, weekly mileage ramp, training load, and Foster monotony and strain. Strong does not mean certain for an individual runner; it means the underlying relationship has the clearest prospective cohort support in this set.
Performance signals have good evidence. They include aerobic decoupling, efficiency factor, effective VO2max, matched-pace ground-contact-time drift, and stride fade across reps. These measures can show a change in performance or mechanics. Their link to an individual running problem is softer, so their language stays descriptive.
Recovery signals have supporting evidence. HRV is useful for understanding training response, but it is bidirectional and least reliable alone. The cap remains in place even when a recovery measure is striking.
The finding card leads with the observation in plain language. "Your steps are getting shorter across this interval set" comes before the supporting number; "Stride length down 3% across 6 reps" gives the reader the evidence. That order matters. A metric name never has to carry the meaning by itself.
Plover also says when it cannot compare. A card might say "3 of 7 comparable runs" rather than inventing a baseline, or mark a mechanics analysis unavailable when the watch did not record the required dynamics. Missing data stays missing. A 7-day pattern cannot be claimed before the comparable history exists.
Plover makes no injury prediction. It does not diagnose an injury, prescribe a training plan, or turn a cohort-level hazard into a promise about Saturday. It surfaces observations that deserve attention, shows the supporting numbers and windows, and leaves room for the runner, coach, or clinician to make the decision.
References
- Banister (1991).
- Foster (1998), monotony and strain.
- Friel, The Triathlete's Training Bible, Pa:HR and efficiency factor.
- Nielsen et al., running weekly-ramp cohort.
- Plews and Laursen, HRV smallest worthwhile change.
- Daniels and Gilbert, running oxygen-cost curve.
- Morin, running stiffness and ground-contact-time work.
- Impellizzeri et al. (2020), acute:chronic workload ratio critique.