Running Fitness

How these are calculated

Load is HR-zone-weighted minutes: each minute in Z1 counts 1, Z2 counts 2, and so on up to Z5 = 5. Runs without real HR count 2.5 per minute.

Long-term / recent load are exponentially weighted averages of daily load over 42 and 7 days (often called fitness and fatigue, CTL and ATL). They measure how much you train, not how fit you are: steady training keeps them flat even while you improve. Fitness gains show up in pace at AeT and VDOT. History begins Aug 19, 2026, so both start at the average load of the first four weeks.

Form measures training fatigue only. Sleep, stress, heat, and illness aren't included, so read single runs with that in mind.

Pace at AeT uses grade-adjusted pace (Minetti et al. 2002) from seconds with HR 138–152 bpm, moving faster than walking, on grades under 5%, scaled to exactly 145 bpm. It skips the first 10 minutes (warm-up), everything after 90 minutes (cardiac drift), and everything after a hard bout, when HR stays high for the rest of the run. A hard bout is 4 minutes at 155 bpm or 2 at 165, on an exponential curve through those points (shorter the higher the HR), counted after the warm-up. Runs within one day per 3 km of a race (Daniels' easy days after racing: 3 days for a 10K) and trail runs (marked in notes.md; grade adjustment doesn't account for rough footing) show as open circles and stay out of the trend. A run needs at least 4 minutes of those seconds. The trend weights each run by how many seconds qualified over the past 21 days (at least 3 runs), and each point is drawn at the center of its window, so the line stops about 10 days short of today. Only runs within 1,000 ft of Flagstaff (7,000 ft) count, so the number is a Flagstaff-altitude pace.

Zones come from config.json: LT1 (AeT) 145 bpm and LT2 (LT) 168 bpm. The intensity chart puts wider breaks at those two thresholds.

VDOT is Jack Daniels' formula applied to your most recent race (races.csv, or a planned race) or time trial (a planned session named "time trial"). Training efforts never replace it: they are run below race effort by design, and a fastest window can include recoveries. The next-race estimate uses that current fitness with no projected gains. Elevation comes from the run's own file.

HR-based estimate (Fitness vs. goal) estimates fitness from pace and HR together, so any steady running counts. Each steady lap gives an effective VO2max: the oxygen cost of its grade-adjusted pace (Daniels' equation, the one inside VDOT) divided by the share of VO2 reserve it used, which Swain & Leutholtz (1997, Med Sci Sports Exerc 29:410) found equal to the share of HR reserve (resting 50, max 188 bpm, from config.json). A steady lap leaves out its first 2 minutes (HR still catching up with pace), the run's first 10 minutes, everything after 90 minutes (cardiac drift), grades over 5%, and laps after a hard bout that aren't hard themselves (HR stays raised on jogs and cool-downs). It needs 4 minutes left and grade-adjusted speed that varies less than 10% (no strides or surges). Trail runs are left out. A population formula doesn't fit one runner exactly, so estimates are scaled to match your last 3 races or time trials (Runalyze calibrates its effective VO2max the same way), then converted to sea level like a race. Lap each rep so a workout's reps and jogs land in separate laps.

Load ramp is the acute:chronic workload ratio: load over the last 7 days divided by the weekly average over the last 28. Gabbett (2016, Br J Sports Med 50:273) calls 0.8–1.3 the sweet spot and ≥ 1.5 the danger zone for injury. The ratio has been criticized as a predictor (Impellizzeri et al. 2020, Int J Sports Physiol Perform 15:907), so read it as a guide to how fast load is rising, not a risk score. It needs four weeks of history.

Projection uses plan.csv after today, with each planned mile worth your recent load per mile for that kind of run (planned workout days, and before the plan, runs with ≥ 10% of the time in Z5, count as workouts). Planned hikes and XC skis (named in the session, with minutes in plan.csv's cross_min) count at your load per minute from recorded sessions of that kind or, until you have any, a default from config.json: hike 1, xc ski 2.5 per minute. The hike default is the Z1 weight: the Pandolf load-carriage equation and Swain's %VO2max-to-%HRmax relation put a 35-lb-pack hike on 0 to 10% grades at about 105 to 130 bpm (data/ref/README.md). Sea-level days and steady finishes aren't modeled.

Cross-training (hikes and XC skis recorded on the watch) adds HR-zone load the same way runs do, but no miles. HR load misses the muscle damage from long loaded descents, so expect to feel more tired after the Grand Canyon than form shows.

Decoupling (Pa:HR) compares grade-adjusted pace per heartbeat in the first and second halves of a run, after the 10-minute warm-up, counting running seconds on grades under 5%. Below 5% means your aerobic endurance holds for that long (Friel, TrainingPeaks). Only steady runs count (planned easy or long days without a steady finish, or before the plan, runs under 10% in Z5), with at least 45 minutes of qualifying running.

Workout reps come from laps, so press lap at the start and end of each rep. Grade-adjusted lap pace uses the same Minetti model as pace at AeT.

Altitude is converted two published ways and shown as a range. Daniels: paces slow 4–5 s/mi per 1,000 ft above 3,000 ft (15–18 s/mi at 7,000 ft). NCAA 2009 altitude tables: % added by venue altitude (0–7,703 ft) for the mile, 5K and 10K, interpolated by altitude and distance, and held at the nearest of those distances outside that span. Neither accounts for acclimatization.