Coaching Methodology & Sources

Steev is a coaching tool, not a medical service, and nothing on this page is medical advice. These rules are how the coach is instructed to work — AI guidance can still be incomplete or wrong. If something hurts, see a clinician. Read the full terms →

The models we run

These ten run in production; the runner-specific blocks appear once there are results to feed them. The equations below are the ones the server evaluates, not a simplified retelling of them.

VDOT — the Daniels–Gilbert model

Fitness estimate · from your best run of the last 12 months

Steev takes your strongest run of the last twelve months — the steady effort of 3 km or more with the highest VDOT it implies — and works backwards to one number: the oxygen cost of the pace you held, divided by the fraction of your maximum you could have held it at. That quotient is VDOT. It is the fitness trend on your stats screen, and the coach is handed it — along with race equivalents computed from that same run — before it prescribes a pace. The window is the point: fitness fades, and a number anchored to your fastest year would keep prescribing paces you cannot hold today. The VO₂max trend chart is a looser thing: it runs the same equations over each individual training run, and a training effort is not a maximal one, so its points usually sit below this headline and are worth reading as a shape rather than as a number.

VO₂ = -4.60 + 0.182258 × v + 0.000104 × v² v = velocity in metres per minute %VO₂max = 0.8 + 0.1894393 e^(-0.012778 t) + 0.2989558 e^(-0.1932605 t) t = duration of the effort in minutes VDOT = VO₂ / %VO₂max

Source Daniels & Gilbert (1979), Oxygen Power; Daniels' Running Formula, 4th ed., Human Kinetics

Race equivalents — the Riegel formula

Estimates · not targets

From one result Steev projects the others: 5K, 10K, half marathon, marathon. Riegel fitted the 1.06 exponent to world-record performances across distances, so the formula describes how endurance decays with distance — it does not predict what you will run. It assumes you have actually trained for the longer distance, which is exactly what a runner reading it usually has not done yet. Read these as a snapshot of current fitness. Never as a goal time.

T₂ = T₁ × (D₂ / D₁)^1.06 T = time, D = distance; 1 = the result you ran, 2 = the race being estimated

Source Riegel (1981), American Scientist 69(3):285-290

Rolling training load

Your own history · no external model

Eight rolling seven-day buckets of distance, the longest single run inside that window, and the change between the two halves. There is no citation here on purpose: this is arithmetic on the runs you synced, not a score against a published index. It is deliberately not an acute:chronic workload ratio — that index has been substantially challenged, and a contested number dressed up as a verdict is worse than a plain one. The coach is given this block before it plans anything.

ramp % = (last 4 weeks - previous 4 weeks) / previous 4 weeks × 100 distance summed per 7-day bucket, most recent 8 weeks

Volume ramp

Your own history · no external model

The ramp on your stats screen is one week against its own recent normal: the current seven days of distance divided by the average of the four seven-day buckets before them. It is not the number in the block above — the coach is handed four weeks against the previous four, which is the same idea measured slowly — so the two disagree by design and each says what it says. The buckets are calendar days back from today, so the current one fills up as the week goes on and the ramp reads low on a Monday and climbs; that is the week arriving, not the load dropping. Under 20 km of prior four-week volume nothing is shown at all, because a percentage off a baseline that thin is arithmetic noise, and your history has to reach back into the fifth bucket before there is anything to divide — an absent week is not a rest week, so the weeks before you arrived are left out of the average rather than counted as zeroes.

ramp % = (this week / mean(previous 4 weeks) − 1) × 100 seven-day buckets counted back from today; nothing reported under 20 km of prior volume

Training paces — the Daniels registers

Pace ranges · from your VDOT

With a VDOT in hand the oxygen-cost equation runs backwards: take a fraction of your VDOT as the target oxygen cost, solve for the velocity that costs it, and that velocity is a pace. Easy, Marathon, Threshold and Interval each carry a published range expressed as a share of VO₂max, and the coach is handed all four before it writes a session. The paces are computed from those equations — nothing here is read off the book's hand-smoothed tables, and no range on this page is a quotation from one. We have not verified the register bounds against the book itself; the sources we could reach that reproduce them are secondary and agree with each other, which is corroboration rather than verification. Repetition pace is left out on purpose: it is not defined as a share of VO₂max, so there is no equation to run. Below VDOT 30 the coach is handed no paces at all — that is under the range the equations were fitted to — and the paces that do come out assume flat ground, with nothing in them for hills, heat or altitude.

target VO₂ = f × VDOT f = the register's share of VDOT f(E) = 0.59–0.74 f(M) = 0.75–0.84 f(T) = 0.83–0.88 f(I) = 0.95–1.00 the published %VO₂max bounds for Easy, Marathon, Threshold and Interval 0.000104·v² + 0.182258·v − (4.60 + VO₂) = 0 positive root → v in metres per minute pace = 60000 / v seconds per kilometre

Source Daniels' Running Formula, 4th ed., Human Kinetics

Heart rate at a matched pace

Your own comparison · no external model

The same pace at a lower heart rate is one of the few signs of aerobic change that survives a bad week. Steev takes the median pace of your runs from the last four weeks, keeps every run within 30 seconds per kilometre of it, and compares the average heart rate of that band against the same band from the four weeks before. Runs only, and only runs with a real pace to read — anything under a kilometre or logged without a duration is thrown out, because one junk row moves a median. If either side of the comparison comes up empty, nothing is shown at all rather than a verdict built on a single run. This is not aerobic decoupling or any published index: it is our own comparison of two averages, and heat, sleep and a strap that slipped move it as readily as fitness does.

reference = median pace of the last 4 weeks' runs runs kept within ±30 s/km of it, both windows Δ = avg HR (last 4 weeks) − avg HR (previous 4 weeks) negative Δ = lower heart rate at the same pace

Negative-split rate

Your own history · no external model

The share of your runs whose second half was quicker than the first. The halves are measured by DISTANCE rather than by lap count, so the lap straddling the midpoint contributes its time in proportion to how much of it falls on each side — assigning whole laps by index would read an evenly-run 3 km as a fade. A run has to carry at least two recorded laps covering at least 3 km, and that floor is applied to the laps themselves rather than to the run's total distance: a synced run can arrive with only a fraction of its distance recorded as laps, and a quarter of a fragment is not the end of a run. Structured sessions are left out entirely, on the same grounds as the VDOT entry — they vary pace on purpose. This is a count of your own runs and nothing more: a low rate over a month of easy long runs is a description, not a fault.

negative split = (total time − first-half time) < first-half time halves by distance, the straddling lap prorated; ≥2 laps covering ≥3 km rate = negative-split runs / runs with readable laps × 100

Pace fade

Your own history · no external model

How much slower the last quarter of a run was than that run as a whole — positive is fading, negative is a closing kick. The quarter is measured by distance back from the finish, with the lap straddling the boundary prorated exactly as the halves above are, and it needs four recorded laps over the same 3 km floor. The card shows the MEDIAN of your runs, never their average. Individual laps are never screened for plausibility — the check that rejects an impossible pace reads whole runs, not the laps inside them — so a lap recorded as 60 m in 240 s pushes its own run more than twenty percent, and one such lap in a month would drag an average across the whole card. The median leaves it where it belongs, as one odd run.

fade % = (last-quarter pace − whole-run pace) / whole-run pace × 100 quarter by distance from the end, the straddling lap prorated; ≥4 laps covering ≥3 km card = median(fade % across the runs in the window)

Personal-best detection

Bookkeeping · your records table

Every completed run is checked against six standard distances — 1K, Mile, 5K, 10K, Half Marathon, Marathon. A run within 5% of one of them counts as an attempt at it: the time is scaled in proportion to the standard distance and kept only if it beats what you already had. That scaling is straight proportion, not the Riegel decay above, so an effort slightly short of the distance is flattered and one slightly long is treated harshly — inside a 5% window the gap is small, but it is there. These records are bookkeeping in the plain sense: the Records card and the coach's list of your bests. Your fitness estimate no longer reads them — VDOT comes from your strongest run of the last twelve months, record or not — so an old best stays on the table without setting today's paces. A GPS run recorded long, or a treadmill effort filed as a road result, still moves your training while it is recent: the estimate reads the run itself, not the record it set. A best set within the last 14 days is shown as news on the Records card; past that it is simply your record, and the table reads the same either way.

normalized time = time × (standard distance / run distance) run distance within ±5% of the standard

Leaderboard progress metrics

Your own history · ranking only

Three of the leaderboard boards rank change rather than totals: pace, VO₂max and weekly distance, each as the last four weeks against the four before them. Pace and distance are averages of what you logged, and pace inverts its sign so that getting faster reads as a gain. The VO₂max board carries the caveat from the VDOT entry above, and carries it over sessions rather than runs: it puts race-effort equations over every session you completed, so an easy month reads as a decline. Pace and VO₂max sit out entirely until both windows have sessions in them. The volume board only needs the earlier month to be non-empty, so a recent month you logged nothing in reads as a drop to zero rather than as silence.

change % = (recent 4-week average − previous 4-week average) / previous × 100 pace and VO₂max average per session; volume averages weekly totals; the pace board evaluates (previous − recent) / previous instead, so faster reads as a gain

Training methodology

In novice runners, distance-related injuries were more frequent where weekly distance rose by more than 30% over two weeks, short of significance (hazard ratio 1.59, 95% CI 0.96-2.66); a graded program built on the 10% rule did not reduce injury at all.

Source Nielsen et al. (2014), J Orthop Sports Phys Ther 44(10):739-747; Buist et al. (2008), Am J Sports Med 36(1):33-39
Elite endurance athletes measurably train polarized: about three of every four sessions clearly easy, below the first ventilatory threshold, roughly one in five hard with substantial time above the second, and little in the comfortably-hard middle.

Source Seiler & Kjerland (2006), Scand J Med Sci Sports 16(1):49-56
Neuromuscular function, force production and gait return within a day of a hard session for most runners but not all, so quality days sit apart rather than consecutive, spaced on the slower case.

Source Riazati et al. (2022), Front Sports Act Living 4:830278; Daniels' Running Formula, 4th ed., Human Kinetics
Intensity and volume are not interchangeable: over eight weeks matched for total work, 4x4-minute intervals at 90-95% of maximum heart rate raised VO2max by 7.2% and 15/15s by 5.5%, while the same work at 70% of maximum or at lactate threshold did not.

Source Helgerud et al. (2007), Med Sci Sports Exerc 39(4):665-671
In adults, high-intensity intervals beat moderate continuous training for VO2max (SMD 0.64 in healthy adults), though only work intervals of 2 minutes or longer, at least 15 minutes of work per session, and blocks of 4 weeks or more reached significance.

Source Wen et al. (2019), J Sci Med Sport 22(8):941-947
In highly trained endurance athletes, added submaximal volume does not appear to further improve endurance performance or peak oxygen uptake; the reading of the evidence is that further gains come through high-intensity interval work.

Source Laursen & Jenkins (2002), Sports Med 32(1):53-73
Training raised lactate and ventilatory thresholds far more in sedentary participants than in already-conditioned ones: pooled across 85 groups from 34 studies, effect sizes were 2.32 against a non-significant 0.63, with high intensity trending best among the conditioned.

Source Londeree (1997), Med Sci Sports Exerc 29(6):837-843
Speed and power need no extra hours: in elite runners, replacing 32% of training with explosive strength work — sprints, jumps, light fast lifts — improved 5K time by 3.1% and running economy by 8.1% in nine weeks, with VO2max unchanged.

Source Paavolainen et al. (1999), J Appl Physiol 86(5):1527-1533
Uphill running is movement-specific resistance training: two sessions a week for six weeks improved 5K time by 2.0% in well-trained runners, equally across gradients from 4% to 18%, with the steepest best for running economy.

Source Barnes et al. (2013), Int J Sports Physiol Perform 8(6):639-647
Illness tracks training strain — weekly load multiplied by monotony, the flatness of daily load — so a flat week costs more than an uneven week of the same total, at a threshold that is individual rather than common.

Source Foster (1998), Med Sci Sports Exerc 30(7):1164-1168; Haddad et al. (2017), Front Neurosci 11:612
A taper cuts volume, not intensity: from about two weeks out — shorter for a 5K or 10K, longer for a marathon — a 40-60% drop in volume with session intensity untouched. Two weeks is the default that holds up.

Source Bosquet et al. (2007), Med Sci Sports Exerc 39(8):1358-1365
After a race, recovery precedes rebuilding: about one easy day per 3K raced, so roughly a week after a half and two after a marathon, with no quality until then. Muscle damage outlasts how fresh the runner feels — myofibrillar repair after a marathon runs to 3-4 weeks, and leg power is still down at five days.

Source Daniels' Running Formula, 4th ed., Human Kinetics; Warhol et al. (1985), Am J Pathol 118(2):331-339; Petersen et al. (2007), Eur J Appl Physiol 101(3):385-396
Milestones are tests: run fresh, aimed at a concrete target, and read against that target when planning onward. The first test is the baseline the block is measured from, so it comes as soon as the runner is fresh enough to run it.

Source Daniels' Running Formula, 4th ed., Human Kinetics
A beginner's training is prescribed by feel and time on feet, with walk-run a first-class structure; paces enter only once race or test data exists.

Source Daniels' Running Formula, 4th ed., Human Kinetics
Intensity has five registers: Easy (conversational, the bulk of the week), Marathon (steady, sustained), Threshold (comfortably hard, about 20-40 minutes of work), Interval (hard, 3-5 minute repeats), Repetition (faster still, short and fully recovered).

Source Daniels & Gilbert (1979), Oxygen Power; Daniels' Running Formula, 4th ed., Human Kinetics
Pace tables and the runner's body can disagree: where heart rate on recent runs or reported effort says a prescribed easy band is off, the body outranks the table.

Source Daniels' Running Formula, 4th ed., Human Kinetics
Heart rate reads as a zone only against a known maximum, and an age formula is not a measurement.

Source Daniels' Running Formula, 4th ed., Human Kinetics
The long run is a share of the week, not a heroic outlier: a clear minority of weekly volume, capped by time on feet rather than distance for slower runners, for whom the same mileage means far longer under load.

Source Daniels' Running Formula, 4th ed., Human Kinetics
Fueling: nothing under about 45 minutes; 30-60 g carbs/hour for 1-2.5 h efforts, up to ~90 g/h beyond. Race fueling is rehearsed on long runs — nothing new on race day.

Source Thomas, Erdman & Burke (2016), ACSM/AND/DC Joint Position Stand, Med Sci Sports Exerc 48(3):543-568; Jeukendrup (2014), Sports Med 44(Suppl 1):S25-S33
After time off, fitness is rebuilt rather than resumed: aerobic fitness slips within the first weeks away, newest gains first, so newer runners lose more. Past about four weeks the progression restarts; resuming at the old volume is a ramp spike in disguise.

Source Mujika & Padilla (2000), Sports Med 30(3):145-154 (Part II) and 30(2):79-87 (Part I)
Where injury or life blocks running, pain-free cross-training substitutes for it: deep-water running or cycling holds running fitness for several weeks, and a substitute that hurts is dropped too.

Source Wilber et al. (1996), Med Sci Sports Exerc 28(8):1056-1062; Eyestone et al. (1993), Am J Sports Med 21(1):41-44
Strength work belongs in a running plan: 2-3 sessions a week, sustained six weeks or more, likely improve running economy without adding aerobic load.

Source Blagrove, Howatson & Hayes (2018), Sports Med 48(5):1117-1149
Treadmill and overground running are biomechanically comparable across 33 crossover studies of 494 participants, apart from foot angle at footstrike. A flat belt costs less energy only from about 3.75 m/s upward, where a 1% grade restored road cost in nine trained men.

Source Van Hooren et al. (2020), Sports Med 50(4):785-813; Jones & Doust (1996), J Sports Sci 14(4):321-327

Warm-up and stretching

A warm-up builds in order: submaximal aerobic work, then large-amplitude dynamic movement, then efforts at the session's pace. Across 32 studies, warming up improved 79% of the performance measures examined, with little evidence of harm.

Source Behm & Chaouachi (2011), Eur J Appl Physiol 111(11):2633-2651; Fradkin, Zazryn & Smoliga (2010), J Strength Cond Res 24(1):140-148
Static stretching is not a warm-up: tested minutes later, holds of 60 seconds or more per muscle group cost about 4.6% of the performance that follows and shorter holds about 1.1%. In the few studies adding dynamic work afterwards, no clear deficit remained.

Source Behm, Blazevich, Kay & McHugh (2016), Appl Physiol Nutr Metab 41(1):1-11; Simic, Sarabon & Markovic (2013), Scand J Med Sci Sports 23(2):131-148
Stretching has not been shown to prevent overuse injury: pooled across randomised trials it was the one preventive measure with no protective effect (RR 0.96, 95% CI 0.85-1.10), on three trials in recruits and general exercisers. Its possible benefit is confined to acute muscle injury in sprint-type sport.

Source Lauersen, Bertelsen & Andersen (2014), Br J Sports Med 48(11):871-877; Behm, Blazevich, Kay & McHugh (2016), Appl Physiol Nutr Metab 41(1):1-11
Stretching does not relieve soreness: pooled across randomised trials it moved next-day soreness about half a point before exercise and about one point after, on a 100-point scale, with intervals spanning no effect.

Source Herbert, de Noronha & Kamper (2011), Cochrane Database Syst Rev (7):CD004577

Sleep and recovery

Sleep need is individual: a flat 7-9 hours is unlikely to suit everyone, and the 2021 athlete consensus works from the athlete's own perceived need. Habitual nights under 7 hours are common in elite athletes, and in the wider population that pattern raises susceptibility to respiratory infection.

Source Walsh et al. (2021), Br J Sports Med 55(7):356-368
A performance drop during intensified training normally restores in days to weeks; taking weeks to months means the overload outran the recovery. No resting hormone, blood or questionnaire marker separates the two, so the distinction is made retrospectively from the recovery itself.

Source Meeusen et al. (2013), Med Sci Sports Exerc 45(1):186-205
Early training costs sleep that an earlier bedtime does not recover: elite swimmers on 06:00 starts went to bed about 2.5 hours earlier but rose about 4 hours earlier, sleeping 5.4 hours against 7.1 before rest days.

Source Sargent, Halson & Roach (2014), Eur J Sport Sci 14(Suppl 1):S310-S315

Nutrition

Iron deficiency runs roughly 15-35% in female athlete cohorts against 5-11% in male, and a normal haemoglobin does not rule it out — iron-deficient non-anaemic is its own stage, where treatment raised VO2max (Hedges' g = 0.61). It shows in bloodwork, not in a training log.

Source Sim et al. (2019), Eur J Appl Physiol 119(7):1463-1478; Burden et al. (2015), Br J Sports Med 49(21):1389-1397
Caffeine at 3-6 mg/kg, most often 60 minutes before, benefits aerobic endurance most consistently of any exercise type. Its size varies between people through caffeine metabolism, as do effects on sleep and anxiety; 9 mg/kg adds side-effects, not performance.

Source Guest et al. (2021), J Int Soc Sports Nutr 18(1):1
1.4-2.0 g/kg of protein a day covers most people who train, spread every 3-4 hours rather than banked in one meal. For endurance running it offsets muscle damage; carbohydrate is what supports the performance.

Source Jäger et al. (2017), J Int Soc Sports Nutr 14(1):20

Motivation and adherence

A varied program is adhered to better than a repetitive one: 64% of sessions against 51% over six weeks in inactive adults, and the effect ran through the variety they perceived rather than the variety on paper.

Source Sylvester et al. (2016), J Behav Med 39(2):214-224; Eather et al. (2023), J Sport Exerc Psychol 45(3):148-165
Pleasure declines above the ventilatory or lactate threshold, holds below it, and varies widely close to it. Across 24 studies, affect during a session predicted later activity while affect afterwards did not; self-selected intensity is tolerated better than imposed.

Source Ekkekakis, Parfitt & Petruzzello (2011), Sports Med 41(8):641-671; Rhodes & Kates (2015), Ann Behav Med 49(5):715-731
Among 111 new gym members, at least four sessions a week for six weeks was the minimum associated with an established exercise habit, and training at the same time of day predicted habit growth more strongly than setting, session complexity or enjoyment.

Source Kaushal & Rhodes (2015), J Behav Med 38(4):652-663
Automaticity builds over months, not weeks: among 39 modelled participants with a self-chosen daily behaviour, the median time to 95% of peak automaticity was 66 days, range 18 to 254, and a missed day cost 0.29 points on a 0-42 scale with no lasting effect.

Source Lally et al. (2010), Eur J Soc Psychol 40(6):998-1009
Of 774 novice runners starting a six-week beginners' program, 29.5% had stopped running 26 weeks later, injury the main reason for 48% of them; being unsure about continuing was associated with double the odds of stopping (OR 2.06).

Source Fokkema et al. (2019), J Sci Med Sport 22(1):106-111
Motivation and action are separable: in a randomised trial of 248 participants, raising threat appraisal, coping appraisal and intention produced no significant increase in exercise over two weeks, while adding an if-then plan naming when and where sharply raised it without moving intention.

Source Milne, Orbell & Sheeran (2002), Br J Health Psychol 7(2):163-184
Pooled across 139 studies and 3,599 participants, music raised affective valence (g = 0.48) and physical performance (g = 0.31) and lowered perceived exertion (g = 0.22), at high intensity as much as at low. Tracks at 120 bpm or more carried the larger performance effect.

Source Terry et al. (2020), Psychol Bull 146(2):91-117
Matched for intensity, outdoor exercise is more enjoyable than indoor in mostly healthy adults (g = 1.24, 10 comparisons, 234 participants), while physiological differences were non-significant or inconclusive. Participants declared greater intent to repeat it; no trial measured whether adherence followed.

Source Peddie et al. (2024), Health Psychol Rev 18(4):853-883; Thompson Coon et al. (2011), Environ Sci Technol 45(5):1761-1772
Across 71 studies of adult physical-activity programmes, group delivery held no significant advantage over individual for activity itself (g = 0.086), nor for psychosocial or health outcomes; only functional outcomes favoured groups. Online and in-person groups performed alike.

Source Kritz et al. (2026), Nat Hum Behav 10(6):1109-1121

Safety

Pain is a gate: sharp, localized, worsening, or stride-changing pain means stopping the run, and the days already on the calendar no longer fit the runner they were written for.

Source Brukner & Khan's Clinical Sports Medicine: Managing Injuries, 6th ed., McGraw Hill; Yamato et al. (2015), J Orthop Sports Phys Ther 45(5):375-380
Needing a painkiller to get through a run is a stop sign, and routine pre-race ibuprofen is linked to gut and kidney harm in distance runners.

Source Küster et al. (2013), BMJ Open 3(4):e002090
Soreness and injury differ: dull, symmetric, easing as the runner warms up is trainable easy, with intensity eased for a day or two after a hard session. Sharp, one-sided, or worsening means rest now, and pain persisting past a few days, or any pain in doubt, is a physio or a doctor.

Source Cheung, Hume & Maxwell (2003), Sports Med 33(2):145-164
Fever, chest symptoms, body aches, or gut illness mean no running until symptom-free, then a rebuild from an easy 10-20 minutes at about two-thirds effort, stepping up only after a session and its next day stay symptom-free. Above-the-neck sniffles only: easy is fine if they feel up to it, a screen rather than a green light. What is already planned was planned for someone who was well.

Source Schwellnus et al. (2022), Br J Sports Med 56(19):1066-1088
Chest pain, dizziness, fainting, or unusual breathlessness during a run means stopping immediately and seeking emergency medical care — local emergency services if symptoms are severe.

Source Thompson et al. (2007), Circulation 115:2358-2368 (AHA/ACSM)
Rising mileage with deliberate calorie restriction, a missed period, recurring stress injuries, or training and performance going flat or backwards reads as under-fueling (RED-S), which is a doctor's or a sports dietitian's to settle.

Source Mountjoy et al. (2023), Br J Sports Med 57(17):1073-1097
Hydration is drink-to-thirst, never a schedule: overdrinking on a fixed plan is what causes exercise-associated hyponatremia, and slower finishers who keep drinking are the risk group.

Source Hew-Butler et al. (2015), Clin J Sport Med 25(4):303-320
In heat, effort replaces pace and the session shortens; for a race it is the target that is revised, not the distance.

Source Armstrong et al. (2007), ACSM Position Stand, Med Sci Sports Exerc 39(3):556-572
A long-sedentary runner starts with walk-run and builds from there, and sees a doctor first with any known medical condition — heart, metabolic, or kidney disease especially — or symptoms that concern them.

Source Riebe et al. (2015), Med Sci Sports Exerc 47(11):2473-2479
An uncomplicated pregnancy with the provider's okay means running continues, coached by effort rather than pace; bleeding, regular painful contractions, fluid leakage, dizziness, or chest pain is an immediate stop-and-call. Starting to run while pregnant, from little or no running base, needs the provider's okay before the first session.

Source ACOG Committee Opinion 804 (2020), Obstet Gynecol 135(4):e178-e188
Postpartum, return is gated on healing, not fitness: typically not before about 12 weeks, and often longer — and any leaking, heaviness, or pelvic pain with impact means a pelvic health physio first.

Source Goom, Donnelly & Brockwell (2019), Returning to Running Postnatal — Guidelines for Medical, Health and Fitness Professionals (expert guideline, Level 4)
A running injury is pain that restricts running or needs a clinician; with one standing, recovery outranks pushing.

Source Yamato et al. (2015), J Orthop Sports Phys Ther 45(5):375-380; Brukner & Khan's Clinical Sports Medicine: Managing Injuries, 6th ed.