Ground Truth

The Truth Tracker

How to actually tell if it's working.

Every peptide page asks one honest question: is there any objective way to see the effect? When the answer is yes, the measurement is only as good as your prep. A DEXA scan taken fasted one month and fed the next tells you nothing. This is how to get a baseline you can actually compare against, method by method.

Log it all in the Truth Tracker— our free record for bloodwork, DEXA, genetics, and wearable data. We sell nothing; it's a resource.

DEXA (DXA) for Tracking Body Composition and Visceral Fat: How to Get a Baseline and Retest You Can Actually Trust

DXA tracks whole-body and regional FAT MASS, LEAN (fat-free) MASS, BONE MINERAL, and an estimated VISCERAL FAT (VAT) value. Under a standardized protocol it has good precision and is well-suited to confirming real recomposition trends over months and to flagging whether abdominal (visceral) fat is trending up or down. What it does NOT do, and where honesty matters: (1) "Lean mass" is not muscle protein — it lumps in water and glycogen, so hydration, carb, and creatine swings can move it 1-2%+ with zero change in actual muscle. (2) VAT is a MODELED estimate from a proprietary algorithm (GE's CoreScan, Hologic's InnerCore), not a direct CT/MRI slice; it is the machine's NOISIEST output, and results from GE and Hologic machines are not interchangeable. (3) It cannot see intramuscular fat, cannot distinguish fluid shifts from tissue change, and cannot tell you WHY a number moved. (4) VAT precision degrades at lower body size — the GE-validated method is most reliable at BMI ≥25 / VAT >~500 g. Treat DXA as a precise-but-confoundable trend tool, not a to-the-gram truth machine.

Prep for a reliable baseline

  • SAME MACHINE, SAME BRAND, SAME SOFTWARE. Book the identical scanner and site every time and confirm the software version has not changed. GE Lunar 'CoreScan' and Hologic 'InnerCore' use different algorithms and are NOT comparable for VAT (or reliably for anything else) - never mix brands between baseline and retest.
  • FASTED, EARLY MORNING, EVERY TIME. Recommend a consistent overnight fast of at least 8-12 hours and a first-thing-AM slot. A recent meal adds gut contents and fluid that read as tissue. If you truly cannot fast, replicate the EXACT same fed state (same meal, same timing) each scan - but fasted AM is the standard for a repeatable VAT baseline.
  • HYDRATION HELD CONSTANT. Drink your normal amount on your normal schedule; do not water-load or dehydrate before a scan, and do not scan right after chugging a large volume. Empty your bladder immediately before. Body water reads as lean mass, so hydration swings masquerade as tissue change.
  • NO EXERCISE FOR 24-48 HOURS PRIOR. Skip resistance training, hard cardio, sauna, and anything that shifts sweat, fluid, or glycogen. A workout the day before can inflate or deflate lean-mass and android-region readings.
  • HOLD CREATINE AND CARB-LOADING IDENTICAL. Creatine supplementation and glycogen each pull water into muscle and can raise the lean-mass reading by ~1-2%+ with ZERO real muscle gain (each gram of glycogen carries ~3 g of water). Either be off both with an equal washout before each scan, or keep the exact same dose/intake at baseline and retest. Do NOT start creatine or do a carb refeed between scans - it will manufacture a fake lean-mass gain.
  • AVOID GLYCOGEN/DIET SWINGS THE DAY BEFORE. Don't scan the morning after a big high-carb refeed or, conversely, a very-low-carb depletion day. Keep the preceding 24-48h of eating typical and similar between scans; glycogen and fluid swings move lean mass and can nudge the android/VAT region.
  • MINIMAL, IDENTICAL CLOTHING - REMOVE ALL METAL. Wear a thin gown or the same seamless, metal-free garment each time. Take off jewelry, watches, piercings, underwire bras, belts, zippers, buttons, and phone. Metal and thicker fabric corrupt attenuation and comparability.
  • NO RECENT CONTRAST OR NUCLEAR SCANS. If you've had CT/MRI contrast, a barium study, or nuclear-medicine imaging within about a week, reschedule - residual agent can distort the scan.
  • WOMEN: MATCH THE MENSTRUAL-CYCLE PHASE. Cyclic fluid retention shifts weight and lean readings, so try to scan at the same point in the cycle and record the cycle day.
  • BE WEIGHT-STABLE AT BASELINE. Don't set your baseline in the middle of a crash diet or a rapid-refeed week when fluid is in flux. Record everything: fast length, last meal, last workout, creatine status, hydration, garment, and cycle day - you'll need to reproduce it exactly.

How to run it

  • Establish baseline under the standardized conditions above and WRITE DOWN every condition (machine ID, software version, time of day, hours fasted, last meal, hours since last workout, creatine on/off and dose, garment, hydration, cycle day). This record is what makes the retest comparable.
  • Confirm with the facility that VAT is enabled (CoreScan on GE, InnerCore on Hologic) and that the same scan mode will be used. Empty your bladder right before the scan.
  • Positioning: lie centered and straight on the table with arms and legs in the standard placement; let the tech use positioning aids and make sure hands/feet aren't clipped by the scan edge. Ask that the same auto-segmentation/regions be applied every time.
  • Collect the FULL report: total and regional fat mass, lean mass, %fat, and VAT mass/volume/area. Ask the facility for their site-specific PRECISION ERROR and LEAST SIGNIFICANT CHANGE (LSC) - ISCD guidance is that reports state the LSC so you know the smallest real change.
  • For the retest, reproduce EVERY recorded condition - same scanner and software, ideally same technologist, same time of day, same fast, same 24-48h no-exercise gap, same creatine status, same garment, same cycle phase.
  • Interpret changes ONLY against the LSC. Any move smaller than the facility's LSC is statistical noise, not a real change. For VAT specifically, treat changes below roughly 130 g as noise (population LSC is ~130 g and precision is worse at lower BMI); demand a change that clearly exceeds the LSC before calling it real.
  • Judge the TREND across two or three standardized scans, not a single before/after pair. Cross-check surprising DXA moves against scale weight, waist tape, and training/diet logs so you can separate a fluid/glycogen artifact from a genuine tissue change.
  • Never compare your absolute number to a machine's 'normal' population range as if it were exact - compare each scan to your OWN prior scan on the SAME machine.

Baseline → retest

For total fat and lean mass, allow at least ~8-12 weeks between baseline and retest so a genuine change can outgrow the scan's precision error (some facilities' LSC allows a real signal at ~6-8 weeks if the change is large). For VAT, wait ~12+ weeks and require a change that exceeds the facility LSC (typically ~130 g, and larger in higher-BMI individuals) before believing it. Do NOT rescan every 1-3 weeks - the real biological change over that window is usually smaller than the machine's noise, so you'll be chasing artifacts.

What silently ruins comparability

  • · Switching scanner brands or letting a software update happen between scans - CoreScan (GE) and InnerCore (Hologic) VAT are not interchangeable, and even a version change breaks comparability.
  • · Scanning fasted one time and fed the next (or different fast lengths) - gut contents and fluid read as tissue.
  • · Any workout, sauna, or heavy sweat within 24-48h - shifts fluid and glycogen and fakes a lean-mass change.
  • · Starting, stopping, or changing creatine, or doing a carb refeed, between scans - manufactures a lean-mass swing of 1-2%+ with no real muscle change.
  • · Different hydration or a full bladder at scan time - water reads as lean mass.
  • · Metal left on, or thicker/different clothing - corrupts attenuation between scans.
  • · Different technologist, positioning, or auto-segmentation drift changing where regions are drawn.
  • · Judging a single number against population 'normal' ranges instead of against your own prior scan and the facility's LSC.
  • · Recent CT/MRI contrast, barium, or nuclear-medicine imaging distorting the scan.
  • · Reading VAT as precise to the gram - it is a modeled estimate and the least-precise DXA output, especially unreliable at low BMI / low VAT (GE guidance favors BMI >=25 or VAT >~500 g).
  • · For women, ignoring menstrual-cycle fluid differences by scanning at different cycle phases.

CGM for Detecting a Drug's Glucose Effect: Baseline Fortnight, Controlled Retest, and What It Cannot See

A 14-day continuous glucose monitor (Abbott FreeStyle Libre or Lingo, Dexcom G7 or Stelo) samples interstitial glucose every 1 to 15 minutes and is genuinely good at three things WHEN diet and activity are held constant: (1) directionally large, sustained shifts in mean glucose and GMI; (2) glycemic variability (SD, CV) and the size of post-meal excursions; and (3) within-person standardized-meal challenge responses such as incremental AUC, peak height, and time-to-baseline [evidence: moderate to strong]. It is the right tool for a drug or supplement expected to move glucose meaningfully (corticosteroids, GLP-1 agonists, metformin, berberine, high-dose niacin, SGLT2 inhibitors). What it does NOT do: it does not measure blood glucose (it reads interstitial fluid with a 5 to 15 minute physiologic lag and roughly 8 to 10% MARD versus lab reference), it cannot resolve small mean shifts (a few mg/dL sit inside sensor and biological noise), and it says nothing about a drug's mechanism, blood level, or any non-glycemic effect. In a metabolically healthy person, whole-fortnight time-in-range 70 to 140 mg/dL already sits near 96% (Shah 2019), so TIR has almost no room to move: a ceiling effect that makes it a poor sensitivity metric in non-diabetics. And CGM cannot by itself separate a true drug effect from placebo, regression to the mean, weight change, sleep debt, stress, menstrual phase, or seasonal drift. Only your experimental design can do that.

Prep for a reliable baseline

  • Pick ONE sensor family and keep it for the entire experiment (baseline AND retest). Never compare a Libre fortnight to a Dexcom fortnight: different enzyme chemistries, smoothing algorithms, and factory calibration make the raw numbers non-comparable [strong]. Note that OTC wellness sensors add processing (Lingo smooths and clips its displayed range; Stelo smooths and reports every 15 min), so if you want quantitative data, choose a sensor you can export from (Libre 3, Dexcom G7, or Stelo via app/CSV export).
  • Decide your ONE primary metric before you start, and the minimum change that would count as real. Recommended primary: incremental area-under-curve (iAUC, 0 to 2 h) of a fixed standardized test meal, averaged over 3 or more days per period. Secondary: mean glucose, CV, and time-above-140. Pre-committing to the metric and threshold is what stops you from cherry-picking a favorable number afterward [strong].
  • Build a standardized test meal you will eat identically in both periods: same product, same grams of available carbohydrate (e.g., 50 g), same clock time, after the same overnight fast (8 h or more), with no caffeine and no exercise in the 2 h before or after. This single controlled stimulus isolates the drug far better than free-living fortnight averages, because it removes meal-to-meal food variation [moderate to strong].
  • Standardize the background you are living in: fixed sleep and wake times, fixed (ideally zero) caffeine and alcohol, consistent meal composition and timing, no new supplements, and a written log. The drug is a small signal riding on top of food and activity, which are large signals. Hold the large signals constant or they bury the drug [strong].
  • Weigh yourself daily and record menstrual phase, illness, and unusual stress. A few pounds of weight change or a different cycle phase can shift glucose as much as a modest drug effect; you will use these logs to reject any contaminated day or window [moderate].
  • Get the drug timing right. If the agent needs days to reach steady state (most do), begin dosing far enough ahead that the entire on-drug fortnight is at steady state. If you are doing a washout crossover instead, allow at least 5 half-lives between the off and on windows [moderate].
  • Insert the sensor 12 to 24 h before data collection begins and plan to discard day-1 readings: first-day values commonly run low or erratic during warm-up [moderate]. Place it on the back of the upper arm per the maker's instructions, and use the same arm and site type in both periods.

How to run it

  • BASELINE FORTNIGHT (drug OFF): wear one sensor for 14 days on your standardized routine. On 3 or more non-consecutive days, run the standardized test meal and tag it in the app. Aim for greater than 70% sensor data capture across the 14 days; the international consensus treats 14 days at over 70% capture as the minimum for a representative glucose profile (Danne 2017; Battelino 2019) [strong].
  • Export the data, drop day 1, then compute baseline metrics: mean glucose, GMI, SD, CV (= SD / mean x 100), time-in-range for a range you fix in advance (70 to 140 mg/dL for a non-diabetic, or the clinical 70 to 180), and per-meal iAUC, peak, and time-to-baseline for each test meal. Record BOTH the average and the spread across your test-meal days: that day-to-day spread is your noise floor and you will judge everything against it.
  • Start the drug. Wait until it is at pharmacologic steady state before opening the retest window. Do not measure during the ramp-up unless the ramp itself is what you are studying [moderate].
  • RETEST FORTNIGHT (drug ON): apply a NEW sensor of the same family to the same site, and repeat the identical 14-day protocol, including the same 3-plus standardized test meals at the same clock times, same fasting duration, same everything. Keep weight, sleep, caffeine, alcohol, and activity matched to baseline [strong].
  • Compute the identical metrics for the on-drug fortnight (again dropping day 1). Compare primarily the test-meal iAUC (on versus off) and secondarily mean glucose, CV, and time-above-140.
  • Judge the difference against your noise floor, not against zero. A change counts as real only if it exceeds BOTH (a) the day-to-day spread you measured within each period and (b) the roughly 8 to 10% sensor error plus the several-mg/dL between-sensor bias. A 2 to 3 mg/dL mean shift, or a small TIR wiggle in a healthy person, is noise [strong].
  • To move from suggestive to credible: repeat the on/off blocks at least twice in randomized order (ABAB), have someone else dispense drug versus placebo so you are blinded, and hold to the pre-registered metric and threshold. An open-label single before/after (n=1) cannot separate the drug from placebo, expectation, or regression to the mean [strong].

Baseline → retest

Two measurement windows, each a full 14-day sensor wear (about 10 or more days of usable data after dropping warm-up), because the consensus minimum for a representative glucose profile is 14 days at over 70% capture (Danne 2017; Battelino 2019). Open the on-drug window ONLY after the drug has reached steady state: for most oral agents and supplements that means starting the drug 1 to 2 weeks before the on-drug fortnight, which puts the two windows roughly 3 to 6 weeks apart. Keep them as close together as the pharmacology allows. Windows separated by more than about 2 to 3 months invite seasonal, behavioral, and weight drift that will masquerade as a drug effect. For anyone who menstruates, align both windows to the same cycle phase (a 28-day offset, or a multiple of it), because the luteal phase raises glucose. For a fast wash-in/wash-out drug, a tighter ABAB crossover (on-off-on-off, each block with 3-plus test meals and at least 5 half-lives between switches) is meaningfully stronger than a single before/after pair.

What silently ruins comparability

  • · Treating two different sensors as one instrument. Every 14-day sensor is a separate physical device with its own bias (often several mg/dL, sometimes near 10%), so a baseline-sensor-versus-retest-sensor mean difference can be pure hardware, not drug. Mitigate by leaning on within-sensor standardized-meal contrasts and by treating small whole-fortnight mean shifts as noise [strong].
  • · Mixing brands or models mid-experiment. Libre, Dexcom G7, Stelo, and Lingo read differently and apply different smoothing; their numbers cannot be pooled or directly compared [strong].
  • · Including the warm-up day. Day-1 readings frequently run low or erratic, so keeping day 1 in one window fabricates a difference [moderate].
  • · Compression lows. Lying on the sensor at night depresses readings and inflates time-below-range and CV. If one fortnight had more side-sleeping on the sensor arm, the variability comparison is corrupted [moderate].
  • · Interferents. High-dose vitamin C can falsely raise some sensors; acetaminophen falsely raises older Dexcom chemistry (G6) and is largely mitigated on G7/Stelo but is still worth avoiding around measurements. Keep any such substances constant across both periods [moderate].
  • · Uncontrolled diet and activity, the single dominant confounder. A slightly higher-carb week or a skipped workout moves glucose more than most drugs do; without a fixed test meal and a matched routine you cannot attribute any change to the drug [strong].
  • · Wrong metric choice from ceiling/floor effects. In healthy people TIR 70 to 140 is about 96% (Shah 2019), so it barely moves and hides real effects; use excursion size, time-above-140, or test-meal iAUC instead [moderate].
  • · Drift in the person, not the drug: weight change, sleep debt, illness, menstrual phase, or stress shifting between the two windows can each imitate or mask a drug effect [moderate].
  • · Placebo, expectation, and regression to the mean. An open-label before/after with no blinding and no randomization cannot rule these out, so a positive n=1 result is only ever suggestive [strong].
  • · Dehydration and extreme skin temperature (sauna, ice bath, sunburned site) transiently distort readings; standardize hydration and avoid these around test meals [weak to moderate].

Tracking Health Changes With Sleep (Oura / WHOOP): Reliable Baselines and Honest Limits

What a consumer sleep wearable actually tracks well is RELATIVE change in your own overnight physiology, not absolute sleep architecture. STRONG signals (trust these): your own night-to-night resting/lowest heart rate and nocturnal HRV (rMSSD measured during sleep). Wrist/finger PPG heart-rate error is small (WHOOP HR precision error ~1.5% vs ECG), so a sustained RHR rise of ~5-10 bpm or a sustained HRV drop against YOUR baseline is a real, sensitive read on incomplete recovery, alcohol, illness, or acute stress. MODERATE-STRONG: total sleep time, sleep vs. wake, and especially sleep timing/regularity — modern devices detect sleep with ~90%+ sensitivity, and regularity is one of the more meaningful outputs. WEAK / DIRECTIONAL ONLY: sleep STAGES (deep/REM/light). These are inferred from movement + heart rate, not brain waves (EEG), so agreement with lab polysomnography is only moderate (4-stage Cohen's kappa ~0.45-0.55; stage agreement ~60-79%). Deep-sleep and REM MINUTES are rough estimates — usable as your own multi-night trend, useless as precise nightly numbers. Also weak: wake detection. Wearables overestimate sleep and miss brief awakenings (specificity for wake often ~50%), so 'sleep efficiency' and 'time awake' read optimistically. CANNOT DETECT / NOT FOR: diagnosing sleep disorders. It does not diagnose sleep apnea, insomnia disorder, restless legs/PLMD, or narcolepsy. Newer SpO2 / 'breathing disturbance' flags are screening hints, not diagnosis. Per AASM, consumer sleep technology cannot be used to diagnose or treat sleep disorders and is not a substitute for a clinical sleep study.

Prep for a reliable baseline

  • Pick ONE device and never cross-compare brands. Oura and WHOOP use different sensors and different staging algorithms; a metric is only valid against ITSELF over time. Wear it the same way every night (Oura: same finger, snug; WHOOP: snug band, correct placement).
  • Wear it continuously through a learning window before you trust any baseline. Personal baselines take time: Oura builds its HRV/RHR/Readiness baseline over ~14 days of wear; WHOOP compares against a rolling 30-day HRV baseline. Treat Readiness/Recovery scores in the first 2-4 weeks as unreliable.
  • Baseline during a 'boring normal' stretch: usual bed/wake times, no travel, no illness, typical (not peak, not deload) training, no unusual alcohol. Do NOT baseline over a vacation, race week, or while sick — you will bake a distorted reference into everything you compare later.
  • Standardize the bedroom: cool, dark, quiet, and the same every night. Room temperature and body temperature both move HRV and resting heart rate.
  • Lock in consistent sleep and wake times. Regularity is itself a tracked metric and it stabilizes all the others; irregular timing is the biggest self-inflicted source of noise.
  • Protect the pre-sleep window on baseline nights (this is where confounders hit hardest): no alcohol, no late/large meals, no late intense exercise, and keep caffeine well before bedtime, so the baseline reflects a genuinely rested you.
  • Keep a daily one-line confounder log: last alcohol (amount + time), last caffeine time, last meal time/size, hard workouts, illness symptoms, menstrual cycle day, and time-zone/DST changes. Without this log you cannot interpret any later change.

How to run it

  • 1. Set up one device and wear it continuously through the learning window (Oura ~14 days, WHOOP ~30 days). Ignore all scores during this period — the algorithm is still personalizing.
  • 2. Define your baseline as a ROLLING AVERAGE over multiple typical nights, not a single night. Research on night-to-night variability shows single nights are noisy; use >=7 nights for duration/regularity and 14 nights for HRV/RHR to get a stable number.
  • 3. Record the RELIABLE metrics as your primary read: nocturnal average HRV, resting/lowest heart rate, total sleep time, and sleep regularity/timing. Log deep and REM only as SECONDARY, directional context.
  • 4. Compute your personal normal RANGE, not a point: baseline mean plus/minus your typical night-to-night swing (roughly one standard deviation). A single night outside the range is noise; a sustained shift of the average is signal.
  • 5. Standardize retest conditions to match baseline exactly: same device and firmware, same wear, same bedroom, and the same evening behaviors (same caffeine cutoff, no alcohol, no late heavy meal, no unusual late workout, not jet-lagged, similar sleep timing).
  • 6. Retest as a multi-night average (>=7 nights, ideally 14), then compare NEW average vs BASELINE average. Never conclude a health change from one night against one night.
  • 7. Interpret directionally and honestly: a persistent RHR rise (~5-10 bpm) and/or HRV drop vs. baseline = incomplete recovery, illness, alcohol, or stress. For deep/REM, only weigh LARGE, sustained shifts and disregard small nightly wiggles (the stage estimate is not precise enough to support fine reads).
  • 8. Escalate to a clinician, not the app, for anything diagnostic: chronic short sleep, loud snoring or witnessed breathing pauses, a persistent SpO2/breathing-disturbance flag, unrefreshing sleep, or excessive daytime sleepiness. The wearable screens and flags; a home sleep apnea test or in-lab polysomnography diagnoses.

Baseline → retest

Baseline: a minimum of 7 typical nights, ideally a 14-night rolling average, and only AFTER the device's 2-4 week learning window. For HRV and resting heart rate, the 14-night rolling window is the practical sweet spot. To detect a real HEALTH or lifestyle change (new training block, weight change, alcohol cutback, treatment), compare a baseline multi-night window to a retest multi-night window separated by about 4-8 weeks — long enough for a true physiological shift to show above the noise. Acute events (a night of drinking, one bad night, onset of illness) appear within 1-3 nights, but those are event checks, not health-change retests. The meaningful comparison is always average-to-average over weeks, matched for season, cycle phase, and illness status — never day-to-day.

What silently ruins comparability

  • · Cross-device or post-update comparisons. Comparing Oura to WHOOP is meaningless, and a firmware/algorithm update can shift your 'deep sleep' overnight with zero change in your physiology. Note update dates in your log and re-baseline after a major algorithm change.
  • · Alcohol the night before a retest. Even moderate evening alcohol lowers nocturnal HRV, raises resting HR, suppresses first-half REM, and fragments second-half sleep — while making you FEEL like you slept fine. One drink invalidates a retest.
  • · Inconsistent caffeine timing. In a controlled trial, 400 mg of caffeine even 6 hours before bed cut objective sleep by more than an hour, and subjects did not notice. Varying your caffeine cutoff silently shifts your numbers.
  • · Late or large meals and late intense exercise, which raise overnight heart rate and depress HRV independent of any health change.
  • · Travel, time-zone changes, and daylight-saving shifts, which disrupt circadian timing and skew every metric for several days.
  • · Illness, fever, recent vaccination, and menstrual cycle phase (the luteal phase raises RHR and temperature and lowers HRV). Compare like phase to like phase; do not retest while sick or right after a shot.
  • · Wearing it differently or with data gaps: loose band, different finger, dead-battery nights, or skipping the learning window all corrupt the baseline.
  • · False precision on sleep stages and efficiency. Wearables overestimate sleep and under-detect wake (wake specificity often ~50%), so reading deep-sleep minutes or 'efficiency' to the decimal is reading noise. Treat stage minutes as ballpark trends only.
  • · Orthosomnia: chasing a single night's score and changing behavior on noise, or losing sleep to anxiety about the tracker. React only to multi-night averages.
  • · Assuming a 'good' score rules out a sleep disorder. It does not. Loud snoring, witnessed apneas, or daytime sleepiness still require a clinical evaluation regardless of what the app says.

VO2max: How to Track Real Aerobic-Fitness Change Without Fooling Yourself

VO2max is the single best number for your cardiorespiratory (aerobic) fitness: the integrated capacity of your heart, lungs, blood, and muscle mitochondria to deliver and use oxygen at maximal effort, expressed in mL of O2 per kg of bodyweight per minute. It is one of the strongest predictors of all-cause and cardiovascular mortality known, which is why the American Heart Association argues it should be treated as a clinical vital sign. It tracks the DIRECTION of an endurance-training program (or of detraining, aging, or a major health change) well over months, and it is meaningfully changeable: 6-13 weeks of real aerobic training raises it on the order of ~0.5 L/min in most people. What VO2max does NOT do: it is not a diagnosis and it does not tell you WHY the number moved. A drop can be lost fitness, or it can be anemia, a beta-blocker, poor sleep, heat, altitude, a recent illness, weight gain, or simply a bad test. It is not a heart-disease screen and never replaces a clinical work-up of symptoms. It says nothing about strength, metabolic health, or body composition per se. And critically, it carries a real measurement error band, so small moves, especially from a wrist wearable, are frequently noise, not signal.

Prep for a reliable baseline

  • Pick ONE modality and never switch it between tests. Treadmill, cycle ergometer, and outdoor running all yield different VO2max numbers (cycling typically runs several mL/kg/min lower than treadmill because it uses less muscle mass). A treadmill baseline and a bike retest are not comparable.
  • Standardize your pre-test body state and write it down: no vigorous or unusually long exercise for 24-48h, no alcohol for 24h, no large meal within 2-3h (light meal is fine), hold caffeine constant (either none for 3-6h, or the exact same dose every time), well hydrated, a normal night of sleep, and not sick or within ~2-4 weeks of a real illness.
  • Control the environment: same time of day, similar ambient temperature (avoid heat), same altitude, same shoes, same bike fit. Heat, dehydration, and altitude all depress the number independent of fitness.
  • For a lab (graded/CPET) baseline: choose a facility that uses a metabolic cart (indirect calorimetry) and a standardized incremental protocol (Bruce, modified Bruce, or an individualized ramp) run by a trained technician. This is the gold standard; test-retest ICC is ~0.93-0.95.
  • For a wearable baseline: use a chest-strap HR monitor rather than wrist-only optical HR if the device supports it (wrist HR error feeds straight into the estimate), and let the algorithm settle by doing several steady-effort outdoor runs or walks with good GPS over 1-2 weeks before you record the baseline number.
  • Log the exact setup: device model AND firmware version, lab/cart and protocol, HR source, route, conditions. If you can't reproduce the conditions later, you can't trust the comparison. Record weight too, since VO2max is per-kg.

How to run it

  • LAB PATH (most reliable): Book a graded cardiopulmonary exercise test (CPET) with a metabolic cart. Keep the same modality (treadmill or cycle) for every future test.
  • Arrive rested and fueled per the prep rules, at the same time of day. Do the lab's standardized warm-up, then the technician runs an incremental ramp to volitional exhaustion.
  • Push to a TRUE maximum. The result is only a valid VO2max if maximal-effort criteria are met: respiratory exchange ratio (RER) above ~1.10, a plateau in VO2 despite rising workload, RPE around 9-10/10, and heart rate near your age-predicted max. Ask the technician to confirm these on your report; a test you quit early is a VO2peak that undercounts you.
  • Take home the actual mL/kg/min value AND the raw data, not just a fitness category. You need the number to compare later.
  • WEARABLE / FIELD PATH (accessible, lower resolution): Pick ONE method and freeze it: your watch's estimate (with a chest strap), or a validated field test (Cooper 12-minute run, 1.5-mile run, or Rockport 1-mile walk) using the same published equation every time.
  • For a device estimate: complete 2-3 qualifying steady-effort runs in the days beforehand, then record the stabilized value or a 7-day average, never a single spike. For a field test: run/walk at genuine full effort on the same track/route, capture raw time and HR, and plug into the identical equation each time; do it twice within a few days and average to set the baseline.
  • Write the number down next to the logged conditions (device/firmware, HR source, temperature, sleep, weight, any illness) so that on retest you can separate real change from noise.

Baseline → retest

Give a real training change time to clear the measurement noise. LAB/CPET: retest no sooner than about 8-12 weeks after a consistent training block. Even the gold standard has a standard error of measurement around 1.7 mL/kg/min, and the smallest change you can confidently call real in an individual is roughly 3-5 mL/kg/min (in some clinical cohorts the individual smallest detectable change is as large as ~4.6 mL/kg/min, i.e. >20%). So treat a shift of at least ~3-5 mL/kg/min (or >~8-10%) as the threshold for a real change; smaller moves are within the error band. WEARABLES: never trust a single reading. The error is larger (mean absolute percentage error typically ~5-10%, often 3-5+ mL/kg/min) and it drifts with heat, HR source, and pace, so watch the rolling 6-12 week trend and treat sub-2-3 mL/kg/min wiggle as noise. Re-establish a fresh baseline any time you change device, firmware, HR strap, modality, or terrain, because those break comparability instantly.

What silently ruins comparability

  • · Changing modality between tests (treadmill vs bike vs outdoor run). Different muscle mass and mechanics produce systematically different numbers, so a 'change' can be pure method artifact.
  • · Accepting a submaximal effort as a 'max' test. Without RER >1.10, a VO2 plateau, or near-max HR/RPE confirmed on the report, you likely quit early and the value undercounts your true VO2max.
  • · Mixing wrist-only optical HR with chest-strap HR. Wrist HR error propagates directly into the estimate, and switching sources between baseline and retest silently corrupts the comparison.
  • · Firmware/algorithm updates. Garmin (Firstbeat) and Apple periodically revise their models, which can rescale your estimate overnight; a sudden 'drop' may be software, not fitness.
  • · Uncontrolled conditions. Heat, altitude, dehydration, a big meal, caffeine, poor sleep, and recent illness all move the number and masquerade as fitness change.
  • · Demographic and range bias. Consumer algorithms were trained largely on younger male data; women, older adults, the very unfit, and elite athletes (whose high values get underestimated) all see larger errors.
  • · Weight change. VO2max is expressed per kg, so losing or gaining weight shifts your mL/kg/min even when the heart-lung 'engine' (absolute L/min) is unchanged. Track weight and, ideally, absolute VO2 alongside it.
  • · Comparing across different labs, metabolic carts, or protocols, or across different field-test equations, without noting the change. Cross-lab differences alone can swamp a real training effect.
  • · Over-reacting to a single wearable reading instead of the multi-week trend, and failing to log conditions so you can never tell signal from noise on the retest.

Tracking Health Changes with Urine: A Reliable-Baseline How-To

Urine is an excretion snapshot, not a blood test in a cup. It genuinely tracks a few things well over time and quietly fails at most of what people hope it does. What it tracks reliably: (1) Early kidney damage and cardiovascular-risk via the urine albumin-to-creatinine ratio (uACR) from a first-morning void. Because it is normalized to creatinine, it corrects for how hydrated you are, so it is the single most repeatable, trend-able urine number a normal person can run. (2) Anything quantified over a correctly done 24-hour collection: total protein, kidney-stone risk chemistry (calcium, oxalate, citrate, uric acid, sodium, total volume), creatinine clearance, and 24-hour urinary free cortisol (UFC). A full-day collection averages out the big hour-to-hour swings, IF the collection is complete. (3) A standard urinalysis (dipstick + microscopic) as a yes/no SCREEN for a change of state: new protein, new blood, glucose spilling over, ketones, or signs of a urinary infection. What urine CANNOT do, and where people get fooled: it is not a real-time metabolic monitor. Urine glucose is a poor stand-in for blood glucose (sugar only spills above a personal kidney threshold around 180 mg/dL that varies widely person to person, and it never detects a low), so it cannot guide diabetes control the way blood glucose and HbA1c do. Dipstick results are coarse (negative/trace/1+/2+/3+), not precise numbers, so they are for screening, not fine trending. A raw spot concentration of almost anything is meaningless without creatinine normalization because hydration alone can double or halve it. Single spot hormone readings are noisy, and even 24-hour cortisol swings enough day-to-day that one value cannot be trusted. Consumer wellness or at-home hormone-metabolite urine panels are a lower evidence tier than the standardized lab tests above. Bottom line: use urine to trend kidney/albumin and to quantify a full day of excretion, and to screen for change, not to replace blood work for glucose, hormones, or organ function.

Prep for a reliable baseline

  • Pick ONE test and ONE collection type, then hold it constant forever. For everyday tracking that means first-morning uACR. For quantified panels (protein, stone risk, cortisol) it means a full 24-hour collection. Never compare a first-morning sample to an afternoon random sample; posture, meals, and exercise shift the numbers.
  • Use the same lab and the same assay each time. Albumin immunoassays and creatinine methods are not perfectly harmonized between labs, so switching labs can create a fake trend. Write the lab name and method on your record.
  • For a first-morning uACR, control the 24 hours before: no heavy or prolonged exercise (it causes transient albuminuria that looks like kidney damage), no alcohol binge, and no unusually large protein load or extreme over/under-hydration. Keep the night before ordinary.
  • Do not collect during a confounder: active urinary tract infection, fever or acute illness, or menstruation/visible blood (blood contaminates protein and heme readings). Reschedule instead of collecting through them.
  • For a 24-hour collection, get the CORRECT container in advance and ask the lab which preservative each analyte needs. Cortisol, catecholamines/metanephrines, and oxalate collections often require an acid preservative and must be kept cold and out of light. The wrong tube silently invalidates the result.
  • Have a clean-catch, midstream technique ready: brief external cleaning, start the stream into the toilet, catch the middle portion, finish into the toilet. This is what keeps skin cells, discharge, and bacteria out of the cup.
  • Plan delivery. Urine degrades at room temperature (pH drifts, cells lyse, bacteria multiply). Refrigerate and get the sample to the lab the same morning, or follow the mail-in kit's cold/timing instructions exactly.

How to run it

  • FIRST-MORNING uACR (the everyday tracker): Order 'urine albumin-to-creatinine ratio,' not a bare dipstick. The lab must report albumin AND creatinine and compute the ratio in mg/g.
  • On waking, do a clean-catch midstream collection of your FIRST void of the day, before coffee/exercise. Fill to the container line, cap, and label with the exact clock time.
  • Deliver refrigerated the same morning. Record four things every time: the ACR number, the lab/method, the date and time, and any confounders (workout, illness, cycle day). The number is only comparable alongside those notes.
  • 24-HOUR COLLECTION (protein, stone-risk chemistry, cortisol): Get the right preservative container(s) from the lab first.
  • At a fixed wake time, empty your bladder into the toilet and DISCARD it. Note that exact clock time. This is your start; the discarded void belongs to the previous day.
  • For the next 24 hours, collect EVERY void into the container. Keep it refrigerated (and dark, if acid-preserved) the whole time. Missing even one void understates the true total.
  • At the same clock time the next morning, do one final void INTO the container to close the 24-hour window, even if you have to go early or force it. Deliver promptly.
  • Validate completeness before trusting the result: the lab reports total volume and total creatinine. Adult creatinine excretion should land near 15 to 20 mg/kg/day for women and 20 to 25 mg/kg/day for men. A value well below that means you missed voids and the panel understates your real excretion, so repeat it.
  • For 24-hour cortisol (UFC) specifically, plan at least TWO separate collections up front, because a single day's value is too variable to act on.

Baseline → retest

"Match the spacing to how noisy the marker is. uACR: short-term biological variability is large (day-to-day coefficient of variation often 30 to 50 percent), so a single change is not a signal. KDIGO defines persistent albuminuria as 2 of 3 abnormal uACR results over roughly a 3-month window, so treat about 3 months as the minimum baseline-to-retest interval, and only believe a change that clearly exceeds normal variation (roughly a 30 to 40 percent shift, in the same direction, confirmed on a repeat). 24-hour quantified panels (protein, stone-risk chemistry): after a real intervention like a diet or fluid change, wait about 8 to 12 weeks so the change can register, and re-run the completeness check each time. 24-hour urinary free cortisol: at least two collections at baseline, and retest only when there is a clinical reason, because day-to-day variability is very high. Routine urinalysis screen: annually or when symptoms change; it is a change-detector, not a fine-trend instrument."

What silently ruins comparability

  • · Not normalizing a spot sample to creatinine. A raw protein or concentration reading swings with hydration alone, so an afternoon-vs-morning comparison of un-normalized dipstick values is noise, not a trend.
  • · Mixing collection types across time points (first-morning once, random later). Posture, exercise, and meals change albumin excretion, so this manufactures fake changes.
  • · Incomplete 24-hour collection. Studies find over 30 percent of collections are incomplete and therefore understate true excretion; a missed first-discard or missed final void is the single biggest error source. Always check the creatinine to confirm completeness.
  • · Wrong or missing preservative, or letting acid-sensitive analytes (cortisol, catecholamines, oxalate) warm up or sit in light. This silently corrupts the value with no visible sign.
  • · Switching labs or assays between tests. Albumin and creatinine methods are not fully harmonized, so a lab change can look like a health change.
  • · Collecting through a confounder: recent heavy exercise (transient albuminuria), fever or acute illness, an active UTI, or menstrual blood/vaginal discharge contaminating protein and blood readings.
  • · Reading dipstick color blocks outside the timed read window or by eye in poor light; over-reading past the window gives false positives. Dipsticks are semi-quantitative screens, not measurements.
  • · Substance interference: vitamin C (ascorbate) can cause false-negative dipstick blood and glucose; beets, rifampin, and other foods/drugs change color; stale or alkaline urine degrades cells and analytes before testing.
  • · Treating urine glucose or ketones as a substitute for blood glucose or blood ketones. Urine glucose only spills above a personal renal threshold (~180 mg/dL, and higher in insulin resistance) and never detects a low, so it is unfit for glycemic monitoring.
  • · Acting on one value instead of the trend, and ignoring the reference-change threshold. Confirm before you conclude.

Stool Testing on Ground Truth: An Honest Guide to Tracking Gut Changes Over Time

What stool actually tracks well (validated, regulator- or guideline-backed): (1) Gut mucosal inflammation via fecal calprotectin, a quantitative neutrophil marker that reliably separates inflammatory bowel disease from IBS and tracks IBD activity (pooled sensitivity ~93%, specificity ~96% in adults; AGA uses a <150 ug/g cutoff to rule out active inflammation) [evidence: STRONG]. (2) Occult blood and colorectal-cancer / advanced-adenoma risk via the fecal immunochemical test (FIT) and multitarget stool DNA (Cologuard), both FDA-cleared and USPSTF-endorsed screening tools [STRONG]. (3) Pancreatic exocrine function via fecal elastase-1 [MODERATE-STRONG]. (4) Active GI infection via stool pathogen PCR/culture, C. difficile toxin, ova-and-parasite exams, and H. pylori stool antigen [STRONG, but for acute diagnosis, not trend-tracking]. What stool CANNOT reliably do (do not track these as personal metrics): diagnose disease or grade "gut health" from direct-to-consumer microbiome composition (16S/shotgun) tests; detect "food sensitivities," "leaky gut," "candida overgrowth," "dysbiosis scores," or "detox" capacity; localize where in the gut a problem is; or replace endoscopy. Calprotectin is a flag, not a diagnosis: it cannot distinguish Crohn's from ulcerative colitis from an NSAID injury. Microbiome relative-abundance readouts are dominated by lab/extraction/bioinformatics artifacts and shift within 24-48h of a diet change, so they are exploratory, not a repeatable metric. A single ova-and-parasite exam misses infections that intermittent shedding and PCR would catch. Net for a normal person tracking change over time: fecal calprotectin is the one genuinely trackable quantitative signal; anything marketed as a "microbiome tracker" or "gut health score" is not.

Prep for a reliable baseline

  • Decide the QUESTION before buying any kit, because it dictates the test: monitoring known/suspected gut inflammation points to calprotectin; colorectal-cancer screening points to FIT or Cologuard (on a schedule, not as a tracker); curiosity about your microbiome has no validated tracking answer. Buying a random 'gut test' first is how people generate noise they then over-interpret.
  • Lock ONE lab and ONE assay/kit brand and write down the exact product (e.g., BUHLMANN fCAL for calprotectin). Results are method-dependent: a calprotectin value from one assay is not interchangeable with another, and microbiome results from different labs/pipelines are effectively different measurements. Retests must use the identical method or the comparison is meaningless.
  • For calprotectin, plan to collect the FIRST bowel movement of the morning. Calprotectin rises the longer stool sits in the colon between movements, so first-morning stool is both highest and the most standardizable time point (Lasson 2015).
  • Clear known confounders in the days before sampling: stop NSAIDs/aspirin (they raise calprotectin), note PPI use, avoid sampling during or right after an acute GI infection, and (for FIT/blood-based readouts) avoid menstruation and obvious rectal bleeding. If you want a stable baseline, do not sample during a flare or a stomach bug.
  • For any microbiome test, hold your diet stable for at least several days beforehand (the gut community shifts within 24-48h of a diet change) and record any antibiotics in the prior weeks-to-months (a single antibiotic course perturbs the community for weeks and sometimes far longer). Without this, a 'change' is just your last few meals or a prescription.
  • Use the correct stabilized collection device and read its instructions before collection day (e.g., BUHLMANN CALEX Cap stabilizes calprotectin and elastase for safe room-temperature transport). Uncontrolled room-temperature delay degrades the sample; storing stool at room temperature more than ~3 days is not advisable for calprotectin.
  • Prepare a simple log to fill in every time: date, exact time of collection, Bristol stool form, recent meds (NSAIDs, PPIs, antibiotics), recent travel/illness, menstrual status, and the specific product/lot used. This log is what makes two data points comparable months apart.

How to run it

  • Confirm the test matches your question and is validated for it. Calprotectin = inflammation monitoring. FIT/Cologuard = cancer screening. Microbiome composition = exploration only, not a metric. If a product promises a 'gut health grade' or 'food sensitivity map,' treat it as entertainment, not data.
  • Register the method: same lab, same assay/kit, same reported units, and the reference cutoff that lab uses (for calprotectin the common decision points are roughly <50 ug/g normal, 50-150 borderline, >150 suggests active inflammation; cutoffs are assay-specific, so use YOUR lab's).
  • Collect the baseline on a non-confounded day: first-morning bowel movement, sampled per the device instructions. Because a single spot of a formed stool is not representative, follow the kit's sampling pattern (multiple sites) rather than one dab.
  • For calprotectin specifically, consider two samples on two separate mornings and average them. Day-to-day biological variability is large, and a single value can mislead; two well-separated samples materially reduce the odds of a spurious high or low.
  • Ship the same day per instructions and keep cold if the kit requires it. Do not let the sample sit over a weekend.
  • Record the numeric result WITH units and the lab's cutoff, plus your full log entry. A bare number with no method, units, or conditions is not trackable.
  • Retest under identical conditions only: same lab, same kit, same first-morning timing, diet-stable, no new medications, not during an illness. Change any one of these and you are no longer measuring the same thing.
  • Apply a real-change threshold before reacting. For calprotectin, treat a shift as meaningful only if it crosses a clinical threshold (e.g., moves from <150 to >150 ug/g) or is roughly a two-fold change, because biological plus analytical variability easily produces swings like 60 to 110 with no true change. For microbiome outputs, treat any single-taxon difference as a hypothesis, not a result.
  • Confirm surprises by repeating before acting, and route anything clinical to a clinician: a positive FIT or Cologuard requires a diagnostic colonoscopy (do not just re-run the stool test), and a rising or persistently elevated calprotectin warrants medical evaluation, not self-treatment.

Baseline → retest

Fecal calprotectin: never trust one value; a monitoring retest is typically 8-12 weeks apart (e.g., before vs. after a treatment or a deliberate dietary change), and a real signal must exceed known variability (crossing a clinical cutoff such as <150 vs. >150 ug/g, or roughly a two-fold change) rather than a small wiggle. Sampling the same first-morning stool with two samples per timepoint tightens the comparison. Microbiome composition: there is NO validated self-tracking cadence; if you insist on repeating, keep it many months apart under identical diet/kit/lab conditions and interpret differences as hypotheses, not measured change, because method and handling artifacts dominate. FIT and Cologuard are screening intervals, not trends: FIT every 12 months and multitarget stool DNA every 3 years per USPSTF; do not repeat them sooner as a "health metric," and never substitute a repeat stool test for the colonoscopy a positive result requires.

What silently ruins comparability

  • · Switching assay, lab, or kit brand between baseline and retest. Calprotectin values are method-dependent, and microbiome results vary most with biospecimen handling, DNA extraction, and bioinformatics pipeline (MBQC/Sinha 2017) rather than with your actual gut, so a cross-method comparison measures the lab, not you.
  • · Sampling at different times of day or from a different bowel movement. Calprotectin increases with a longer interval between movements, so a first-morning baseline vs. an afternoon retest can look like a real change when nothing changed (Lasson 2015).
  • · Sampling one spot of a formed stool instead of following the kit's multi-site pattern. Both calprotectin and microbes are unevenly distributed within a single stool, so single-dab sampling adds avoidable noise.
  • · Letting the sample sit at room temperature or ship slowly. Degradation before analysis (>~3 days at room temperature for calprotectin) quietly lowers or distorts the result unless a stabilizing device is used.
  • · Transient confounders you forget to log: NSAIDs/aspirin and recent GI infection raise calprotectin; menstrual blood and rectal bleeding affect blood-based readouts; a recent colonoscopy prep disturbs everything. Any of these can masquerade as a trend.
  • · Recent antibiotics before a microbiome test. A single course perturbs the community for weeks to months, so a post-antibiotic 'baseline' is not your steady state.
  • · Diet changes right before a microbiome sample. The community responds within 24-48h (Allaband 2019), so your last few days of eating, not a stable trait, drive the result.
  • · Over-interpreting a single value given large intra-individual biological variability, or reading meaning into small numeric moves that fall inside assay noise.
  • · Treating a direct-to-consumer 'gut health score,' 'dysbiosis index,' or 'food sensitivity' output as a validated, trackable biomarker. These are not clinically validated and are not comparable across time or across companies.
  • · Comparing microbiome relative abundances across timepoints without absolute quantification: relative abundance of one taxon can move purely because a different taxon changed, creating phantom trends.

None of this is medical advice. Tracking tells you whether a number moved, not whether a change is safe or worth it. Talk to a clinician before acting on anything you measure.

← Back to the peptides

New posts, every couple weeks

We're still wiring up the mailing list. Want the next post when it goes up? Email us and we'll add you by hand. No spam, nothing for sale, unsubscribe whenever.

Email to get on the list →