Why Fitness Can Feel Worse Before It Feels Better — A Research-Grade Synthesis

Abstract

A new training block often feels worse before it feels better, and the peer-reviewed literature treats this as the rule, not the exception. The foundational Banister fitness-fatigue model — first formalised in 1980 and refined in 1990 — separates a training response into a slow-building fitness component and a fast-decaying fatigue component, and the net performance at any moment is the difference between them ([1] Banister et al. 1980, Level 5; [2] Busso et al. 1990, Level 5; [3] Halson 2014, Level 5). The practical read: a rower who climbs training load by more than about 10 percent per week will see measurable fatigue signals within 7 to 14 days, and performance will often dip before the underlying fitness adaptations catch up ([4] Hulin et al. 2014, Level 2b; [3] Halson 2014, Level 5; [5] Plews et al. 2017, Level 2b). The honest read: feeling worse first is a physiological prediction, not a sign that the training is broken — and the rower who reads the signals in their logbook, widens the corridor, and waits for the supercompensation window is reading the model correctly.

The fitness-fatigue model in plain language

The 1980 [1] Banister, Calvert, Savage & Bach paper introduced the foundational model: training load is an impulse that drives a positive fitness response and a negative fatigue response, and the difference between them is the observable performance at any moment. The 1990 [2] Busso et al. paper formalised the model mathematically — fitness builds slowly (a time constant of about 50 to 60 days) and decays slowly (about 60 to 90 days), while fatigue builds quickly (a time constant of about 7 to 14 days) and decays quickly (about 14 to 21 days) ([1] Banister et al. 1980, Level 5; [2] Busso et al. 1990, Level 5).

The 2014 [3] Halson review consolidated the operational consequence: any training load increase raises both fitness and fatigue in the short term, and because fatigue decays faster than fitness builds, the net performance can dip for one to four weeks before the underlying adaptations catch up ([3] Halson 2014, Level 5). The 2022 [22] Foster, Rodriguez-Marroyo & de Koning review is the modern consensus: polarised training distribution — about 80 percent below LT1 and about 20 percent above LT2 — is the distribution that maximises the fitness component per unit of fatigue incurred ([22] Foster et al. 2022, Level 5; [16] Seiler & Kjerland 2006, Level 4; [17] Guellich et al. 2009, Level 2b).

The practical read: the rower who starts a new training block and feels worse in the first two to four weeks is reading the model correctly. The fatigue is real, the fitness is also being built, and the net performance will catch up once the fatigue component decays faster than the fitness builds. The 2008 [11] Ingham et al. paper is the rowing-specific experimental anchor: in 18 trained rowers, the LOW group (polarised distribution) gained 23.5 ± 12.2 W at LT over 12 weeks vs 5.1 ± 5.0 W in the MIX group — the polarised pattern is the training shape that converts accumulated load into measurable fitness gains ([11] Ingham et al. 2008, Level 2b).

The acute:chronic workload ratio (ACWR)

The 2014 [4] Hulin, Gabbett, Lawson, Caputi & Sampson paper formalised the ACWR: the ratio of acute training load (typically the past 7 days) to chronic training load (typically the past 28 days). The paper found that an ACWR above about 1.5 was associated with elevated injury risk in elite rugby, and below about 0.8 was associated with de-training patterns ([4] Hulin et al. 2014, Level 2b). The 2020 [23] Impellizzeri et al. review is the methodological anchor for how to interpret ACWR: it is one signal among several, not a verdict, and the practical sweet spot is between about 0.8 and 1.3 ([23] Impellizzeri et al. 2020, Level 2b).

The 2014 [3] Halson review is the consensus anchor: ACWR is the most-cited summary index for whether load has escalated faster than the body has adapted, and the rower who reads their logbook with ACWR above 1.5 should treat the next 7 to 14 days as a high-risk window for injury and overreaching ([3] Halson 2014, Level 5; [4] Hulin et al. 2014, Level 2b). The 2020 [20] Seiler review is the modern practical anchor: the optimal ACWR is not the maximum — it is the ratio that allows the fitness component to build without the fatigue component exceeding the body's recovery capacity ([20] Seiler 2010, Level 5).

The practical read: ACWR and the 10-percent-per-week rule are two different guardrails, and it is worth being precise about how they relate. A steady 10 percent weekly increase does not on its own push ACWR above 1.5 — taking acute load as the current week and chronic load as the trailing 4-week mean, steady 10 percent growth settles at an ACWR of about 1.15, and it would take roughly 35 percent per week sustained to reach 1.5. What actually drives the ratio above 1.5 is a spike: a single week that jumps well clear of the preceding month, or a return to full load after a break has pulled the chronic average down. That second case is the common one for a rower coming back from a layoff, and it is where the first fatigue signals (sleep disruption, soreness, mood dip, HRV drop) tend to appear. The 2014 [3] Halson review is the methodological anchor: the ACWR is most useful as a leading indicator of overreaching risk, not as a measure of the underlying training response.

The two-timescale fatigue signal

Fatigue signals come in two timescales: acute (days) and chronic (weeks). The 2014 [3] Halson review is the consensus anchor: acute signals — sleep disruption, increased soreness, mood dip, morning heart-rate elevation, HRV drop — appear within days of a load increase; chronic maladaptation signals — sustained mood decline, persistent motivation loss, sustained HRV drop over weeks — appear on the chronic timescale ([3] Halson 2014, Level 5). The 2016 [14] Bellenger et al. meta-analysis is the methodological anchor for HRV: across 39 studies, daily rMSSD is highly individual, and the actionable signal is the 7-day rolling trend against the 60- to 90-day baseline ([14] Bellenger et al. 2016, Level 2b).

The 2017 [5] Plews et al. paper is the rowing-specific anchor: in world-champion rowers, daily HRV is very individual, and the reference values from one athlete do not transfer to another ([5] Plews et al. 2017, Level 2b). The 2014 [15] Plews et al. paper added the rowing-specific training-distribution angle: HRV tracks polarised vs threshold training distribution over a season in elite rowers ([15] Plews et al. 2014, Level 2b). The 2013 [13] Buchheit review is the methodological anchor: rMSSD and Ln rMSSD trends over weeks, not single sessions, are the actionable signal ([13] Buchheit 2013, Level 5).

The practical read: the rower who sees a 7-day rMSSD drop of 10 to 15 percent below a 60- to 90-day baseline is in the consensus overreaching warning band. The rower who sees a daily rMSSD drop of 10 percent on a single day is in the noise. The 2014 [3] Halson review is the methodological anchor: the gap between acute and chronic signals is the rower's most useful warning lead time.

Test results and the moving-average protocol

A single test result is a snapshot of the current fitness-fatigue balance, not a verdict on the underlying training response. The 1999 [10] Schabort, Hopkins & Hawley paper is the methodological anchor: under standard conditions, a 2K rowing-ergometer test has a within-subject coefficient of variation around 1 percent — and that figure applies to familiarised trials, not to the first effort post-layoff ([10] Schabort et al. 1999, Level 2b). The 2009 [21] Hopkins et al. paper added the statistical-significance framework: the smallest worthwhile change in 2K time is about 0.3 to 0.5 percent, and a single test that drops by more than that is not necessarily a real change ([21] Hopkins et al. 2009, Level 5).

The 2002 [29] Ingham et al. paper is the rowing-specific anchor: in 41 elite rowers, a regression model with power at VO2max, VO2 at lactate threshold, power at 4 mmol/L lactate, and peak power explained 98 percent of 2K speed variance ([29] Ingham et al. 2002, Level 2b). The 2004 [12] Bourdin et al. paper added the peak-power anchor: in 54 male rowers, peak power output correlated with 2K time at r=0.92 ([12] Bourdin et al. 2004, Level 2b). The 2012 [9] Smith & Hopkins paper is the methodological anchor for predicting 2K performance from independent variables ([9] Smith & Hopkins 2012, Level 2b).

The practical read: the rower who watches a 2K drop from 7:00 to 7:05 in the first month of a new training block is reading the fitness-fatigue model correctly. The 5-second drop is well within the within-subject corridor plus the model's predicted fatigue dip, and the moving average of the next two to three tests is the honest read. The 2009 [21] Hopkins et al. paper is the methodological anchor: a single test below the smallest worthwhile change is noise; a sequence of tests below the change is a real shift.

Fatigue and the lactate-threshold shift

The [19] Poole, Rossiter, Brooks & Gladden 2021 review is the methodological anchor on what the lactate threshold actually measures: it is a transition zone, not a single line, and the threshold can shift up or down depending on the current fitness-fatigue balance ([19] Poole et al. 2021, Level 5). The 2022 [22] Foster et al. review is the consensus anchor: a training block that successfully builds the fitness component will see the lactate threshold move up over weeks, but the fatigue component can drag it down in the short term ([22] Foster et al. 2022, Level 5).

The 2008 [11] Ingham et al. paper is the rowing-specific experimental anchor: in 18 trained rowers, the LOW group (polarised distribution) gained 23.5 ± 12.2 W at LT over 12 weeks vs 5.1 ± 5.0 W in the MIX group — the polarised pattern is the training shape that converts accumulated load into measurable threshold gains ([11] Ingham et al. 2008, Level 2b). The 2019 [24] British Rowing Moseley article is the federation anchor: in elite rowers, the polarised distribution is about 80 percent of sessions below 2 mmol/L and about 20 percent above 4 mmol/L ([24] British Rowing — Moseley 2019, Level 5). The 2009 [17] Guellich et al. paper is the rowing-specific anchor for the polarised pattern: in 36 German junior finalists over 37 weeks, 95 percent of rowing time was below 2 mmol/L lactate ([17] Guellich et al. 2009, Level 2b).

The practical read: a rower who trains within the polarised distribution will see the lactate threshold shift up over weeks, but the fatigue component will hold it back for the first two to four weeks of a new block. The 2010 [20] Seiler review is the methodological anchor: the polarised distribution is the training shape that maximises the fitness component per unit of fatigue incurred ([20] Seiler 2010, Level 5).

The federation and manufacturer guidance points the same way. The [25] Concept2 training resources are the manufacturer anchor for structuring training blocks and tapers, and the [27] USRowing education hub is the US federation anchor for training-block and recovery guidance ([25] Concept2 — Indoor Rowers training, Level 5; [27] USRowing — Education resources, Level 5). The [28] British Rowing Performance Talent testing protocols are the federation anchor for testing cadence: tests are scheduled at fixed points in the block rather than whenever the rower feels fast, which is what makes the moving average readable at all ([28] British Rowing — Performance Talent testing protocols, Level 5). The [26] World Rowing disciplines page is the international-federation anchor for indoor rowing as a recognised discipline, which is why this cross-sport fitness-fatigue literature transfers to the erg ([26] World Rowing — Disciplines, Level 5).

Tapering and the inverse curve

Tapering is the inverse of the feel-worse-before-better curve. The 2003 [7] Mujika & Padilla paper is the foundational tapering review: a 7 to 14 day taper converts accumulated training into measurable performance gains by letting the fatigue component decay faster than the fitness component ([7] Mujika & Padilla 2003, Level 5). The 2000 [6] Mujika & Padilla review is the detraining-side anchor: fitness detrains faster than previously thought, but the rate depends on the underlying training status and the duration of the taper ([6] Mujika & Padilla 2000, Level 2b). The 1985 [18] Coyle et al. detraining paper is the methodological anchor: short detraining reduces VO2max within 12 days and the lactate threshold buffer earlier ([18] Coyle et al. 1985, Level 2b).

The practical read: the rower who plans a 7 to 14 day taper before a goal event will see measurable performance gains not because the training has changed but because the fatigue component has decayed faster than the fitness component. The 2003 [7] Mujika & Padilla paper is the methodological anchor: the taper is the model's most reliable way to convert accumulated training into a single competition result ([7] Mujika & Padilla 2003, Level 5). The 2014 [3] Halson review is the consensus anchor: the taper is also the most sensitive window for overreaching detection, because the training load drops while the body is still recovering ([3] Halson 2014, Level 5).

Overreaching and the warning band

The 2013 [8] Meeusen et al. consensus statement is the diagnostic anchor: functional overreaching is a planned, beneficial training state that resolves within a few weeks of tapering; non-functional overreaching is a maladaptive state that lasts for weeks to months and requires a significant recovery period; overtraining syndrome is a chronic maladaptive state that requires months of recovery ([8] Meeusen et al. 2013, Level 5). The 2014 [3] Halson review is the methodological anchor: HRV, hormonal markers, mood-state questionnaires, and performance tests are the multi-marker diagnostic inputs, and a single low reading is not a verdict ([3] Halson 2014, Level 5).

The 2014 [4] Hulin et al. paper is the practical anchor: ACWR above 1.5 is the published threshold for elevated injury and fatigue risk, and the rower who reads ACWR above 1.5 should treat the next 7 to 14 days as a high-risk window ([4] Hulin et al. 2014, Level 2b). The 2020 [23] Impellizzeri et al. review is the methodology anchor: ACWR is one signal among several, and the practical sweet spot is between about 0.8 and 1.3 ([23] Impellizzeri et al. 2020, Level 2b).

The practical read: the rower who reads the logbook with ACWR above 1.5, a 7-day rMSSD trend down 10 to 15 percent, and a 2K test that drops more than the within-subject corridor should treat the next 7 to 14 days as a recovery window, not a continued climb. The 2013 [8] Meeusen et al. consensus is the methodological anchor: the gap between acute and chronic signals is the rower's most useful warning lead time.

What the AI coach does with the fitness-fatigue signal

For an AI coach that reads the rower's logbook and writes the rower's session, the fitness-fatigue model is the operational anchor. The coach's rule is:

• A rower who has just started a new training block and feels worse — the coach widens the corridor, watches the 7-day rMSSD trend and the 2K moving average, and waits for the supercompensation window. The 1980 [1] Banister paper is the methodological anchor: a 1 to 4 week dip is the model's prediction, not a verdict on the underlying training response ([1] Banister et al. 1980, Level 5; [2] Busso et al. 1990, Level 5).
• A rower whose ACWR has climbed above 1.5 — the coach widens the recovery load, reduces the planned intensity, and watches the trend for 7 to 14 days before resuming the planned progression. The 2014 [4] Hulin et al. paper is the methodological anchor ([4] Hulin et al. 2014, Level 2b; [3] Halson 2014, Level 5).
• A rower whose 7-day rMSSD trend has dropped 10 to 15 percent below the 60- to 90-day baseline — the coach widens the recovery load, watches for the trend to reverse, and waits for the autonomic baseline to recover before resuming the planned climb. The 2016 [14] Bellenger et al. meta-analysis is the methodological anchor ([14] Bellenger et al. 2016, Level 2b; [13] Buchheit 2013, Level 5).
• A rower whose 2K test has dropped more than the within-subject corridor — the coach notes the result, watches the next two to three tests, and waits for the moving average to confirm a real downturn before reading the result as a fitness drop. The 1999 [10] Schabort et al. paper is the methodological anchor ([10] Schabort et al. 1999, Level 2b; [21] Hopkins et al. 2009, Level 5).
• A rower in a planned taper before a goal event — the coach holds the taper window, lets the fatigue component decay, and waits for the supercompensation to register on the next test. The 2003 [7] Mujika & Padilla paper is the methodological anchor ([7] Mujika & Padilla 2003, Level 5).
• A rower whose logbook shows persistent mood, motivation, and HRV drop over weeks — the coach widens the recovery window, leans on the 2013 [8] Meeusen et al. consensus multi-marker diagnostic, and prescribes a 7 to 14 day recovery block before resuming the planned progression ([8] Meeusen et al. 2013, Level 5).

The coach that treats a single bad session as a calibration of physiology, treats a 2K dip in the first month of a new block as a verdict, or ramps load faster than about 10 percent per week is over-fitting the model. The coach that reads the logbook as a fitness-fatigue signal, treats single-session variability as noise, and waits for the moving average to confirm or refute a real downturn is reading the model correctly.

Limitations and open questions

The fitness-fatigue model is a parsimonious approximation. The 1980 [1] Banister paper and the 1990 [2] Busso et al. paper formalised it as a two-component impulse-response model, but the underlying biology is more complex — peripheral fatigue, central fatigue, autonomic balance, hormonal environment, and psychological state all interact in ways the model does not capture. The 2014 [3] Halson review is the methodological anchor: the model is most useful as a first-order summary, not as a complete description of the training response ([3] Halson 2014, Level 5).

The ACWR has known methodological limitations. The 2014 [4] Hulin et al. paper introduced ACWR, and the 2020 [23] Impellizzeri et al. review is the methodological anchor for the 1.5 threshold: the original paper used a 7-day-to-28-day definition, but other definitions (acute 3-day or 7-day, chronic 21-day or 42-day) produce different thresholds, and the practical sweet spot is between about 0.8 and 1.3 ([23] Impellizzeri et al. 2020, Level 2b). The 2014 [4] Hulin et al. paper is the most-cited anchor, but the rower who reads ACWR as a verdict is over-fitting the model.

The HRV-based overreaching warning band is one signal among several. The 2016 [14] Bellenger et al. meta-analysis is the methodological anchor: across 39 studies, rMSSD trends over weeks track autonomic readiness, but the actionability of a single 10 to 15 percent drop is bounded by the rower's individual baseline ([14] Bellenger et al. 2016, Level 2b). The 2013 [8] Meeusen et al. consensus is the diagnostic anchor: HRV is one input alongside hormonal markers, mood-state questionnaires, and performance tests, and the multi-marker diagnosis is the honest read.

The rowing-specific fitness-fatigue model literature is small. Most of the foundational papers are in cycling, running, or general exercise physiology, with rowing-specific work concentrated on the 2008 [11] Ingham et al. LOW-vs-MIX trial, the 2002 [29] Ingham et al. 2K-determinant paper, and the 2017 [5] Plews et al. world-champion-rower HRV paper. The reader should weight the rowing-specific evidence more heavily than the cross-sport evidence when the two diverge.

What to do with this article

Read the principle: a new training block often feels worse before it feels better, and the Banister fitness-fatigue model predicts this as the rule, not the exception. The 1980 [1] Banister paper (Level 5) and the 1990 [2] Busso et al. paper (Level 5) are the foundational anchors; the 2014 [3] Halson review (Level 5) is the modern consensus. Read the evidence: the 2014 [4] Hulin et al. paper (Level 2b) anchors the ACWR with a 1.5 threshold; the 2016 [14] Bellenger et al. meta-analysis (Level 2b) and the 2013 [13] Buchheit review (Level 5) anchor HRV trends over weeks; the 1999 [10] Schabort et al. paper (Level 2b) anchors the within-subject test-retest variation; the 2003 [7] Mujika & Padilla paper (Level 5) anchors the taper as the inverse curve; the 2013 [8] Meeusen et al. consensus (Level 5) anchors the multi-marker overreaching diagnosis. Read the practical read: the rower who climbs training load by more than about 10 percent per week will see measurable fatigue signals within 7 to 14 days; the rower who watches a 2K test dip in the first month of a new block is reading the model correctly; the rower who watches ACWR above 1.5 is in the 7 to 14 day high-risk window; the rower who plans a 7 to 14 day taper before a goal event will see the fatigue component decay faster than the fitness component.

When you want to read the model in your own logbook, the practical recipe is: track the daily rMSSD and the 7-day rolling trend against the 60- to 90-day baseline; track the ACWR (acute 7-day load divided by chronic 28-day load) and watch for values above 1.5; track the 2K moving average across the most recent two to three tests and watch for dips below the within-subject corridor (about 1 percent); track the mood, sleep, and motivation against the training load; and treat a single bad session as noise, a single bad week as a signal, and a single bad month as a verdict. The AI coach that reads the logbook as a fitness-fatigue signal, treats single-session variability as noise, and waits for the moving average to confirm or refute a real downturn is reading the model correctly.