“Why has my recovery worsened?”

A sample, run on the app’s invented example record. No person’s data is in it; where it says “you”, it means that example.

Worked on example data: every reading behind these weights is one the app supplied, not yours. The arithmetic is real; what it says about you is nothing yet.

Viewing it sends nothing anywhere: the sample is framed by the labelled stand-in (no model is called), and its literature searches and the analysts’ review are not run.

Needs youRecovery trend · 8 steps taken by itself · run Oct 12, 2026

"No real change — recovery is moving within normal day-to-day variation" leads at 0.31 of the evidence weight, left open: two records disagree about resting heart rate and sleep.

Four explanations remain plausible; what is on file cannot tell "Shorter sleep is holding recovery down", "An energy deficit is limiting recovery" and "Low-grade inflammation or illness is weighing on recovery" apart. The step that would reduce the uncertainty most is "Keep recording HRV, resting heart rate and sleep for 14 days" (0.39 bits).

How it read the question: Recovery is read as heart-rate variability and resting heart rate — the two recovery markers a wearable records. How recovered you feel is not on file as such; the journal's mood score is the nearest thing, and it is a different quantity.

The explanations in competition

A share of the evidence weight among the explanations listed — not a probability. The likelihoods behind it are the pack's declared expectations, not calibrated against outcomes. The tick on each bar is where it started; they all started level.

0.31
No real change — recovery is moving within normal day-to-day variationLeft open
Two records disagree on something it predicts strongly (resting heart rate, sleep); left open rather than averaged.
What bears on it (3)
For it
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Held — two records disagree
Two records of your resting heart rate disagree over 2026-06-15 → 2026-07-18: the trial "Zone 2 base → resting HR" logged -3.23 bpm between its windows; the wearable recorded -1.18 bpm over the same days. The smallest meaningful change is 1.5 bpm, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.unresolvedexample dataTwo sources disagree; the investigation leaves it open rather than averaging.
Two records of your sleep disagree over 2026-06-15 → 2026-07-11: the trial "Magnesium glycinate → sleep" logged +0.86 h between its windows; the wearable recorded -0.03 h over the same days. The smallest meaningful change is 0.3 h, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.unresolvedexample dataTwo sources disagree; the investigation leaves it open rather than averaging.
0.16
Shorter sleep is holding recovery downLeft open
Two records disagree on something it predicts strongly (resting heart rate, sleep); left open rather than averaged.
What bears on it (4)
For it
HRV rises and falls with sleep of the night before (ρ 0.76, permutation p 0.0005, 28 days). A co-movement, not a cause. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Against it
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Held — two records disagree
Two records of your resting heart rate disagree over 2026-06-15 → 2026-07-18: the trial "Zone 2 base → resting HR" logged -3.23 bpm between its windows; the wearable recorded -1.18 bpm over the same days. The smallest meaningful change is 1.5 bpm, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.unresolvedexample dataTwo sources disagree; the investigation leaves it open rather than averaging.
Two records of your sleep disagree over 2026-06-15 → 2026-07-11: the trial "Magnesium glycinate → sleep" logged +0.86 h between its windows; the wearable recorded -0.03 h over the same days. The smallest meaningful change is 0.3 h, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.unresolvedexample dataTwo sources disagree; the investigation leaves it open rather than averaging.
0.16
An energy deficit is limiting recoveryIn play
The evidence so far has not moved it far from where it started.
What bears on it (2)
For it
Weight -0.64 kg, the later 14 days against the earlier 14 (95% CI -0.78 to -0.49) — beyond the 0.4 kg smallest meaningful change, CI clear of zero. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Against it
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
0.16
Low-grade inflammation or illness is weighing on recoveryIn play
The evidence so far has not moved it far from where it started.
What bears on it (3)
For it
hs-CRP 3.4 mg/L on 2026-07-11 — above its range (under 3). One draw. Its within-person variation is 42.2%, so a repeat could land on the other side of the line. These are example readings the app supplied — the arithmetic is real and the readings are not yours.measuredexample datacounted weak
Against it
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Held — two records disagree
Two records of your resting heart rate disagree over 2026-06-15 → 2026-07-18: the trial "Zone 2 base → resting HR" logged -3.23 bpm between its windows; the wearable recorded -1.18 bpm over the same days. The smallest meaningful change is 1.5 bpm, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.unresolvedexample dataTwo sources disagree; the investigation leaves it open rather than averaging.
0.08
Training and dieting together are more than recovery can absorbWeakened
Its share fell from 0.15 to 0.08 — on readings that are not yours, so it cannot be ruled out. · proposed by the stand-in model (no model was called)
What bears on it (3)
For it
Weight -0.64 kg, the later 14 days against the earlier 14 (95% CI -0.78 to -0.49) — beyond the 0.4 kg smallest meaningful change, CI clear of zero. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Against it
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Weekly loaded volume 41.8 t → 42.6 t → 43.4 t → 44.2 t over 4 weeks (+5.7%), with no lighter week. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
0.02
Training load has outpaced recoveryWeakened
Its share fell from 0.15 to 0.02 — on readings that are not yours, so it cannot be ruled out.
What bears on it (4)
Against it
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Weekly loaded volume 41.8 t → 42.6 t → 43.4 t → 44.2 t over 4 weeks (+5.7%), with no lighter week. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
HRV -0.33 ms on the 16 mornings after training against the 12 others (95% CI -3.23 to 2.81) — inside the 3 ms smallest meaningful change. A contrast, not a cause. These are example readings the app supplied — the arithmetic is real and the readings are not yours.observedexample datacounted weak
Held — two records disagree
Two records of your resting heart rate disagree over 2026-06-15 → 2026-07-18: the trial "Zone 2 base → resting HR" logged -3.23 bpm between its windows; the wearable recorded -1.18 bpm over the same days. The smallest meaningful change is 1.5 bpm, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.unresolvedexample dataTwo sources disagree; the investigation leaves it open rather than averaging.
0.10
Something not on this listIn play
The evidence so far has not moved it far from where it started. · structural

What it checked by itself

Read-only work on the record — the only kind it does without anyone.

HRV the morning after training→ same
HRV -0.33 ms on the 16 mornings after training against the 12 others (95% CI -3.23 to 2.81) — inside the 3 ms smallest meaningful change. A contrast, not a cause. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Re-analysisobservedexample data
HRV and the night before→ positive
HRV rises and falls with sleep of the night before (ρ 0.76, permutation p 0.0005, 28 days). A co-movement, not a cause. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Re-analysisobservedexample data
hs-CRP→ high
hs-CRP 3.4 mg/L on 2026-07-11 — above its range (under 3). One draw. Its within-person variation is 42.2%, so a repeat could land on the other side of the line. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Existing recordmeasuredexample data
Weight trend→ fell
Weight -0.64 kg, the later 14 days against the earlier 14 (95% CI -0.78 to -0.49) — beyond the 0.4 kg smallest meaningful change, CI clear of zero. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Re-analysisobservedexample data
Weekly training volume→ steady
Weekly loaded volume 41.8 t → 42.6 t → 43.4 t → 44.2 t over 4 weeks (+5.7%), with no lighter week. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Existing recordobservedexample data
Sleep trend→ flat
Sleep -0.03 h, the later 14 days against the earlier 14 (95% CI -0.31 to 0.26) — inside the 0.3 h smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Re-analysisobservedexample data
Held: Two records of your sleep disagree over 2026-06-15 → 2026-07-11: the trial "Magnesium glycinate → sleep" logged +0.86 h between its windows; the wearable recorded -0.03 h over the same days. The smallest meaningful change is 0.3 h, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.
Resting heart rate trend→ flat
Resting heart rate -1.14 bpm, the later 14 days against the earlier 14 (95% CI -1.57 to -0.71) — inside the 1.5 bpm smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Re-analysisobservedexample data
Held: Two records of your resting heart rate disagree over 2026-06-15 → 2026-07-18: the trial "Zone 2 base → resting HR" logged -3.23 bpm between its windows; the wearable recorded -1.18 bpm over the same days. The smallest meaningful change is 1.5 bpm, so one says it moved and the other that it did not. Neither is used, and they are not averaged. Both records are examples the app supplied.
HRV trend→ flat
HRV +2.43 ms, the later 14 days against the earlier 14 (95% CI -0.36 to 5.5) — inside the 3 ms smallest meaningful change. These are example readings the app supplied — the arithmetic is real and the readings are not yours.
Re-analysisobservedexample data

What is still uncertain

Two records of one quantity disagree — both are held, neither is averaged away

resting heart rate: the trial "Zone 2 base → resting HR" logged -3.23 bpm between its windows; the wearable recorded -1.18 bpm over the same days.
sleep: the trial "Magnesium glycinate → sleep" logged +0.86 h between its windows; the wearable recorded -0.03 h over the same days.

The newest reading behind the evidence is 92 days old — the record speaks for 2026-07-12, not for now.

Where it stopped, and who it needs

It never places an order, files bench work or starts a trial. On a real record, each of these waits for the person.

Keep recording HRV, resting heart rate and sleep for 14 daysRequires you
Fresh days test whether the change you notice shows in the markers now. Import them from Apple Health or a CSV as they arrive; the investigation wakes on them.
Repeat hs-CRP, with ferritin, on a well dayRequires you, and a lab-order consent you issue
One draw of a marker that varies this much from week to week cannot say much alone. A repeat taken with no infection or hard training in the two weeks before tells a lasting inflammatory load from a passing one.
- Mock Labs (sandbox) — Inflammation + iron: Mock Labs (sandbox) carries hs-CRP and Ferritin ($29), but it is a sandbox: its values are generated, not measured, and would not count here.
- A clinician you share with: No clinician has accepted your sharing yet — Sharing is where you invite one.

Proposed, and unable to run here

Measure morning and evening salivary cortisol on the bench · stand-in modelno declared check reads its result, so it cannot move a weight $50, a result in 2 days, asks a little of you (feasibility 0.42) a needle, a sample, or something taken, some risk if it goes wrong, cannot be undone (safety 0.25) Cannot run here. No calibrated assay for cortisol on this bench. Its plate readers measure absorbance and fluorescence; turning either into a cortisol level needs a standard curve and a validated kit, and none is on file. The sandbox is not a substitute: what it returns is generated.

Refused at the door

The stand-in model (no model was called) proposed these; the engine refused them before they could carry any weight.

Low testosterone is slowing recovery — It predicts on "testosterone-level", a check this pack does not declare — nothing on file could test it as stated.

What this cannot establish

Cause. A trend or a co-movement is a difference, not a because; only a trial's own windows can say that, and then only for that trial.
A diagnosis. Overtraining, an illness or a deficiency is a clinician's call.
How you feel. The markers here are what a wearable records; a change you feel and they do not show is still real, and it is not on file.