The guide, statement by statement · How the Numbers Feel

Tested Pre-registered and run; the finding is on the site.
Via a proxy Tested through a stand-in signal (this watch has no HRV).
Testable Could be tested on this data; not a headline yet.
Partly Only a coarse, daily-summary version is visible here.
Descriptive (POTS) A descriptive read of a circulation pattern, not a crash test.
Not built yet A real claim this data could check, not done yet.
Never tried here The intervention was never done, so there is nothing in this data to measure.
Off this watch The Forerunner 245 cannot measure it.
Needs outside data Not a watch metric; needs an outside source.
Not in the export Absent from this device's data.
She flags it unreliable Wiggers herself calls the watch poor at this.
Method check A data-quality check, not a claim about the body.

Heart rate A

A1 PEM Tested

Resting heart rate rises after overexertion, and scales with how much you overdid it.

Pre-registered morning-RHR drift + dose test. · morning-RHR finding →
A2 both Tested

It can deviate the other way too: a lower resting heart rate can also follow too much movement.

Tested both directions against my personal baseline. The deviation does show up around crashes, but it is just as common on ordinary nights, so it is present without being a warning. · morning-RHR finding →
A3 PEM Tested

The lowest night-time heart rate rises on crash days.

Peri-event night-RHR alignment. · morning-RHR finding →
A4 PEM Tested

Hours of sustained elevation, not a brief spike, marks real overexertion.

Tested at the bout level, hours of sustained elevation rather than a brief spike. The direction matched, more sustained load around crashes, but the effect fell short of the bar I set in advance, so I read it as too small to call on my data, not as a miss. · walls-of-orange finding →

Heart-rate variability (HRV) B

This watch records no HRV at all, so the whole family is read sideways through the night-to-night variability of sleep stress.

B1 PEM Via a proxy

A day-over-day drop in HRV predicts a crash.

Sleep-stress spike as an HRV proxy. · HRV finding →
B2 PEM Via a proxy

In PEM, HRV declines over several days even with rest.

Rolling sleep-stress slope across the peri-crash window. · HRV finding →
B3 PEM Not built yet

A slowly rising HRV baseline coincides with improving periods.

The per-phase recovery data carries this (the HRV-proxy baseline read by phase); I have not isolated it as its own read. · the recovery in six phases →
B4 PEM Via a proxy

A sudden HRV spike is bad news, not good: the parasympathetic swing.

Sleep-stress negative outlier after overexertion. · the swing finding →
B5 — Via a proxy

HRV can rise at the onset of an acute illness (distinct from PEM).

Conditioned on illness-onset days; the one COVID window went the other way. · the COVID check →

Stress C

Garmin 'stress' is an HRV-derived arousal index, blind to the cause. The U-dip and the low-blood-volume half of stuck stress are orthostatic; see the O family below.

C1 PEM Via a proxy

More orange night stress on crash days and their lead-in.

Sleep-stress mean / high-duration peri-event. · the variability finding →
C2 PEM Testable

High total daily stress costs the next day's recharge.

Daily stress vs overnight body-battery gain and next-day felt-state. · still on my list →
C3 PEM Tested

The stress cost is non-linear: a 30 to 40 step costs far more than it looks (a stair-step).

Binned / spline stress to felt-state (not a straight line). · Stress, and how I felt →
C4 both Tested

Walls of orange: after overexertion, stress stays stuck high even at rest ('PEM or low blood volume').

Stress-not-settling test on the PEM interpretation (weak). · walls-of-orange finding →

Body battery D

D1 PEM Tested

The bare body-battery number is a weak indicator of how you feel.

Tested the bare daily body-battery number as a crash signal, and it did not hold across the record. On its own the daily figure is a weak read of how I felt, which is what her guide says of it too. · the swing finding →
D2 PEM Tested

The dynamics (overnight gain, drain rate) matter more than the level.

Overnight gain / daytime drain vs felt-state. · the swing finding →
D3 PEM Tested

Living 'at the top' (a 70 to 80% floor) coincides with fewer crashes.

Checked how my body-battery floor tracks crashes. The direction fits, crash days sit lower, but her specific 70 to 80 percent floor is above anything my battery reaches, so on my body the honest version is a personal-baseline floor rather than her number. That recast is still on my list.
D4 PEM Tested

Body battery drains steeply around a crash and leads the felt dip.

Two halves, and they part ways. The steep drain around a crash is real: the morning peak collapses right at the crash. But the leading half is not there, the drain does not run ahead of the felt dip; it arrives with the crash, not before it. · A discriminator, not a predictor →
D5 PEM Tested

Paradoxically, a high morning body battery after overdoing it precedes a crash (false energy / the swing).

Morning body battery conditioned on prior-day overexertion. · the swing finding →

Steps & activity E

E1 PEM Tested

There is a personal step threshold above which crash risk climbs, and you must respect the lag.

Lagged step-load in the days before vs crash. · exertion-lead-up finding →
E2 PEM Testable

Rising steps without rising crashes marks genuine improvement.

Step-average slope vs crash rate over the same window. · still on my list →
E3 PEM Tested

Intensity minutes may track your load better than raw steps.

Compared the axes head to head. In my recent years intensity minutes did track my load better than raw steps, as she suggests; in my earlier years it ran the other way. Neither single axis wins across the whole record, and a combined effort measure does better than either alone. · exertion-lead-up finding →

Sleep F

F1 PEM Tested

Longer-than-usual sleep is typical in a crash.

Compared sleep length on crash nights against ordinary ones, and on my body there is no real difference. The one night everyone remembers, a sleep of nearly sixteen hours inside a bad crash, was my longest ever, not my typical crash. An honest negative for me; the wider illness picture still stands.
F2 PEM Testable

Too little or too much deep sleep tracks worse days.

Deep-sleep deviation vs next-day felt-state. · still on my list →
F3 PEM Not built yet

The Garmin sleep score predicts next-day capacity.

Garmin's own score is proprietary and needs REM the Forerunner 245 does not record, so I can't reproduce it. But a home-built sleep-quality proxy (duration deviation + deep-sleep fraction + awake minutes + bedtime consistency) is constructible from what this watch does record; queued.
F4 PEM Testable

Inconsistent bedtimes cost the next day's energy.

7-day bedtime variance vs next-day felt-state. · still on my list →

The other sensors G

G1 both Testable

Faster breathing marks a stuck-sympathetic, poor-recovery state.

Respiration rate (sleep + waking) vs crash / felt-state. · still on my list →
G2 PEM Off this watch

Skin / estimated temperature shifts around a crash (mostly up).

No skin-temperature sensor on the Forerunner 245. · out of reach →
G3 both Needs outside data

Low or falling barometric pressure tracks worse days and headaches.

Not a Garmin metric; needs an external weather join. · out of reach →
G4 POTS She flags it unreliable

Blood oxygen dips on exertion, and (her interesting case) after standing.

She calls the watch's SpO2 unreliable; the interesting signal is positional. · out of reach →

Mechanism & lead/lag H

These are the decisive method questions the guide implies rather than states outright.

H1 PEM Tested

The wearable signals lead the felt crash (earlier warning than self-report).

Tested channel by channel. For almost all of them the signal marks the crash as it arrives, not before it, so the 'earlier warning' reading mostly does not hold. The one exception is fragmented sleep, which does run high in the days ahead. · A discriminator, not a predictor →
H2 PEM Via a proxy

A meaningful share of crashes are 'activity-invisible' (mental PEM the watch can't see).

Low-activity crashes with a sleep-stress signature; her own concession. · what the watch catches (beyond) →
H3 — Tested

Acute-illness crashes have a different signature than PEM sags.

Characterised one acute-illness crash from the record, the COVID window, and a physically-driven crash type does show up. The full head-to-head, an illness signature against a PEM one, I could not run: there were no clean household-illness dates to build it on. · two kinds of crash (beyond) →
H4 PEM Tested

The parasympathetic-swing signature precedes a felt dip within a day or two.

Tested the 'precedes within a day or two' timing, and it did not hold: the swing signature shows at the crash itself, not a day or two ahead of the felt dip. The state is real; the head start is not. · the swing finding →
H5 PEM Via a proxy

Each metric has its own characteristic lag versus the exertion; the lags differ.

Per-channel cross-correlation ordering (BB/stress <= RHR < HRV). · A discriminator, not a predictor →

Data-quality checks I

Not claims about the body; checks the research runs on itself.

I1 — Method check

Re-run results excluding the first weeks of a device and imputed periods.

Sensitivity re-run; conclusions stable. · the seven limits →
I2 — Method check

Mark device-change points and test for level shifts.

N/A here: one Forerunner 245 throughout four years. · the seven limits →
I3 — Method check

Confirm the overlap window: rich metrics exist only from certain dates.

Per-column coverage documented. · what this watch can't see →

The orthostatic (POTS) family O

Half the guide is about circulation, not crashes: how the body handles standing, blood volume, and heat. Wiggers treats this descriptively, and it waxes and wanes as symptoms are managed. Most of it this watch cannot see.

O1 POTS Partly

Resting heart rate can rise while you lie still, from too little movement ('you needed more blood flow').

Two legs. The rise-from-too-little-movement leg comes up empty at the episode level, the same null as the morning heart-rate finding. The while-lying-still leg is off this watch, which records no posture. A rest-day version is still queued. · out of reach →
O2 POTS Off this watch

Standing up makes the heart race (the NASA-lean rise), a read of the autonomic nervous system.

The watch records no posture, so it cannot tell a standing rise from any other. · out of reach →
O3 POTS Tested

The stress U-dip: a dip in stress while body battery rises, which she treats with electrolytes.

The U-dip count; present in my data and time-varying. · the U-dip finding →
O4 both Via a proxy

Stress stays high on standing or after eating ('PEM or low blood volume').

Stress-not-settling test; weak, and it can't separate the POTS cause from the PEM one. · walls-of-orange finding →
O5 POTS Never tried here

Managing POTS with compression, water, and salt significantly lowers daytime stress scores.

Her claim, an intervention effect. I never did the management (no compression, salt, or extra fluids), so there is nothing in my passive record to test; it would need a prospective trial. · out of reach →
O6 both Needs outside data

Low or dropping barometric pressure leaves you dizzy and exhausted (below ~980 mBar for her).

Not a Garmin metric; needs an outside weather join. · out of reach →
O7 POTS Off this watch

Blood oxygen drops after standing or moving, and rises again lying down.

Positional, and she calls the watch's blood-oxygen reading unreliable anyway. · out of reach →
O8 POTS Method check

POTS raises the watch's data-gap and high-heart-rate estimation errors.

A racing heart can make the watch drop or estimate readings. · out of reach →

PEM statements not built yet PG

Load-and-recovery statements in her guide that the register named but had not yet turned into a test. One has since been built: the awake-minutes readout (PG5) was run and turned out to be the standout of the whole set, so it now reads 'tested'. It stays here beside its siblings for now.

PG1 PEM Not built yet

A long time to fall asleep after overexertion warns of PEM the next day (a rare predictive claim).

Sleep-onset latency; timestamps exist, constructable.
PG2 PEM Not built yet

The heart-rate x HRV night-typology (high-HR-fluctuating, initial-high-then-falling, both-fluctuating).

A five-state night typology; only the severe-PEM and swing states map today.
PG3 PEM Not built yet

Night heart rate is higher and more variable when disrupted, not just higher.

Two legs. The higher-night-heart-rate leg lines up with the resting-heart-rate finding, which came up null. The more-variable leg needs a per-night variability measure I have not built yet.
PG4 PEM Not built yet

A second personal dose anchor: two HRV dips past ~2,000 steps; HRV at 40 = not going well.

A steps/HRV anchor distinct from E1's 1600/3000.
PG5 PEM Tested

Awake / restless minutes per night as a status readout.

The awake-minutes side is the quiet standout. Crash days carry more awake time inside the night than ordinary days. And, unusually for my data, it shows up early: the nights start to fray up to six days before a felt crash, then peak on the crash day. Among every channel I have checked for a run-up like this, it is the only one that behaves like a real lead-up. A pre-registered test, pooled across the whole record, put numbers on it: across the four days before a crash the fraying fires on about 73 percent of crash weeks against roughly half of ordinary ones, a real gap that does not quite clear the chance bar (p=0.12); narrowed to the last two nights, it does clear it (+27 points, p=0.04). So the closer to the crash, the sharper the fraying. Still a modest signal, Tier C, the closest thing to a forecast I have rather than a strong one. The restless-minutes side is off this watch, which does not record it. · the nights that fray early →
PG6 PEM Not built yet

In a flare-up the heart rate runs high even lying down ('how upset my body is').

Resting HR / intensity-minutes while recumbent.

Distilled from the research register: wiggers_testable_hypotheses.md ↗. The mechanism labels and the "can I measure it" status are framing, not verdicts; the verdicts live in the crash tests, and the signals walked one at a time are in the field guide.

The guide, statement by statement.