The undiagnosed months Watch

Nobody looked at the knee for six months. A few weeks after the tear life felt normal again; what lingered was an inability to push off or hop on the leg, and less than full range of motion. Would the data have said "this is serious"? The test: the injured window (Feb 18 – Aug 14, 2023) against the same dates in healthy 2022, with alert rules that use only pre-injury data as their reference.

• The clinicians didn't catch it by hand either. Both the 8/15 PT exam and the 8/29 orthopaedic exam recorded a negative Lachman and anterior drawer; they suspected a PCL or posterolateral-corner injury, and the ACL tear was found on the 9/5 MRI ("chronic appearing", with a ganglion), alongside a displaced bucket-handle meniscus tear the surgeon believed had been displaced since February. That is the fair benchmark for what passive data could have caught.
• The tear was unmistakable in the data for about six weeks: six of eight gait metrics sat beyond their pre-injury 5th/95th percentile for three or more consecutive weeks, starting day 16, and the composite rule fired at day 15. The same rules on the same metrics never fired in the healthy 2022 window.
• Then it faded. By roughly May every alert had switched off. Through weeks 9–25 step length, walking speed, both stair speeds and flights climbed all stayed measurably below baseline (Cliff's δ 0.25–0.40 vs the immediate pre-injury months), but inside the 5–95% band, so a threshold alert would have called it "injury, then recovery", not "torn ACL with a bucket-handle meniscus".
• The push-off deficit you felt has counterparts: stair ascent, step length, fast-end walking speed and flights were all depressed in weeks 9–25. Hop and range of motion have no passive proxy at all. "Walking felt normal" is partly contradicted by the data: double support stayed elevated (δ +0.26 vs pre-injury, +0.73 vs 2022) and mean speed lower. Asymmetry, the obvious limp signal, did normalise: 44% in the injury week, 0.1% by weeks 3–8.
• Two one-off tells: the six-minute-walk estimate dipped below its 500 m cap only twice in three years, at 444 m and 489 m in weeks 4–5 after the tear; and the unknowing prehab is plain in the workout log, 54 cycling and 47 strength sessions between May and surgery, with no running.

The prehab, unknowing Watch

From May 2023 he rode an indoor bike almost every other day and did band and bodyweight work on ride days, not knowing what he was rehabbing. The Watch logged it all. Was it training, or gentle spinning?

• The prehab was real: 128 minutes a week for 19 weeks, five times the previous year's bike-plus-strength volume and 60% of the later post-op rehab block. Rides ran at 79% of max heart rate, every one of them above 70%.
• It held load rather than building it. Ride length never changed, weekly minutes drifted down, and ride heart rate fell from 145 to 136 bpm at the same energy cost, which reads as adaptation to a fixed workload. The strength sessions (median 32 min, 119 bpm, 5.0 kcal/min against 6.0 for gym lifting) look like band and bodyweight work bolted onto ride days, which is what he remembers.
• The consequence shows in the pre-op baseline used in section 02: resting HR 64, HRV 52 ms, the best numbers of the whole 2022–23 record. Whatever the surgery did to him physiologically, it did to a fitter person than the one who tore the ligament.

What the ring and the watch recorded

Daily values from 31 days before to 60 days after, against the one-month pre-op Oura baseline (mean ± 1 SD). The ring saw the surgery clearly; the watch barely did.

• Sleep efficiency fell from 88% to 61% (42% on day 4), temperature ran +0.9 °C above baseline on day 4, REM halved. All three were back inside the band by day 14.
• Readiness was the slowest to settle: 49 days. Watch resting HR spiked to 84 on day 3 and was back by day 5; Watch HRV never left the ±1 SD band.
• The day-14 return of sleep efficiency, REM and temperature coincides exactly with the end of a two-week oxycodone course (20 × 5 mg, the first of two orders written; the second, for 30, was not taken). Opioids suppress REM and disturb sleep and thermoregulation, so the shape fits; but the first two weeks are also the acute inflammatory, immobile period, and the data cannot separate stopping the drug from healing. Aspirin 325 mg twice daily ran for 30 days and cephalexin for four.

How it compared with the tear Watch

The plan predicted the tear would hit gait harder and the surgery would hit physiology harder. Compared like-for-like on the Watch and iPhone (no Oura existed in February 2023), neither half held.

• Surgery was the larger perturbation on every comparable gait/load metric: walking speed d −2.64 vs −1.74, step length −1.50 vs −1.05, steps −1.65 vs −0.44. Four gait metrics are not comparable at all because there were only 1–2 phone-walking days in the post-op week (crutches).
• Watch physiology moved little at either event. The vivid surgery signal in 03 is Oura-only, so that half of the prediction cannot be tested symmetrically.
• The tear itself was visible within 24 hours: asymmetry went from 0–4% the prior week to 93% on day 1 (z = 18.8), walking speed from 2.65 to 1.68 mph, resting HR from 63 to 75–90 bpm on days 2–6. The last logged ski run that day was 82 minutes.

The first weeks, against the protocol

The discharge protocol set the calendar: immobilizer and partial weight-bearing to 10/27, flexion limited to 60° until 10/13, brace off allowed at the 11/6 visit with range of motion 5–95°, a PT referral for December. A family event out of state one week post-op. The Watch saw each transition about a week before the protocol date.

• Protocol dates show up, each about a week early. Stair-speed samples reappear the week after the 60° flexion cap lifted; steps jump from 3–4k to 5–6.7k and flights double in the two weeks after brace-off on 11/6. The 12/12 PT start has no counterpart at all (steps fell through December and January), which matches the note that PT lapsed after that one visit.

Sleep through the post-op months Oura

The December PT note says he was occasionally waking from knee pain. Oura recorded every night. What actually happened to sleep between the operation and the spring?

• The "waking from pain" does not show. October to December 2023 sleep was as consolidated as the recovered year: efficiency 88% both, awake time slightly lower, fewer wake bouts, fewer restless periods. December 2023 was one of the calmest months in the record. Sub-five-minute arousals are invisible to the ring, so brief pain wakes could still have happened; long ones did not.
• What stayed abnormal was autonomic, not structural. Sleeping heart rate ran 64–67 bpm against a baseline of 60 every month from October to March, and sleep HRV 34–40 against 53 ms. Both came back in June–July 2024, which is also when steps recovered from 3.4k to 8k a day; deconditioning and recovery cannot be separated here.
• The March setback is the one sleep event after the opioid fortnight. Awake time nearly doubled for two weeks, with five nights over two hours awake, while readiness rose and sleeping HR fell. A local pain signal, not a systemic one, on the same date the stair-descent speed dropped.
• Daytime load did not cost sleep: more flights climbed went with slightly better efficiency (ρ +0.21), and steps had no relationship to awake time.

Gait Watch

Six gait measures, weekly medians against the pre-injury interquartile band. The tear dented every one of them; surgery dented them again, harder.

• Double support recovered first (59 days), walking speed last (213 days), and the Jan–Mar 2024 trough that delays it is not winter (season explains 6% of it, see below) but the months PT lapsed: one visit in December, the surgeon follow-up missed, then around 3/11 a pivot produced a pop, swelling and pain going downstairs (the lapse and setback, below).
• Every metric also recovered before surgery, 23–86 days after the tear. Walking on a torn ACL normalised within three months; the reconstruction then reset the clock.
• Walking asymmetry is the one metric fully indistinguishable from baseline at 12+ months (Cliff's δ +0.03, CI −0.09 to +0.16).

• Season explains 6% of the Jan–Mar 2024 trough. Raw walking speed 2.32 mph against an expected-for-date 2.75; residual −0.43 mph (16% below expectation, δ −0.40 vs pre-injury residuals). After the 3/11 setback the residual is −0.10 and not distinguishable from zero, so the setback did not worsen walking. The trough belongs to the months PT lapsed.
• Adjusting for season barely moves the recovery clock: on a first pass walking speed re-enters the band at 59 days (an edge effect; it sat a hair under the raw threshold), then relapses for twelve weeks during the lapse and settles at 199 days against 213 raw. Step length 157 → 157, double support 59 → 59 first pass, asymmetry 108 → 101. Both stair speeds never settle by either method. The earlier "winter-padded" reading was wrong and has been removed from this page.
• For the 2023 undiagnosed months the correction cuts the other way: weeks 9–25 are spring and summer, when expectation runs above the annual level, so the persistent deficit there gets larger, not smaller (walking speed δ −0.40, step length −0.47).

Load and activity Watch

Watch-only counts (Oura excluded). The tear alone cut stair-climbing by a third; surgery took it to near zero for a month.

The lapse and the March 2024 setback

After one PT visit in December he "fell off" PT, missed the December surgeon follow-up, and was doing well on his own until, around 3/11/24, he pivoted on a bug, felt a pop, and had swelling and pain going downstairs. The 3/25 exam found no re-injury (flexion 130°, negative Lachman, tunnels fine on X-ray), renewed the PT referral, and on 4/15 the surgeon noted a mild-to-moderate effusion and set the sequence: bike, then elliptical, then running, with a single-leg hop test before any sport clearance.

• The setback is real and it is a stair-descent event, which is exactly the symptom in the note ("posterolateral pain going downstairs"). Stair descent ran 1.5–1.85 ft/s through 3/9, dropped to 1.27 on 3/10 and 0.99 on 3/11, and sat near 1.26 through April. Ascent fell too (δ −0.59). Walking speed, step length, double support, steps and flights did not move; resting HR and readiness improved. The Jan–Mar walking-speed trough was already there before the setback (and, per the seasonal analysis under Gait, is not winter); stair descent had recovered to 1.5–1.6 by January and was cut back to November levels.

The PT block and the return to running

Twelve weeks of outpatient PT from 5/2/24, one visit a week: LEFS 66 at intake with symmetrical range of motion, not cleared to run on 6/7 (failed an 8-inch step-down), cleared to jog ten minutes on soft ground on 7/12, a walk-4-minutes/run-30-seconds program on 8/16 after treadmill pain at 40 seconds, and by 9/13 "20–30 minutes of run/walk every other day, up to 7 mph". Single-leg press stayed at 0.87–0.89 left over right; the hamstring caught up by July.

• The run log is the PT plan, executed. Zero runs between the 7/12 soft-surface clearance and 8/16 (unlogged or not done). First run 8/29 at 15:58 /mi, 13 days after the walk-4/run-30s program whose blended pace works out to 15:24–15:54. September–October: 9 runs in 49 days (1.3/week, not every other day), 22–32 minutes each (matches "20–30 min"), 14.3 /mi, which implies the running share had grown from ~11% to ~20–27% of session time.
• Clinical and passive agree on the trajectory, not the mechanism. Lowest LEFS (52) sits on the slowest walking, shortest step, highest double support and fewest steps; highest LEFS (66) on the fastest. But single-leg press (L/R ratio 0.87–0.89 May–August) and hamstring strength (17.8 → 24.1 lb) moved with no matching change in stair ascent, which stayed 1.03–1.11 ft/s. Strength gains in the gym did not show in stair mechanics on the Watch.

Strength, leg by leg Strong

Lifting was logged only once rehab moved onto real machines, from December 2024, fifteen months after surgery. There is no pre-injury unilateral baseline; the right leg is the comparator. Each session, each exercise: left ÷ right estimated one-rep max, a limb symmetry index.

• By the numbers the gym could see, symmetry was already near 0.95 when logging began (day 467) and reached 1.00 in the second half of 2025. Both the quad and the hamstring side started one or two reps short on the left at the same load and converged; the hamstring, which donated the graft, has the cleaner monotone trend and finished its deficit first (June 2025); the quad's last short session was January 2026.
• The seated leg press, the one lift PT had measured, was logged identically on both legs in every session (LSI 1.00 × 11), which reads as a logging habit rather than a measurement. So the clinic's 0.87–0.89 press ratios from mid-2024 have no gym follow-up; the extension and curl carry the story instead.
• Absolute strength rose more on the left than the right (extension +22% vs +15%; curl +8% vs 0%), which is what closing a deficit looks like, and also what a detrained comparator looks like. The right leg is a control that trained too.
• Strength and gait still don't meet. Against the same fortnight's Watch and iPhone gait, the only nominal correlation is left-extension e1RM vs stair-descent speed at ρ −0.66, which is two opposite secular trends (load up, stair speed down in "Two years on"), not a strength effect. Consistent with the PT block above: strength gains in the gym never showed on the stairs.

A leg day, the morning after StrongOura

With a gym log and a ring on the same calendar, one more question is cheap: does a session on the reconstructed leg cost more the next morning than an upper-body session, and did that cost shrink over the year?

• A leg day does not register as a bigger insult than an upper-body day: readiness, sleep HRV, sleeping HR, resting HR, total sleep and efficiency show no leg-vs-upper difference, and the leg-vs-rest comparison is null too. The only reliable next-morning cost in the log belongs to upper-body sessions, which are longer.
• The cost did not shrink with time; the twenty leg sessions trend the other way (first-half median +1.0, second-half −4.0 readiness points), with no matching drift in upper-body sessions and no dose-response with leg volume. Read as a small-n curiosity, not a finding: the second half of the log is mostly light bilateral work, and session days are self-selected.

Two years on: what came back, and what didn't Watch

Pre-injury (Aug 2022–Feb 2023) against a later window. Cliff's δ with bootstrap 95% CI; filled markers survive Benjamini–Hochberg at q<0.05.

• Gait, worse (season-matched 2024–25): stair descent −13%, stair ascent −6.5%, step length −2.4%, double support +2.3%. Gait, unchanged: walking speed, asymmetry.
• Everything else, better: steps +47%, exercise minutes +63%, resting HR 70 → 63, HRV +18%. A fitter person, two years on.
• In the most recent six months step length, double support and walking speed are all back at baseline; only the stair speeds and the cardiovascular gains persist across every window.