Editor's Note
Physical therapists, athletic trainers, and movement specialists have spent careers learning to read a body in motion — the hitch in a gait, the compensation a shoulder makes for a hip, the thing a patient does not say but demonstrates on the table. AI movement-analysis and rehabilitation-planning systems are now reading those bodies first, faster, and in ways that do not require hands. What this issue presses on is not accuracy. It is what was built inside the person who used to be the reader, and whether that building was ever separable from the reading.
In This Issue
Load Bearing
A reported institutional account of AI pre-screening in university athletic training programs. The essay enters through documentation — pre-participation screening records, trainer caseload data, and program policy language — and works outward to the question of what a first appointment is for when its conclusion has already been drafted. The affirmative case for the system's efficiency is present and not dismissed.
Gait
A first-person account of a practitioner whose central professional skill has been made adjacent rather than necessary. The essay moves through specific clinical encounters — patients whose gaits she read correctly before the system did, patients whose gaits she read wrong and the system did not — and arrives at a question about professional identity that the essay refuses to answer cleanly.
Range of Motion
A longitudinal reported personal account organized around the gap between what an AI rehabilitation platform optimizes for and what post-stroke recovery actually requires. The essay is structured around the therapist's self-assigned project of documenting the cases the platform handles poorly — and what that documentation reveals about the platform's training data and about what her own expertise was built from.
Faster Better
A reported account of AI triage in a resource-constrained clinical setting where the alternative to algorithmic intake was not a skilled human practitioner — it was a waiting list that outlasted the injury. The essay makes the affirmative case directly and holds it. It does not end in ambivalence.
Load Bearing
A first evaluation used to begin in ignorance. The athletic trainer did not know the incoming freshman's body, and the appointment existed to end that condition — slowly, by hand, in an order that varied from trainer to trainer and was itself a kind of signature. At three Division I programs in the Pacific Northwest, the ignorance now ends before the appointment begins.
The policy language at the three schools is similar enough to suggest a shared vendor template. Incoming athletes complete a baseline movement assessment during orientation week, before the pre-participation physical examination and before any member of the sports medicine staff has conducted a hands-on evaluation. The assessment is a camera-based battery: overhead squat, hurdle step, single-leg landing, lateral bound, a trunk-stability sequence. Software extracts joint angles, measures left-right asymmetries, and sorts the athlete into a risk tier. The report — valgus collapse on left knee descent, restricted thoracic rotation, contralateral hip drop in single-leg stance — lands in the trainer's queue, often the same afternoon, formatted for the electronic medical record and time-stamped for the compliance file.
A hand that searches is a different instrument from a hand that confirms.
None of this is trivial. August at a Division I program is a compression event. Rosters arrive in waves, the pre-participation window is fixed by conference rule, and athletic training staffs operate at ratios the profession has complained about for a generation. Trainers at all three programs could name an athlete the screen caught early — an asymmetry that would have taken weeks of practice observation to surface, flagged on day two. The documentation is uniform, legible, and defensible, a word that appears in the risk-management memo at two of the three programs. The affirmative case is real and the programs make it without embarrassment.
What has changed is the first appointment, and what has changed about it is not its length but its direction. The trainer now opens the report before opening the door. The intake interview, which once wandered — sport history, injury history, the sofa the athlete slept on in July — is organized around the flags. Athletes read their own reports through the portal before they meet their trainer. They arrive fluent. A freshman middle blocker describes her dynamic valgus before she has taken off her shoes. The old appointment extracted information the athlete did not know she was giving: the way she sat, the shoulder she guarded while describing a different injury, the pause before the word fine. The new appointment receives information the athlete has already organized, in vocabulary she learned from a PDF, about a body she has been told how to narrate.
Then there are the clean reports. At all three programs, policy permits an abbreviated evaluation for athletes in the lowest risk tier, a provision written in the language of staff allocation and hard to argue with on those terms. But the screen reads a particular body: warm, rested, performing at midday, aware that a camera is grading it. A compensation that appears only under fatigue does not appear in a four-minute battery. A trainer at one program described a distance runner whose screen was unremarkable and whose hip mechanics deteriorated, visibly, in the back half of a tempo run — the kind of thing a first appointment used to be positioned to go looking for, precisely because the trainer had nothing else to go on and so had to look at everything. The clean report does not merely fail to flag such an athlete. It licenses the abbreviation of the one encounter that might have.
The deeper problem is structural, and the trainers themselves name it most precisely. A hand that searches is a different instrument from a hand that confirms. The search moves broadly, settles where resistance appears, doubles back. The confirmation goes where the report said to go, finds what the report said was there, and stops. The liability framework hardens this asymmetry. Policy at all three programs requires that clinical findings inconsistent with screening output be documented with supporting rationale; at one program, the rationale requires a supervisor's co-signature. Agreement requires a checkbox. Disagreement requires a paragraph, and the knowledge that the paragraph becomes discoverable if the athlete is later injured in a way the screen predicted. Over a season of caseload pressure, the path of least documentation is agreement, and the programs' own policy language has made it so.
Senior staff can still push back. They were formed before the system arrived, across thousands of unseeded first evaluations, which means thousands of occasions to search, to guess, to palpate the wrong structure and learn why it was wrong. The graduate assistants and early-career hires at these programs have, in most cases, never conducted a first evaluation that did not begin with a report. Their hands are being trained as confirming instruments because confirming is the only task the workflow assigns them. What the senior trainers know, they know because they were wrong, over and over, with their hands on bodies that trusted them anyway, and no report, however accurate, can hand that down.
The programs are aware of some of this. One is piloting a protocol in which trainers conduct the hands-on evaluation before reading the screen output, then reconcile the two — a blinding procedure, borrowed from research design, applied to a training room. It takes longer. That is, depending on where you sit, either its flaw or its entire point.
The vendor contract at one program comes up for renewal in the spring, and the abbreviated-evaluation policy will be reviewed alongside it, together with the tier thresholds. In ten years, the staff who can disagree with the report will have retired or moved into administration, and the question none of the three programs has put on any agenda is smaller and harder than a liability framework: who, by then, will still know how to be wrong?
Second Opinion
Last week a kid told me his knee was fine, and his quad told me otherwise about four seconds after I got my thumbs on it, which is the whole job in one sentence. Everyone writing policy about screening tiers should have to spend an August doing intake with sixty bodies and a schedule that was impossible before the software and is still impossible after it, because the software did not hire anyone. The report is not my enemy, the checkbox is, and the day agreeing stops being faster than disagreeing is the day I'll believe anyone upstairs actually read this.