Mobility Atlas research: exoskeleton-guide
Data and method for the article at mobilitatlas.com/blog/exoskeleton-guide/
License: CC BY 4.0 (credit "Mobility Atlas")
Prepared: 18 October 2026

WHAT WAS READ
Every FDA MAUDE medical device report with text filed under product code PHL (Powered Exoskeleton; 21 CFR 890.3480), IQZ (Hand, External Limb Component, Powered), PAE (Upper Extremity Prosthesis With Multiple Simultaneous Degrees Of Freedom And Controlled Via Cutaneous Electromyography; 21 CFR 890.3450), IQW (Hook, External Limb Component, Powered), IRE (Joint, Elbow, External Limb Component, Powered) or IMK (Wheelchair, Stair Climbing; 21 CFR 890.3890), received at any date up to the latest openFDA update read on 18 October 2026; plus every report already read for earlier Mobility Atlas studies under IOR (manual wheelchairs, 2019 to 2026) or ITI (powered wheelchairs, 2024 to 2026) whose text names an all-terrain, off-road, beach, tracked or trail chair. Pool: 161 reports, all coded with one codebook: 56 about powered exoskeletons, 11 about powered arms and hands, 81 about stair-climbing wheelchairs, 1 about an all-terrain chair, 12 about other devices. This data folder is shared by the Mobility Atlas exoskeleton and stair-climbing wheelchair guides.
All FDA recall records for these product codes, downloaded through openFDA on 18 October 2026.

HOW IT WAS CODED
1. A codebook (codebook.txt) was written before any report was coded. Its SHA-256 fingerprint, recorded at 2026-10-18T14:30:29Z, is 2f3c3bce072c9ae8dbce04c9b00f22a3ad9f0fc1462dee67f46f2f5daa53da72. Any later change to the codebook would change this fingerprint.
2. The reports were split into chunks. Two coders (AI language models working separately, each in a private scratch folder, each without access to the other's answers, told to read every report and not to use keyword scripts) coded every report in every chunk against the codebook.
3. Agreement between the two coders: device 99.4% (kappa 0.99); problem 88.8% (kappa 0.859); injury 100.0% (kappa 1.0); fall 98.8% (kappa 0.974); setting 90.1% (kappa 0.843); what was found 97.5% (kappa 0.954).
4. Every report where the coders differed was decided by the lead analyst under the written rules below (40 reports marked adjudicated). A consistency pass then re-applied the rules across the whole table and changed codes in 18 more reports (marked consistency).
5. Counts in the article were produced by script from the final table, not by hand. Some sub-counts (parts, places or words named in reports) are keyword counts over the report text; the exact word patterns are listed under KEYWORD COUNTS. Different wording would give somewhat different counts.

ADJUDICATION RULES
S1: DEVICE (all-terrain vs ordinary chair): T only for a chair built or fitted for off-road, beach or all-terrain use (an off-road attachment such as a front freewheel counts) or one that runs on tracks; an ordinary wheelchair ridden on gravel or rough ground, or one whose rubber tyre or tread is called a track, is O.
S2: PROBLEM (F vs U for tips and falls): F when the reporter alleges the device did something on its own beyond the tip itself (locked up, kept rotating, rolled or moved unprompted, lurched, stopped, shut down, transitioned for no reason, made an unusual noise) or a part broke, came loose or detached and the user fell or the chair tipped with the user, because F comes before S in the codebook list; U when the report only says the chair tipped or fell (on stairs, a curb or while backing up) and names no device behaviour, or puts the event down to the user, the assistant, an object or the surface. The maker's log verdict belongs in FINDING, not PROBLEM.
S3: PROBLEM (automatic responses): an automatic device response (auto-transition, safety stop) that the report itself ties to an outside trigger (wheels slipping on a loose rug or a wet surface, hitting an object) and presents as protective or expected is U; C, or F with a fall, only when the reporter presents the response as unexplained or wrong.
S4: PROBLEM (damage from the event): damage to the device caused by an outside object or impact during the event (a cable snagging and breaking an armrest, a handle catching a railing) is not a structure fault alleged; the event is coded U.
S5: PROBLEM (U vs X): U when the report places a fall or an injury in an episode of using or training with the device (during a session, a sit-to-stand, a walk, a fall from the chair) and alleges no device fault, even when it gives no further circumstances; X when the harm is only noticed after use with nothing said to have happened during it, has no stated link to use, is caused by an outside party or event (a vehicle, a house fire the maker judges unrelated), or is an illness the report ties to a medical condition.
S6: PROBLEM (K): K when the report attributes the harm to how the device fits or sits on the body (sizing, length adjustment, misalignment, a strap, belt, footplate, footrest, bracket or socket pressing on the body or letting a limb slip through), with no other fault, including a fracture the report places at a device contact point; harm from a sudden spasm or an outside cause is not K.
S7: PROBLEM (C for felt or seen movement): an unexpected movement or behaviour of the device that the user or staff report (a buckle, jerking, shuddering, hesitation, getting stuck part-way through sit-to-stand) is C, even when the maker later finds the device worked as intended.
S8: PROBLEM (implant revisions, device O): a revision done to exchange a named component (a poly exchange) means that component was at issue, so with no reason given it is Q; X only when the stated reason lies outside the device (a fracture of the bone around it).
S9: FALL: y when the user fell, was thrown or went down, or the chair tipped over with the user in it; n when the chair went down the stairs without the user, who stayed holding the railing.
S10: SETTING: H when the text names a place outside healthcare (home, yard, house, street, store, bar, workplace, vacation, car) or describes the moment of the event in the user's own everyday use of a personal device outside any session (driving, sitting in, wearing or charging it, solo stair use, help from family or friends, partying with friends); C when it happens in a clinic, hospital or rehab centre, or in a therapy, training, trial, test or study session run by clinicians or the maker's trainers, unless the text puts that session in the home; U when neither is stated (a bare injury, fall or malfunction report with no episode, a revision, a session of unstated kind, a demonstration with no place). Revision surgery is care after the event, not its setting.
S11: FINDING: D when the maker's, service or clinic inspection confirms a fault in the reported device or a part is replaced as faulty; N when the maker examined the device or its logs and found no fault, or concludes there was no malfunction, or names an outside cause (another device such as a faulty wheelchair power module, misuse, an obstacle), even if the reported damage itself is confirmed and even without a full evaluation; X when nothing was evaluated and no conclusion is drawn, when evaluation is pending, or when the review says it cannot confirm or rule out a fault or that the cause is unknown.

KEYWORD COUNTS (Python regular expressions, matched against the upper-case report text without its header line)
KW = {'fracture': r'FRACTUR', 'skin': r'SKIN|ABRASION|BLISTER|PRESSURE (SORE|ULCER|INJUR)|BRUIS|REDNESS', 'therapist': r'THERAPIST|CLINICIAN|TRAINER', 'stairs': r'STAIR', 'battery': r'BATTER', 'curb': r'CURB', 'balance': r'BALANCE', 'tip': r'TIP(PED|S)? (OVER|BACK|FORWARD)|TIPPED'}

QUALITY NOTES (disclosed in full)
- The article went through fact-check passes by checkers who had not written it (a full pass with a reader review, then a control pass over the changes), each recomputing numbers from the files here and reading quotes against their sources; every error found was corrected before publication.

AMOUNTS
Every report text was searched for money figures; none gives one. Prices in the article are prices shown on makers', dealers' and retailers' pages and in the Mobility Atlas catalog, listed in exoskeleton-guide-documents.csv and the article sources.

LIMITS
- Reports are not rates. Nobody publishes how many devices of each model are in use.
- A report is an allegation or a description; the FDA says the submission of a report "itself does not necessarily demonstrate that the device caused or contributed to the adverse outcome or event."
- One maker filed most reports in each device group (Lifeward/ReWalk for exoskeletons, Independence Technology and Mobius Mobility for the iBOT), which reflects filing, not safety.
- Most exoskeleton reports come from clinics and studies, where events are watched and reported more closely than at home.
- Only one report concerns an all-terrain chair, and only already-read wheelchair reports were searched for them: the data say almost nothing about all-terrain chairs.
- The same event is sometimes filed twice; we count reports, not events.

FILES
codebook.txt        the codebook as fingerprinted
coded_reports.csv   one row per coded report: both coders' answers, the final code, and how a disagreement was resolved (resolution, rule_applied)
report_pool.csv     the reports in the pool with product code, brand, date received and event type
The full report texts are public: search MAUDE or openFDA by report key.
exoskeleton-guide-documents.csv  the documents and pages read, with links