# Wheelchairs & Mobility Scooters: research notes

Date: 2026-10-25. Scope: manual wheelchairs (self-propelled and attendant-propelled/transport chairs), power (electric) wheelchairs, and mobility scooters (3-wheel, 4-wheel, folding/travel, heavy-duty). One register, three mechanism families, because the taxonomy leaf is a single label and the splits divide it, not separate registers. Excluded, and left to sibling leaves in `supplies > personal`: walkers, rollators, canes, crutches (`canes` — "Canes, Walking Sticks, Walkers & Crutches"), stair lifts (`stair-lifts`), shower benches/transfer seats (`shower-benches`), medical alert pendants (`medical-alarms`), orthopedic supports (`support-braces`), incontinence aids (`bed-pans`). Also out of scope: wheelchair cushions, ramps, and replacement batteries sold as standalone accessories — this register covers the chair/scooter unit itself, which is what the existing add-on schema (`wheelchair.jsonc`) targets.

Quote labels: **verbatim** = I read the sentence myself in text I fetched (curl + HTML-strip, or pdftotext on a downloaded PDF) and can point at the line number in my scratchpad copy. **fetched excerpt** = the WebFetch tool returned a model's summary of the page, not raw text I read myself. **search excerpt** = from a search-engine summary only (none used here — WebSearch's session budget was exhausted before I ran a single query for this category; see "Could not reach"). A search-excerpt-only number does not go in category.json; in this run I did not use any non-fetched numbers.

## Seed

Seed run output (verbatim from `seed.py --category wheelchairs`):

- Example used: car-tyres `treadwear_grade` — a UTQG-style facet, actually olive-oil `harvest_date` is what the script returned this run: "a facet that experts rank first, that almost no listing carries, and that is expensive to backfill... list it, mark it researched, and limit backfill to the top of the catalog."
- Persona: "A ship's cook, who thinks in shelf life, storage and what works without the ideal setup."

1. The wheelchairs equivalent of harvest date: **drive wheel position** on power wheelchairs (front/mid/rear-wheel drive). It is the single fact that most changes how a power chair actually handles, clinicians and rehab engineers treat it as a primary spec (MSKTC's own factsheet gives it a figure and a paragraph each), and it is essentially never on a general retail listing — it lives on brand spec sheets and in rehab literature. Survived the data check: kept as the `drive_wheel_position` facet, source `researched`, backfill limited by how many brand spec pages the pipeline can reach.
2. What the ship's cook notices first: not speed or battery range, but **whether it fits through the actual doorways and hallways at home and whether someone can lift it into a car**. A cook without the ideal galley thinks in constraints, not features. That is exactly the buyer question the ADA's own 32-inch doorway clear-width standard and the existing schema's `overall_width_raw` / `product_weight` fields are built for. Survived: kept as the `overall_width` and `transport_weight` facets, and the `doorway_clearance_margin` derived metric.

Nothing from the seed was dropped outright; both lines fed directly into facets already justified by independent sources below.

## Sources and what each supports

### Academic / clinical: MSKTC (Model Systems Knowledge Translation Center, funded by NIDILRR)

Three factsheets, part of a peer-reviewed-adjacent clinical series for spinal cord injury patients and their clinicians. Fetched as raw HTML and stripped (not just WebFetch summaries); line numbers below are from my scratchpad copies.

- `msktc_getting` — https://msktc.org/sci/factsheets/wheelchairs/Getting-The-Right-Wheelchair — **verbatim** (scratchpad `msktc_getting.txt`, confirmed via curl of the same URL at lines 139, 141, 143): "Age affects endurance and strength and may be a deciding factor between a manual or power wheelchair." / "Height and weight will affect what size wheelchair you will need." / "Functional abilities: Your level and completeness of spinal cord injury will affect how much function you have in your arms and hands and whether you will need a power or manual wheelchair." Supports: the manual-vs-power decision is about strength/function, not just preference; weight/height drive sizing.
- `msktc_manual` — https://msktc.org/sci/factsheets/wheelchairs/The-Manual-Wheelchair — **verbatim** (scratchpad `msktc_The-Manual-Wheelchair.txt` line 126): "Wheelchairs are generally classified by Medicare (which sets the industry standards) as lightweight (less than 34 lbs.) and ultralight (less than 30 lbs.). (These weights do not include footrests or armrests.)" Line 127: ultralight chairs are pricier, highly adjustable, harder to get insurance to cover; "Clinical practice guidelines recommend the lightest chair possible. Titanium and aluminum chairs can weigh under 20 lbs." Line 129: "Frames may be folding or rigid. Rigid frames tend to perform better when maneuvering. Folding frames are often easier to transport..." Line 143: tire types (pneumatic/foam-filled/solid) and their maintenance/comfort trade-offs.
- `msktc_power` — https://msktc.org/sci/factsheets/wheelchairs/The-Power-Wheelchair — **verbatim** (scratchpad `msktc_The-Power-Wheelchair.txt` lines 129-146): the three drive-wheel positions, quoted in full in "Branches" below, plus: "These controls are usually programmable and can be operated using various kinds of joysticks or switches (such as sip-and-puff)." No battery-chemistry content in this factsheet — I looked, it isn't there.

These are the only sources I found this run that are written for people who actually use the equipment rather than sell it, which is why they carry most of the mechanism and the drive-wheel trade-off. The 34 lb / 30 lb Medicare weight-class figures are single-sourced here; I tried four different routes to a second source (CMS's own LCD page, two DME-MAC policy pages, an HCPCS code lookup site) and all four were blocked or 404'd (see "Could not reach"). I kept the figures but demoted them to a judgement (medium confidence) rather than a `range` on a facet, per the "two sources for a number in the JSON" rule — the lightweight/ultralight thresholds appear only in prose (judgements), not as a `range.min/max` anywhere in `category.json`.

### Primer / community reference: Wikipedia

- `wikipedia_wheelchair` — https://en.wikipedia.org/wiki/Wheelchair — **verbatim** (scratchpad `wheelchair.txt`, fetched via curl, not WebFetch's summary). Line 317: "A bariatric wheelchair is one designed to support larger weights; most standard chairs are designed to support no more than 250 lb (113 kg) on average." Line 264: "Where an ultra-lightweight manual chair may weigh under 10 kg, the largest outdoor power-chairs may weigh 200 kg or more." Line 262: the indoor / indoor-outdoor / outdoor power-chair capability classes, quoted in "Branches." Line 346: "ANSI-RESNA WC19... is a voluntary standard for wheelchairs designed for use when traveling facing forward in a motor vehicle. ISO 7176/19 is an international transit wheelchair standard that specifies similar design and performance requirements."
- `wikipedia_scooter` — https://en.wikipedia.org/wiki/Mobility_scooter — **verbatim** (scratchpad `scooter.txt`). Line 87: "There are two types of mobility scooters: front-wheel drive (FD) or rear-wheel drive (RD)... Rider weight capacity is a minimum of 77 kg (170 pounds) generally upwards to 440 kg (980 pounds) maximum [front-wheel drive]. The rear-wheel drive is used both indoors and outdoors with rider weight capacity of 160 kg (350 pounds). A heavy duty rear-drive can carry up to 230 kg (500 pounds), varying by manufacturer." Lines 103-105: the 3-wheel vs 4-wheel vs folding/travel trade-offs, quoted in "Branches" and in `tradeoffs`. Line 120: UK Class 2/Class 3 invalid-carriage registration rule, corroborated independently below by gov.uk.

I treat Wikipedia here as a primer/community source per `references/sources.md`, not as a standards body — the numbers it gives for weight capacity are single-sourced (I could not independently confirm the exact 250 lb / 170-980 lb figures against a second fetchable page this run; four brand bariatric-chair pages I tried all 404'd). They go into `category.json` only inside judgements (medium confidence, single source stated) or inside `why_experts_care` prose phrased as approximate ("about"), never as a facet `range`. **Single-source flag, added at the blind check:** the four-wheel-scooter weight-capacity/battery-life claim in `scooter_wheel_count`'s `why_experts_care` ("Four wheels trade some turning radius for stability, a higher typical weight capacity, and longer battery life") is also single-sourced to this Wikipedia article and belongs in this accounting alongside the 250 lb and 170-980 lb figures, same as A2/A3 — it just wasn't listed here originally.

### Regulator: UK gov.uk — mobility scooter and powered wheelchair classes

- `govuk_classes` — https://www.gov.uk/mobility-scooters-and-powered-wheelchairs-rules/classes — **verbatim** (scratchpad `govuk_classes.txt`, lines 67-83): "class 2 if it has a maximum speed of 4mph or less" / "class 3 if it has a maximum speed of 8mph" / "Class 3 mobility scooters and powered wheelchairs can also be used on the road. You need to register them with DVLA." / "Mobility scooters and powered wheelchairs in class 2 and 3 cannot be used on cycle lanes." This independently corroborates Wikipedia's 4 mph / 8 mph figures for the UK classification (two sources, same numbers, one of them the regulator itself).

### Regulator: CMS (US Medicare) — power mobility device coverage

- `cms_lcd` — https://www.cms.gov/medicare-coverage-database/view/lcd.aspx?lcdid=33789 — **fetched excerpt**. I could not confirm this one with my own raw-text read: curling the same URL returns an empty JavaScript-app shell (confirmed — scratchpad `cms_lcd.html`/`cms_lcd.txt`, 73,011 characters of nav chrome and a "please enable JavaScript" notice, zero mentions of any K-code). WebFetch's own fetch returned substantive content (the K0800-series POV groups, K0813-K0891 power-wheelchair groups 1-5, and the rule that "the fundamental distinction involves whether beneficiaries have sufficient upper extremity function to self-propel an optimally-configured manual wheelchair"), which I am treating as a fetched excerpt, not verbatim, and citing at medium confidence accordingly. It is not used for any specific K-code's weight or dimension threshold, only for the existence and logic of the group structure, which I describe in the mechanism/vocabulary sections and one judgement.

### Regulator: US DOT/ADA — doorway clear width

- `ada_standards` — 2010 ADA Standards for Accessible Design, PDF, fetched directly and run through `pdftotext` (scratchpad `ada_standards.txt`). **verbatim**, lines 4890-4892 and 4944: "403.5.1 Clear Width. Except as provided in 403.5.2 and 403.5.3, the clear width of walking surfaces shall be 36 inches (915 mm) minimum." and "404.2.3 Clear Width. Door openings shall provide a clear width of 32 inches (815 mm) minimum." This is the standard itself, not a secondary report of it, so I am treating 32 in / 36 in as sourced facts even though only one document states them (it is the regulation, not a claim about the regulation).

### Regulator: FAA — batteries in air travel (general, not wheelchair-specific)

- FAA PackSafe (https://www.faa.gov/hazmat/packsafe) — **verbatim** (scratchpad `faa.txt`, lines 301-306): general lithium-battery and spare-battery carriage rules ("spare nonspillable wet (gel cell, absorbed electrolyte) batteries are limited to two per person," spare lithium batteries must go in carry-on). This page is about batteries carried as cargo/baggage in general, not about batteries installed in an assembled wheelchair or scooter — I looked for the wheelchair-specific guidance (the actual airline rule for an installed battery: whether it can stay connected, and the watt-hour threshold that changes handling) on five further pages (FAA's own wheelchair page returned 403 to every fetch method I have; TSA's wheelchair-battery pages 404'd; a specialist disability-travel site 404'd; `wheelchairtravel.org`'s battery-rules page 404'd; a DOT consumer page 403'd) and could not reach it. I did not put any specific watt-hour threshold in `category.json`, and since this page ended up supporting no claim I actually used, I dropped it from the `references` array rather than force a citation — it is not in `category.json`, only here. Battery chemistry (sealed lead-acid/AGM vs. lithium-ion) is kept as a facet because it is a real, commonly-specified product attribute with obvious weight and handling consequences, but I have **no fetched source for the exact airline rule**, so `battery_chemistry`'s `why_experts_care` only says that chemistry affects travel handling in general terms and does not cite a number.

### Could not reach

- Every brand spec sheet I tried for a leading power-chair, scooter or manual-chair maker (Pride Mobility — two product pages 500'd; Quantum Rehab — no wheel-configuration content on the page that loaded; Drive Medical — two URL guesses 404'd; Golden Technologies — redirect loop; Invacare — 404; Permobil's blog — 404). I had no working search to find the *correct* URL for any of these (see below), so I was guessing paths, and all of them failed. This means I have **no direct spec-sheet confirmation** of which fields brand pages list, in what order — a gap the other finished registers (lamps, power-bank) filled with brand PDFs. I compensated with the MSKTC clinical factsheets, which describe the same specs from the clinician's side instead.
- All retail listings (Amazon search page: 503; SpinLife: 403 on the category page, no product detail in the WebFetch summary of a direct product URL; 1800wheelchair: 403). Consistent with the lesson in `checks.md` that retail listings reliably block fetches. I could not confirm any facet's presence on a retail listing this run, so every facet that isn't already a structured field in the existing add-on schema is marked `researched`, not `listing`, even where I believe (without being able to check) that it is commonly shown.
- Enthusiast/community and specialist-review sources: United Spinal Association's own page pointed me to MSKTC instead of containing content itself; New Mobility magazine's search turned up gift guides and hack articles, nothing resembling a buying guide; I found no RTINGS/Wirecutter-equivalent specialist reviewer that measures wheelchairs or scooters (this is plausibly because the category is clinically prescribed and insurance-funded in the US, so there is less of a consumer-review culture than for e.g. headphones — a judgement, not a sourced fact).
- WebSearch: this session's WebSearch budget (200 calls) was already exhausted before I ran my first query for this category, so every search-shaped step above was done as a direct URL guess, a curl, or a WebFetch of a hand-picked URL, never a search engine, and Bing results fetched through WebFetch came back as unrelated (JS-rendered results that did not reflect my query). No search-excerpt number is used anywhere in `category.json`.
- ISO 7176 (the full standard, any part): iso.org returned 403 to every fetch attempt. I only have Wikipedia's secondary description of what ISO 7176-19 / WC19 cover, not the standard's own text, and RESNA's own standards page listed section names ("Wheelchairs (WCS)") with no content.

## Data points used for derived metrics

- `doorway_clearance_margin` = 32 in (ADA minimum door clear width, `ada_standards`, verbatim) minus the chair/scooter's `overall_width`. Depends only on a sourced constant and a facet; no additional data point needed.
- `capacity_headroom` = `weight_capacity` minus the rider's stated weight. No external data point beyond the two facets; I deliberately did not attach a sourced "safe buffer percentage" because I have no source for one (see Judgements — I describe the practice in general terms only).

## Against the existing add-on schema

Field-kind vocabulary from `zamana-expertise/AGENTS.md`: **vertical anchor**, **category quality currency**, **hard filter**, **sizing** (variant boundary), or **noise** (cut on sight).

| This register's facet | Existing field (`wheelchair.jsonc` unless noted) | Current kind | Proposed kind / change |
|---|---|---|---|
| `weight_capacity` | `personal-standard.jsonc` → `sizing.weight_capacity` (value + unit lb/kg) | hard filter, already structured | Keep as-is. This register just confirms it is the right call — it is the facet that gates the bariatric segment. |
| `overall_width` | `overall_width_raw` (string, commented "doorway compat") | sizing, unstructured string | Promote to `value`/`unit` like `weight_capacity` already is. A raw string can't be compared against the ADA 32 in figure or sorted; the comment shows the schema's author already knew what it was for. |
| `transport_weight` | `sizing.product_weight` (value + unit lb/kg, commented "transport weight") | sizing, already structured | Keep as-is. Already correctly a hard filter / currency field. |
| `battery_range` | `battery_range_raw` (string) | unstructured string | Promote to `value` (numeric) + `unit` (mi/km). It is a category quality currency number (what buyers compare power chairs and scooters on) and a free-text field can't be sorted or filtered. |
| — (no equivalent) | `max_speed_raw` (string) | unstructured string | Not one of this register's top facets (low narrows_set once `mobility_type` and `use_class` are known — speed differences within a class are modest and mostly regulatory, see the UK class rule), but the same promote-to-numeric argument applies if it is kept. |
| `drive_wheel_position` | none | — | New categorical field, power-wheelchair only. Proposed as `researched`/annotation+filter, not a hard filter, because I could not confirm it is reliably stated even on brand pages this run. |
| `scooter_wheel_count` | none (implicit in product title) | — | New categorical field, scooter only. Near-zero backfill cost — it's almost always in the title — so worth a real field rather than leaving it to text search. |
| `battery_chemistry` | none | — | New categorical field. Backfill cost medium; I have no confirmed source for how reliably brand pages state it. |
| `frame_type` | none | — | New categorical field, manual-wheelchair only (rigid vs folding). |
| `self_propelled` | none | — | New boolean. Distinguishes self-propel chairs (large wheels, push-rims) from attendant-propelled/transport chairs (small wheels both ends) — these are different products with different weight, price and independence implications, and "transport wheelchair" is a distinct, common listing title, so this is close to free to derive. |
| `use_class` | none | — | New categorical field (indoor / indoor-outdoor / outdoor), power-wheelchair primarily, loosely informative for scooters. |
| `seat_width_raw`, `seat_depth_raw` | present, unstructured strings | sizing | Not promoted to a top facet in this register — see "Trimmed detail." Still worth structuring to value+unit for consistency with the rest of the `sizing` block, but that is a sizing-fit concern, not category expertise. |

No field in the existing `wheelchair.jsonc` or the shared `personal-standard.jsonc` sizing block struck me as noise under the AGENTS.md rules; the add-on is already small and each field maps to something real.

## Branches

All branches are on the `mobility_type` split (the one split with branches). Each leg traces to the mechanism split itself — the three device types genuinely do not share most of their other facets.

- `when: ["manual_wheelchair"]` → `then_ask: ["self_propelled", "frame_type"]`, `makes_irrelevant: ["drive_wheel_position", "battery_chemistry", "battery_range", "use_class", "scooter_wheel_count"]`. Trace: manual chairs have no battery or drive wheels (mechanism section; MSKTC's manual-chair factsheet covers frame/tire/backrest, never a battery or a drive position), and `use_class`/`scooter_wheel_count` are defined only for power chairs and scooters respectively (Wikipedia line 262, line 87/103-105).
- `when: ["power_wheelchair"]` → `then_ask: ["use_class", "drive_wheel_position"]`, `makes_irrelevant: ["frame_type", "self_propelled", "scooter_wheel_count"]`. Trace: `drive_wheel_position` and `use_class` are both defined and sourced specifically for power chairs (MSKTC power-chair factsheet lines 129-146; Wikipedia line 262); `frame_type` (rigid/folding) and `self_propelled` are manual-chair-specific terms in the sources I read (MSKTC's manual factsheet line 129; "self-propelled" vs "attendant-propelled" is a manual-chair distinction in `wikipedia_wheelchair`'s taxonomy of wheelchair types). `battery_chemistry` was removed from this leg's `then_ask` at the blind check (2026-10-25) — see the corrections log below.
- `when: ["mobility_scooter"]` → `then_ask: ["use_class", "scooter_wheel_count"]`, `makes_irrelevant: ["drive_wheel_position", "frame_type", "self_propelled"]`. Trace: `scooter_wheel_count` (3-wheel vs 4-wheel) is scooter-specific (Wikipedia lines 103-105); `drive_wheel_position` as MSKTC defines it (front/mid/rear relative to center of gravity, with joystick control) is specific to power-wheelchair bases, not scooters, which steer by tiller, not a drive-position choice between three named configurations. `use_class` on this leg is now also traced to `wikipedia_scooter` directly, not just borrowed from the power-chair source — see the corrections log below. `battery_chemistry` was removed from this leg's `then_ask` at the blind check (2026-10-25) — see the corrections log below.

I considered a fourth leg — chaining `use_class = outdoor` to a recommended `drive_wheel_position` or `scooter_wheel_count` value — but dropped it. The sources describe trade-offs (midwheel "more effective for indoor mobility, but not as good outdoors"; 3-wheel scooters "designed for riders who keep on flat surfaces like a home or for shopping") rather than a hard rule that one value makes another facet irrelevant, so this stays in `tradeoffs`, not `branches`, per the instruction to drop any leg that isn't a real "stops mattering" change.

## Trimmed detail

- `seat_width_raw` / `seat_depth_raw`: real clinical sizing (hip width plus 1-2 in is the standard rule of thumb I know from general domain knowledge, but I have no source I fetched this run that states that specific rule, so I am not putting it in `category.json`). These are body-fit measurements, not an expert "I didn't expect that" split — closer to a shirt size than to a fact that changes what else to ask. Left as sizing-block fields, not promoted to facets.
- Exact Medicare weight thresholds for manual chairs (34 lb lightweight / 30 lb ultralight): single-sourced (MSKTC only); kept as a medium-confidence judgement, not a facet `range`.
- Scooter/manual-chair weight-capacity numbers from Wikipedia (250 lb standard average; 170-980 lb FWD scooter range; 350 lb RWD; 500 lb heavy-duty RWD; <10 kg ultralight manual / 200 kg+ outdoor power-chair): single-sourced; kept as medium-confidence judgements and softened ("about," "roughly") in prose, not as facet `range` objects.
- Airline/TSA rules for an installed wheelchair/scooter battery (the actual rule buyers ask about — can it fly without being disassembled, what watt-hour figure changes the answer): I could not reach a source that states it this run (see "Could not reach"). Not in `category.json` at all, including judgements — I would rather say nothing than restate the general-battery FAA page as if it answered the wheelchair-specific question.
- ISO 7176 standard parts list: could not reach iso.org; kept the WC19/ISO 7176-19 transit-wheelchair mention (Wikipedia-sourced) as a judgement, not a facet, since I can't point to the standard's own text.
- RESNA's standards page: fetched, contained only navigation section names, no usable content; not cited.

## Corrections after the blind check (2026-10-25)

A blind checker (`check.md`) found 7 required fixes. Fetch attempts this pass used a plain WebFetch/curl path, not WebSearch — this session's WebSearch budget was already exhausted (same constraint as the original research pass; see "Could not reach" above).

1. **Required fix 1 (`battery_chemistry` uncited, wired into two branch legs).** Removed `battery_chemistry` from `then_ask` on both the `power_wheelchair` and `mobility_scooter` legs of the `mobility_type` split. Tried eight further sources to confirm it's a stated, askable spec: FAA PackSafe (403), TSA Cares disabilities page (screening-procedure content only, no battery-type detail), FAA's wheelchair-battery resource page (403), Delta's mobility-aids page (404), United's mobility-aids page (timeout), Pride Mobility's power-wheelchair page (redirect, no battery content), a Monarch Mobility blog URL (404), and a scooter-battery retailer URL (unresolvable domain). None confirmed it. Corrected `backfill_cost` from "medium" to "high" and rewrote `backfill_note` to say the facet is unconfirmed and not wired into any question, rather than implying a measured medium-cost estimate. The facet stays in `facets` (it's a real attribute with real travel-handling consequences per general domain knowledge) but is no longer asked about anywhere until a real source turns up.
2. **Required fix 2 (`use_class` on the `mobility_scooter` leg borrowed a power-chair-only source).** Fetched `wikipedia_scooter` again and found it does carry its own indoor/mixed/outdoor-equivalent tiering, just under different labels: "small, light scooters for travel" fold for transport (indoor/portable end), "mid-range scooters, which are intended for both indoor and outdoor use" (middle), and "heavy-duty" scooters that "better handle uneven ground... and journeys where dropped curbs are in short supply" but "cannot fit into public transport or inside most shops" (outdoor end). Added `wikipedia_scooter` to the `use_class` facet's `evidence` array and to its own `supports` entry in `references`. This is a real, scooter-specific source now, not an unstated analogy from the power-chair article.
3. **Required fix 3 (`weight_capacity` split question sets a different quantity than the facet).** The question ("Roughly, how much does the rider weigh?") correctly elicits the rider's body weight, which is not `weight_capacity`'s own value. Rather than invent a separate `rider_weight` facet — the sibling `sleds` and `watercrafts` registers both handle the same pattern by naming the buyer-side quantity directly inside a derived-metric formula without giving it its own facet entry (`sleds`: `max_user_weight - (body weight + pack weight)`; `watercrafts`: `weight_capacity − hull_weight − rider weight`) — I followed that precedent and rewrote the split's `rationale` to say explicitly that the answer is checked against `weight_capacity` via `capacity_headroom`, not set as `weight_capacity`'s own value. No facet added; `capacity_headroom`'s formula and `depends_on` are unchanged and were already consistent with the sibling pattern.
4. **Required fix 4 (`use_class` question is a generic indoor/outdoor trap).** Reworded from "Mostly indoors, in and out, or outdoors too?" to "Flat floors only, indoors and out, or curbs and rough ground?" in both the split's `question` and the facet's `cold_start_question` (they must match). Names curbs and rough ground, the actual mechanism at stake for a wheeled mobility device, instead of a sentence that would read the same for furniture or tents.
5. **Required fix 5 (`transport_weight`'s blended numbers).** Re-read `msktc_manual` and `wikipedia_wheelchair`. MSKTC's "under 20 lbs" is explicitly a frame weight ("Titanium and aluminum chairs can weigh under 20 lbs"), not a stated whole-chair weight, so `why_experts_care` now says "an ultralight manual chair's frame can weigh under 20 lb" rather than implying the whole chair. Wikipedia's outdoor power-chair figure is "200 kg or more" ≈ 440.9 lb; the register had rounded this down to "400 lb or more," which is not wrong (441 ≥ 400) but understates the source. Corrected to "440 lb or more," matching the source instead of a rounder but less accurate number.
6. **Required fix 6 (`battery_range`'s uncited "ceiling, not a promise" claim).** Checked `msktc_power` and `wikipedia_scooter` again for any statement about range test conditions (ideal/flat-ground/rider-weight assumptions) — neither source says anything about how range is measured. Removed the specific unsourced claim "(flat ground, average rider weight)" from `why_experts_care` and kept only the general "ceiling, not a guarantee" framing, which is now also logged as its own judgement entry (confidence low, no evidence, noted explicitly as ours and not sourced to a wheelchair-specific fact) — the same treatment the original register already gave the airline-battery gap.
7. **Minor flag (checker: "no action needed beyond awareness").** Added A7's four-wheel-scooter weight-capacity/battery-life claim to the single-source accounting in the Wikipedia sources paragraph above, alongside A2/A3, for consistency.

No finding was rejected; all seven were actioned as above.