The ADA obliged existing rail to retrofit key stations only — 310 of 475 New York City subway station records are not accessible

What is happening?

The Americans with Disabilities Act of 1990 did not require existing rail systems to make their stations accessible. It required them to make key stations accessible, on a list each operator drew for itself. In the largest legacy system in the United States the state open data portal records 310 of 475 subway station records as not accessible, against 156 fully accessible and 9 partially accessible, as of a dataset update on 2026-05-08. The New York City subway operator designated 54 key stations in the federal plan it filed under that statute.

The structure sits in two layers, and only the second is a deadline. Under 42 U.S.C. 12147(b) key stations were due within three years of 1990-07-26, and the implementing rule at 49 CFR 37.47(c) sets the date at 1993-07-26 with an extension the FTA Administrator may grant as far as 2020-07-26, conditioned on two-thirds being accessible by 2010-07-26. Commuter rail runs the same way under 49 CFR 37.51 with an outer limit of 2010-07-26. A legacy station that is not a key station was never on any clock at all. It is not a missed deadline and not a grace period. It is a lawful steady state.

One mode in the same statute got no carve-out. Under 42 U.S.C. 12162(e)(2) all intercity rail stations were due within 20 years of 1990-07-26. The national intercity rail operator missed that date, and on 2020-12-02 the Department of Justice entered a settlement agreement covering over 400 of the roughly 514 stations it serves. Where an obligation existed, an enforcer appeared. Where the exemption applies there is nothing to enforce.

Whose problem is this?

The party that decides which stations carry an obligation is the party that would have to pay for it. Nothing in the rule makes anyone else draw that line, and nothing in the rule reopens it once it is drawn.

Where does this problem end?

The boundary matters here because the exemption is the law rather than a failure to obey it. A reader who arrives expecting a compliance story will look for a violation and find none.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no federal source opened here names a target number of accessible stations, a completion year, or a share of the network to be reached. The only dated target found anywhere is a settlement under a city antidiscrimination ordinance covering one system — 95 percent of the 364 then-inaccessible New York City stations by 2055 — and the federal government is not a party to it.

How big is it?

About 1.6 million to 3.15 million adults, and that band covers only the two states whose station-level data was opened. The floor scales the 2022 count of adults with a mobility disability in each state by that system inaccessible-station share — 1,970,414 in New York State against 310 of 475, and 1,177,533 in Illinois against 38 of 144, giving about 1,596,700. The ceiling drops the scaler and counts every adult with a mobility disability in both states, 3,147,947.

The two ends answer different questions. The floor asks how much of each network is closed to a person who cannot use stairs. The ceiling asks how many people live under a rule that governs the rail they would otherwise use, whether or not they ride today.

Neither is a national figure, and the true national number is larger than the ceiling rather than smaller. Pennsylvania, Massachusetts, New Jersey, Ohio, Maryland and every other state with pre-1990 rail is outside both ends. So are children, since the prevalence survey covers adults 18 and over.

The chain scales prevalence by an inaccessible-station share, which is a proxy for a population share and not a measurement of one. It also mixes 2022 prevalence with station status recorded in late 2025 and mid 2026. Nothing in it is measured at a single moment.

Under what conditions does it arise?

1. The operator draws its own boundary. 49 CFR 37.47(b) supplies criteria — high boardings, transfer points, interchanges with other modes, end stations, stations serving major activity centers — but the public entity identifies its own key stations and files the list. One criterion shrinks the list from inside, since a high-ridership station may be omitted when it is close to another accessible station.

2. The list is never reopened. Nothing in the rule revisits a key-station designation as ridership, land use or the age structure of the population change over the following three decades. A line drawn in 1992 still governs in 2026.

3. The one live federal lever fires only on renovation. Under 49 CFR 37.43 an alteration that affects usability must be made accessible to the maximum extent feasible, and path-of-travel work is deemed disproportionate above 20 percent of the primary alteration cost. An agency that never renovates a legacy station never owes anything on it. Leaving a station alone is the cheaper legal position.

4. Cost is real and does not explain the pattern. Secondary reporting puts an elevator installation near USD 5 million and records a USD 965 million contract for 21 elevators at 8 stations in November 2022. Under the same cost structure and the same exemption, Boston heavy rail records 152 of 154 stop records as accessible while the New York City subway records under a third. The exemption did not produce one outcome. It produced a lottery settled afterwards by state law, city law and litigation.

What has been tried?

Two rows above share the year 2020 and must not be read as one fact. The federal row is the date on which the extension window closed for rapid and light rail, and no opened source reports which agencies finished or which key stations remain. The state row is the status of a 100-station list created by New York law in 1994, which is a different obligation with a different number.

What was found?

Why is it still unsolved?

Institutional exemption — the undesirable state is the lawful one, so there is no violation to find and no deadline to miss.

The first movement is in the statute itself. Congress wrote 42 U.S.C. 12147(b) so that an existing rail system owes accessibility at the stations it designates as key and nowhere else. That sentence removes the entire apparatus a rights claim normally runs on. There is no complaint to file about a station that was never required, no agency with jurisdiction over it, no date on which anything came due. The intercity rail comparison is the proof rather than a rhetorical flourish. Intercity rail got the same statute in the same year with no key-station limitation, missed its absolute deadline, and drew a federal enforcement action a decade later. Enforcement follows obligation. The problem here is upstream of enforcement.

The second movement is that the exempt set was drawn by the party that would pay to shrink it, and then frozen. The regulation supplies criteria and the operator applies them to itself, files the list, and is done. One criterion lets a busy station out when another accessible station is nearby. Nothing reopens the list when a neighborhood changes or the riders age. The only federal lever that still moves is 49 CFR 37.43, which attaches to alterations rather than to conditions, and which therefore creates an edge nobody designed — renovating a station generates an obligation while leaving it untouched generates none. Every case in the New York litigation record that produced an elevator produced it that way, at a station somebody had chosen to rebuild.

The third movement is what happens when federal law does not reach the majority of the problem. Every real remedy had to be built somewhere else, and everything built elsewhere is repealable. The 100-station floor in New York is a state statute passed in 1994. The 2055 schedule is a settlement under a city human rights ordinance. The money is a discretionary grant program whose appropriation runs out and which has issued three notices in five years. None of these has the property that made the ADA matter, which is that it applies whether or not anyone is currently paying attention. And underneath all of it the federal government funds the retrofit without measuring the backlog. The national station inventory counts 6,341 passenger stations by whether they sit above, at or below street grade, and has no column for whether a person using a wheelchair can enter one. A program can be sized in dollars indefinitely without ever being sized in stations.

What observation would mean it is solved?

Candidates — (a) a federal source publishes a station-level accessibility inventory and the inaccessible count falls toward zero (b) the milestones in the New York settlement are met and published on schedule (c) Congress removes the key-station limitation so that every legacy station carries an obligation.

(a) alone counts flags, not journeys. Every dataset opened here treats accessibility as a permanent attribute of a station, and the separate elevator-outage litigation exists precisely because it is not. The Chicago accessibility column carries no description at all in its own metadata, so nobody publishes what it means there. Nine New York stations are coded partially accessible and no source says what a rider can actually do at one.

(b) alone is one city under a city law. The schedule runs to 2055 at roughly nine stations a year against a 364-station backlog, and it says nothing about Philadelphia, Cleveland, Baltimore or Pittsburgh. It also rests on a base nobody has reconciled, since 364 inaccessible plus 113 accessible does not match either published total for the same system.

(c) alone repeats the intercity rail result. An obligation without money or a schedule produced, in the one mode that had it, a deadline missed by a decade, a design timetable rather than a completion date, and per-person compensation capped below USD 2,500. Removing the exemption would create the hook that is missing. It would not by itself supply the elevator.

What is it connected to?

Fills with researchparatransit demand and cost, the geography of employment access, elevator maintenance and reliability regimes, station accessibility in bus and ferry networks, and the same retrofit exemption pattern in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• No federal source states how many rail transit stations in the United States are inaccessible. The national station inventory has 37 columns and not one concerns accessibility, elevators or wheelchair access, verified at the column schema rather than from a dataset description. The three funding notices for legacy stations contain no such figure either. The program is sized in dollars and never in stations, so nobody can say whether USD 1.372 billion is 5 percent of the job or 50 percent.
• No published list of designated key stations, of the extensions granted, or of the ones still incomplete. A full-text search of Federal Register documents returns planning rules, a 2015 FTA circular notice and the funding notices, and no compliance report. That search reaches back only to about 1994, so the 1991 rule and any early extension approvals could not appear in it. The defensible statement is that no notice since 1994 reports which extensions were granted or which key stations remain.
• For the largest legacy system in the country, nobody publishes whether the federal obligation ever came due. The New York City subway operator designated 54 key stations with a 2010 date. No opened source says whether an extension was granted, what the extended date was, or whether those 54 were completed. The widely repeated status of 97 accessible belongs to the separate 100-station obligation created by state law in 1994, and the two tracks are routinely fused in secondary writing.
• No source reports whether the first milestone of the 2055 schedule was met, and the two opened sources disagree on when it fell due — one says 81 stations by 2024 and the other says 81 before 2025. Neither states the counting unit, so the comparison cannot be closed. A 33-year remedy schedule with no published status against its first checkpoint is the sharper finding than any answer would be.
• No source states whether 95 percent by 2055 means 95 percent of stations or of station complexes. Both bases are published for the same system, 163 of 472 stations against 133 complexes, and the choice moves the target by roughly forty stations. The settlement figure of 364 inaccessible stations reconciles to neither published total.
• No source names the residual 5 percent that the settlement treats as technically infeasible, states the criteria for that judgment, says who makes it, or says whether it can be revisited before 2055.
• Nothing counts the trip not taken. Ridership data counts boardings, which are journeys that happened. No federal source, no agency source and no litigation document opened here reports how many journeys are abandoned, rerouted or never attempted because a station has no elevator. The measure that would size the harm is the one measure nobody collects.
• No uptime target, outage cap or baseline outage rate exists in anything opened here. The elevator-outage settlement reached in July 2026 sets notice obligations rather than reliability obligations, and the case record carries no count of elevators in the system, no outage rate and no mean outage duration.

See the evidence

Five sources carry the load and were read directly — the statute, the two implementing regulations, the funding notices in the Federal Register, and three open datasets queried through their own APIs. The strongest single finding, that the federal station inventory has no accessibility column, was confirmed at the column schema rather than taken from a dataset description. Where sources overlap they agree on magnitude and disagree on the exact pair, and those disagreements are left visible. The New York denominator is 472 stations in one source, 475 subway records in another, 477 implied by a third and 496 records including Staten Island Railway, so no numerator here is divided by a denominator from a different source. The accessible count is 163 in one and 156 full plus 9 partial in the other. The first milestone of the 2055 schedule is due in 2024 by one account and before 2025 by another. The 1984 and 1994 history, the status of 97 accessible against the state list, and both cost anchors rest on a single secondary source and are attributed as such rather than promoted. Two claims are deliberately not made — the completion year and cost of the Chicago program, because that page could not be opened, and any statement that the federal key-station obligation in New York was met, because the widely reported figure belongs to a separate state-law list. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 1,600,000–3,150,000 As of 2022 prevalence against station status of 2025-11-19 and 2026-05-08

Sensitivity The width is the distance between two questions rather than a confidence interval. The floor of about 1.6 million asks how much of each network is closed to a person who cannot use stairs, and the ceiling of about 3.15 million asks how many people live under the rule that governs the rail they would otherwise use, whether or not they ride today. This is not an apportionment — nothing splits a national figure by population share, and the scalers are inaccessible-station shares measured directly in each system. The chain mixes 2022 prevalence with station status recorded in late 2025 and mid 2026, so nothing in it is measured at a single moment. It fails to count children, because the prevalence survey covers adults 18 and over; the trip never attempted, because ridership data counts only journeys that happened and no instrument anywhere measures suppression; the broken elevator, because every dataset opened treats accessibility as a permanent attribute of a station while the separate outage litigation exists precisely because it is not; non-mobility disabilities for which stairs and platform gaps are barriers, against an outer envelope of 73,434,428 US adults with any disability in 2022 that no source apportions to station access; and everyone with a stroller, luggage, a delivery cart or a temporary injury, who appears in no disability statistic at all. The limit in the opposite direction is that both ends are state-bounded rather than national. Pennsylvania, Massachusetts, New Jersey, Ohio and Maryland all have pre-1990 rail and are outside both numbers, so the true national figure is larger than the ceiling rather than smaller. A metropolitan-level estimate was attempted and abandoned because the Census API now requires a key, which is why the range is bounded by state.

Regional breakdown Regional values could not be confirmed. Station-level accessibility was opened for only three systems — New York City, Chicago and Boston — and the Boston figure is stop records on heavy rail only, which is not comparable to the other two. The federal station inventory that would supply a national regional breakdown has 37 columns and none concerns accessibility, so there is no source from which regional shares could be derived without apportioning a national total, which this document does not do.