Penn Station’s Missing Model

RPA asks New York to accept a capacity case it has not tested.

Liam Blank · The City Club of New York · April 2026 · Updated weekly through the master developer selection.

The Regional Plan Association's April 2026 report, New York Penn Station: Constraints and Considerations for Meeting Future Demand, reaches a clear conclusion. Converting the station entirely to revenue-to-revenue through-running within the existing footprint, the report argues, would reduce regional rail capacity rather than expand it. The peak-direction throughput RPA derives for full through-running at existing platform widths is sixty-six trains per hour. A wider-platform variant that reduces total track count yields sixty-three. Both fall short of the ninety-train benchmark RPA sets as the post-Gateway target. The report concludes that meeting projected demand will require physical expansion somewhere in the regional system, and that operational reform alone cannot close the gap.[1]

RPA does not support that conclusion with the analysis a serious reader would need. The report asserts the dwell-time parameters that drive the headline figures instead of modeling them. It presents a ninety-train tunnel maximum as a demand-derived target. It applies a connectivity logic that no through-running system in international service uses. It treats yard expansion as a cost unique to through-running, even though NJ TRANSIT and Amtrak documents already place a Gateway storage yard at or near Secaucus under the expansion paradigm. It tests one narrow condition—full revenue-to-revenue through-running inside today's Penn Station footprint—and leaves readers to treat that test as an answer to ReThinkNYC's Regional Unified Network, or RUN, proposal. RUN is not simply "run trains through today's Penn." It is a capital and operating program. It combines platform reconstruction, widened vertical circulation, edge terminals, fleet transition, service planning, yard strategy, and governance reform. In RUN, edge terminals are peripheral stations where trains can be turned, stored, or originated rather than sent to traditional outer-end yards. RPA does not evaluate that program.

But constraints are not conclusions. They become conclusions only after someone tests them. This page lays out those gaps. RPA has not built the case for foreclosing through-running on the documentary and methodological foundation a decision of this magnitude requires. The Penn Station master developer selection is days away. NJ TRANSIT OPRA records show the railroads already have the federal Service Optimization Study Phase 1 draft, while Phase 2, the final phase, is expected in Summer 2027.[5] The MTA's records schedule calls for productions on April 30 and June 4, but no one can assume those productions will include every underlying file, avoid redactions, or resolve the modeling record by themselves.

The Dwell-Time Question

Section 3 of the RPA report sets the dwell-time parameters that drive every capacity figure that follows. Revenue-to-revenue through-running at existing platform widths, the report says, requires six minutes for passengers to deboard and clear the platform, four minutes for boarding passengers to descend and load, two minutes of schedule buffer, and two and a half minutes for interlocking clearance. The total is twelve minutes of station dwell plus the interlocking margin. A wider-platform variant reduces this to seven minutes of dwell plus the interlocking margin. From these parameters, the report derives capacities of four trains per track per hour and eight trains per platform per hour.[1]

RPA gives no methodological provenance. The report does not cite a simulation. It does not cite an empirical observation at Penn Station or any other facility. It does not cite a peer-reviewed study, an international benchmark, or a sensitivity analysis. It simply states the parameters. The capacity figures that follow multiply those assumptions by track counts.

The pattern is not new. Amtrak, the MTA, and NJ TRANSIT jointly commissioned the October 2024 feasibility study. That study presented its own dwell-time assumptions in similar terms and attributed them to recent simulations conducted by WSP, Amtrak, and NJ TRANSIT.[2] Amtrak's August 2025 and February 2026 public-records responses describe the underlying technical record in materially different terms. The August 2025 appeal denial states that the agency's search did not reveal any simulations or modeling data in Amtrak's possession, and that the dwell-time conclusions relied on publicly available data.[3] The February 2026 response goes further and states that no simulations were performed for the work in question.[4] A November 2025 email released through NJ TRANSIT OPRA shows a Federal Railroad Administration official asking NJ TRANSIT to release a Penn Station-specific RTC model developed in coordination with all three railroads.[5] The three documents do not reconcile on their face. The MTA's records schedule calls for productions on April 30 and June 4; those productions may add to the record, but redactions, withholdings, or gaps could leave the technical question unresolved.[6]

The April 2026 report extends and reinforces the October 2024 study without resolving the contradictions in the public record about that study's methodological foundation. RPA does not show that its parameters rest on the kind of operational simulation comparable Penn Station-area projects have used. The Penn Access analysis used Rail Traffic Controller modeling. The Empire Corridor expansion used RTC. The proposed Penn-to-Scranton service used RTC. Grand Central Madison used RTC. The LIRR Third Track used RTC.[7] The capacity question at the center of the most consequential infrastructure decision in the region's modern history deserves the same analytical treatment.

The figures asserted in the report also diverge sharply from achieved performance at peer through-running stations. The London Elizabeth Line operates twenty-four trains per hour through its central core with scheduled station dwells of forty-five seconds at most stations and sixty seconds at major interchanges.[8] The Paris RER A operates roughly thirty trains per hour through Châtelet-Les Halles with peak station dwells in the sixty- to one-hundred-second range.[9] The Munich S-Bahn Stammstrecke operates thirty trains per hour with a scheduled core dwell of thirty seconds.[10] The Zurich Hauptbahnhof S-Bahn operates with through-platform dwells of approximately fifty-five to sixty seconds.[11] RPA's existing-platform assumption of twelve minutes is roughly an order of magnitude longer than the peer central-tunnel performance.

The peer comparison does not claim that Penn can run at thirty- or sixty-second dwells today. Penn-specific factors matter: Amtrak intercity service, commuter train lengths, platform geometry, vertical circulation, signaling, rolling stock, and crew protocols. But those factors strengthen the case for testing the program advocates actually propose. RPA treats its dwell parameters as structural facts rather than variables that platform widening, vertical circulation upgrades, signaling modernization, and fleet transition could change. International peer systems improved those variables through capital and operational investment. RPA does not test whether the same categories of investment could move Penn toward comparable performance.

The Ninety-Train Benchmark

The report's headline finding rests on a second assumption that warrants attention. RPA writes that future tunnel capacity will deliver forty-eight trains per hour from the west and forty-two from the east, and that to match this combined capacity the station "must be able to reach 90 trains per hour in the peak direction."[1] RPA uses this figure as the benchmark for judging the sixty-six- and sixty-three-train through-running scenarios inadequate.

The ninety-train number is a tunnel ceiling, not a demand-derived service target. The report does not present a service plan that justifies running ninety trains in a single peak hour. It does not analyze whether ninety peak-hour trains is operationally optimal under any reasonable reliability standard. It does not compare the ninety-train target to actual peak service plans at peer through-running networks, none of which size their central-station throughput to tunnel maxima.

Engineers designed the Elizabeth Line's central tunnel for twenty-four trains per hour, and operators run twenty-two to twenty-four in revenue service, preserving reliability margin in the difference.[8] The Paris RER A operates near thirty trains per hour on its central trunk because designers built the central stations to handle that volume; peripheral stations carry less. The Munich Stammstrecke runs thirty trains per hour, and Munich is building a second trunk because the existing core cannot reliably exceed that figure even with extensive operational refinement.[10] The general pattern is that mature through-running systems run their central infrastructure below tunnel maxima for reliability reasons, and they distribute peak service across longer peak windows rather than concentrating it into a single peak hour.

A demand-anchored target derived from projected ridership patterns, rather than from tunnel engineering capacity, would almost certainly land below ninety. A peak-period target distributed across the four-hour AM peak rather than concentrated in the eight-to-nine peak hour would change the capacity test further. RPA does not perform either analysis.

Platform Width, Vertical Circulation, and the Moynihan Record

Among the operational constraints the report lists, vertical circulation receives explicit attention. RPA writes that "on average it takes longer than the National Fire Protection Association 130 (NFPA 130) mandated four minutes for all passengers to exit the train and ascend to the station concourse," and uses this finding to support the broader claim that platform-related constraints make through-running unworkable at the existing footprint.[1]

The 2010 Final Technical Memorandum for the Moynihan Station Development Project documented platform clearance times for every platform in Penn Station under both existing and post-build conditions.[12] The data is unambiguous. In the existing 2008 conditions, expected platform clearance times on seven of the eleven Penn Station platforms were below four minutes. The LIRR platforms ranged from 1.83 to 3.66 minutes. The two southernmost NJT platforms serving the stub-end tracks ranged from 1.86 to 1.92 minutes. Only four platforms exceeded four minutes, all of them the heavily loaded NJT platforms serving Tracks 5 through 12, where clearance times ranged from 5.37 to 5.60 minutes. After the Moynihan project's vertical circulation improvements, expected clearance times on those four platforms improved by 13.7 to 31.5 percent, and three of the four fell at or below four minutes under the build condition.

This record matters in two ways. First, the simple weighted-average claim the RPA report makes does not survive contact with the underlying data. The majority of Penn Station platforms cleared in well under four minutes in the historical baseline. Second, and more importantly, the Moynihan project demonstrated fifteen years ago that pedestrian platform clearance is responsive to vertical circulation investment. The constraint exists, in the form RPA describes, on a specific subset of NJT platforms with documented vertical circulation deficiencies that engineering analysis has shown to be addressable. The October 2024 feasibility study acknowledges the same constraint.[2] The intervening railroads have not pursued the remedy at scale.

RPA's framing treats vertical circulation as a structural condition that disqualifies through-running. The historical record treats it as an infrastructure question with a documented design solution. Whether vertical circulation upgrades on the most constrained platforms would reduce dwell times sufficiently to support the operational regime through-running advocates propose is a question that capacity analysis can answer. The RPA report does not engage with the GPP findings or with the broader question of what platform and circulation investments through-running would require. Treating the constraint as fixed when the documentary record shows it is improvable is the kind of analytical move that requires either a citation or a justification, and the report offers neither.

The Connectivity Argument

The report's connectivity analysis runs as follows. With ten branch lines on either side of Penn Station, there are one hundred possible branch-to-branch combinations. Under that framing, a given Long Island branch has a low probability of sending trains directly onto a given New Jersey branch. Therefore, the report concludes, the connectivity benefit of through-running is also low.[1]

This conflicts with the report's own description of how operators would run through-service. Two paragraphs earlier, the same section acknowledges that "the most effective utilization of through-running would be for relatively short runs making stops within the metropolitan core and then turning back," and that station pairing would be demand-driven. The hundred-combination framing then reverts to a probabilistic logic that no operational planner would actually apply: it treats branch pairings as if planners assigned them at random.

No real-world through-running system uses random branch pairing. The Paris RER pairs branches based on demand patterns. London Thameslink pairs Bedford with Brighton, St Albans with Sutton, and similar high-demand combinations.[13] Munich pairs branches based on ridership analysis. The Zurich S-Bahn assigns lines to specific Stammstrecke slots based on travel-pattern study. In every case, planners identify high-demand origin-destination flows and pair the branches that serve those flows, with through-ticketing and timed transfers handling lower-demand pairings.

A connectivity argument grounded in demand patterns would identify the highest-volume cross-Hudson origin-destination pairs, evaluate which would benefit from one-seat service under various pairing scenarios, and quantify the resulting passenger-minutes saved. The Hempstead-to-Gladstone pairing the report uses to illustrate weak connectivity is precisely the kind of pairing no operational plan in international peer practice has attempted, because the demand does not justify it. Citing it as evidence that through-running cannot deliver connectivity is a strawman.

The Yard Question

The report identifies yard requirements as a structural cost that through-running would impose and terminal expansion would not. RPA writes that through-running "would require additional railyards" and cites advocacy proposals for forty acres in the Bronx and forty-six acres in Secaucus. "Building these yards," the report concludes, "would require significantly more land than expanding Penn Station to create additional track capacity."[1]

NJ TRANSIT's own capital plan tells a different story. The agency's project sheet for the Gateway Storage Yard and Maintenance Facility identifies a new yard as a required component of the Gateway Program, with an estimated cost of $2.4 billion in 2020 dollars and a description that places it "proximate to New York and key NJ TRANSIT hubs."[14] The agency document is explicit that the yard is needed because "to realize both the full potential of the system and keep operating costs to a minimum, a larger rail fleet and appropriately placed storage for it will be required." The project sheet does not condition the yard on a through-running operating paradigm. It conditions the yard on the additional service Gateway is designed to deliver, regardless of whether that service runs through Penn Station or terminates there.

Amtrak's 2015 Gateway System-Level Design technical memorandum identified the same yard requirement at the same general location, in support of an expansion-paradigm operating concept.[15] Gateway-era plans have included a storage facility at or near Secaucus for at least a decade.

The asymmetry in the RPA framing is therefore worth noting. RPA presents yard expansion as a cost unique to through-running, even though the documentary record shows a planned and budgeted Gateway component that the railroads contemplate either way. The forty-six acre figure that RPA assigns to a hypothetical Secaucus yard for through-running overlaps geographically and operationally with the yard NJ TRANSIT has already costed at $2.4 billion. The Bronx alternative is a separate proposal with its own merits and costs. But the report treats yard expansion as a disqualifying liability for one option and an unmentioned line item for the other, changing the analytical treatment of the same physical investment depending on which operating model it prefers.

A defensible comparison would normalize the yard question across alternatives. It would compare cost per acre at candidate locations, displacement impacts (industrial-adjacent parcels at Sunnyside and Secaucus versus residential and ecclesiastical parcels in central Manhattan), construction complexity, schedule risk, and network function. The RPA report does none of this. The acreage comparison it presents is the entirety of the analysis.

What RPA Tested vs. What RUN Proposes

RPA's report does not name ReThinkNYC or claim to analyze RUN. It tests a constrained operating condition: revenue-to-revenue through-running inside today's Penn Station footprint, with dwell times driven by current platform and passenger-flow constraints. That test does not answer ReThinkNYC's RUN proposal.

That wording matters. In the Penn Station debate, "within-footprint through-running" is the operating condition most closely associated with ReThinkNYC's proposal. Testing that condition without analyzing the proposal lets the report appear to answer RUN without actually evaluating it. Current platform limits can explain why today's Penn Station performs poorly. They do not justify treating those limits as permanent when RUN explicitly changes the platform, circulation, fleet, and governance conditions that produce them.

The recommendations at the end of the report acknowledge that limited through-running can improve operations. Special-event through-running—like the defunct Meadowlands "Train to the Game" that ran NJ TRANSIT equipment from New Haven, CT to Secaucus, NJ via New York Penn Station from 2009 through 2016—and peak-shoulder second runs may help. However, as the demise of the Meadowlands pilot proved, bolting through-running onto the margins of a fragmented system is a recipe for failure. Those recommendations do not engage the central question: whether Penn through-running works under the specific package of platform, circulation, fleet, yard, signaling, and governance changes RUN actually proposes. RPA did not produce that study. It produced a constrained scenario, then treated the result as a boundary condition.

Methodology

Three methodological observations follow from the analysis above.

The capacity claims do not appear to rest on operational simulation. Capacity at peer through-running stations is generally evaluated using Rail Traffic Controller modeling for North American applications, Viriato timetabling simulation for European applications, or equivalent tools that capture dynamic interactions among train movements, dwell variability, interlocking conflicts, and recovery margins.[16] A capacity analysis without such modeling is a paper exercise that cannot reliably distinguish robust operating configurations from configurations that work on paper but collapse under real-world variability. The April 2026 report extends the conclusions of the October 2024 feasibility study, whose simulation foundations have themselves been the subject of contradictory agency statements in the public-records record.

The capacity figures are point estimates without sensitivity analysis. Engineering analyses of station capacity routinely report capacity as a function of dwell variability, recovery margin assumptions, train mix, signaling configuration, and crew protocols. The RPA report presents single-point figures with no ranges, no scenario variations, and no quantification of uncertainty. A reader cannot evaluate whether the headline conclusions are robust to parameter changes or whether they collapse under modest revisions to the input assumptions.

The treatment of risk is asymmetric. Through-running implementation challenges, including rolling stock standardization, labor agreement renegotiation, governance coordination across three agencies operating at Penn Station, and yard capacity expansion, receive sustained discussion. The implementation challenges of terminal expansion, including multi-year deep-cavern construction in central Manhattan, displacement of residents and businesses on a city block including a Gothic Revival church listed as eligible for the National Register of Historic Places, the schedule and cost risk of a major capital project on the scale contemplated, and the precedent of the 2022 MTA planning effort that consumed seventy-four million dollars across the three agencies without producing a buildable design,[17] receive minimal treatment. RPA presents expansion as the analytical residual rather than as a comparable proposal subject to the same scrutiny.

Institutional Context

The RPA report appears under the authorship of Rachel Weinberger, Vice President for Research Strategy and the Peter W. Herman Chair for Transportation, and Joshua Berman, Campaigns Manager. RPA identifies itself in the report as the lead organization of the Build Gateway Now coalition.[1] Documents released by Amtrak in response to City Club Freedom of Information Act requests show that Amtrak has paid RPA under contract to manage the Build Gateway Now Coalition and produce communications and research materials in support of Gateway-related advocacy.[18] The Amtrak budget for one of those contracts allocated $350,000 to RPA for Penn Station-related research and communications work. One invoice billed Amtrak $36,000 for a through-running report deliverable aligned with the April 2026 publication.[18] The full documentary record of those productions is available on the RPA-Amtrak Contract page.

The institutional context does not by itself invalidate the analysis. It does mean the methodology has to carry weight that a neutral staff memorandum would not have to carry, particularly where the conclusions align with positions the funding institution has previously advanced. The methodological gaps identified above register more heavily in that light. A report asking the public to accept terminal expansion as the remaining answer to the Penn Station question should show its work. This one does not.

What a More Defensible Analysis Would Look Like

A study capable of foreclosing through-running on the technical record would deploy several elements absent from the April 2026 report.

It would include independent operational simulation of through-running and RUN scenarios at Penn under varying dwell-time, signaling, and platform-width assumptions, with the input assumptions documented and the model files preserved for review.

It would derive the peak service target from projected ridership patterns under various network configurations, rather than from tunnel engineering maxima, and would distribute peak service across the full four-hour peak period rather than concentrating analysis on a single peak hour.

It would calibrate Penn-specific dwell assumptions against achieved performance at the London Thameslink and Elizabeth Line, the Paris RER, the Munich S-Bahn, and the Zurich Hauptbahnhof, identifying which Penn parameters are structurally fixed and which investments can move toward peer values.

It would evaluate three scenarios under common assumptions: status quo operations, the ReThinkNYC Regional Unified Network proposal as actually proposed, and southern terminal expansion.

It would treat displacement, schedule, and disruption risks symmetrically across alternatives, with comparable analytical depth on the through-running side and the expansion side.

It would treat governance reform as a policy variable rather than as an exogenous constraint, evaluating capacity outcomes under both the current three-agency structure and a hypothetical unified regional rail authority on the same time horizon as physical expansion.

It would disclose all funding relationships among study sponsors, study authors, and advocacy coalitions whose positions the report evaluates, then send conclusions through independent technical peer review before publication.

The April 2026 report contains none of these elements. The October 2024 feasibility study that preceded it also omitted several of them. NJ TRANSIT OPRA records show the railroads already have the FRA Service Optimization Study Phase 1 draft as of April 26, 2026, while Phase 2, the final phase, is expected in Summer 2027.[5] The May 2026 master developer selection therefore moves forward before the public has seen the full federal study record.

The April 2026 RPA report has not built the case against through-running on the documentary record. The agencies have not produced the underlying technical files that would let the public evaluate that case independently. Procurement decisions should not foreclose RUN before independent analysis compares it with Penn South under common assumptions. The MTA productions scheduled for April 30 and June 4 may add useful records, or they may arrive redacted, incomplete, or contested. Until the agencies produce the underlying files and explain their assumptions, they ask the public to accept conclusions whose foundations remain outside public view.