A deep-research pass (2026-07-01) across four axes of continued expansion,
building on expansion-research.md (the verified
competitive research), expansion-ideation-2026-07.md
(the horizon scan, now mostly implemented), and the shipped surface as of
today: the six-hub IA, /pulse/, /focus/, /check/, /query/, /compare/, the MCP
server, monthly dataset releases, and the Canada pilot. Everything below is
new relative to those documents, with sources cited inline. Items are tagged
buildable now, partnership-gated, or bet.
The GTFS community adopted the cemv_support field on 2025-09-29 (4 in
favor, none against): agencies and routes can now declare that riders can pay
with contactless bank cards
(PR #545). This is exactly the
shape of signal the What-feeds-publish page tracks. Detection is a column
read in agency.txt/routes.txt, the same pattern as flex.py; adoption today
is near zero, which is the point: the page can watch a brand-new field spread
from day one. Frame as adoption, never quality, as with the rest.
The ideation doc called RT prediction accuracy "the next honest frontier" gated on the always-on archive. The methodology risk is now retired: the community maintains a GTFS-Realtime prediction accuracy metrics definition, MBTA published an open-source transit-performance system measuring prediction accuracy in production, and Mineta published an assessment methodology for temporal accuracy of TripUpdates. When the archive bet is funded, the metric definitions should be adopted from these rather than invented.
The vendor market consolidated around exactly this problem: Optibus acquired Trillium and now markets GTFS services on the claim that "more than 90% of feeds receive first-pass approval from Google" (Optibus); Swiftly acquired Hopthru to sell ridership cleaning and NTD reporting. Vendors now market quality claims a buyer cannot verify. The scorecard is the only national, independent, daily check of those claims, and the procurement page should say so explicitly: the acceptance test works precisely because the referee is not selling the feed.
FTA's 2024 annual data products are out, and monthly ridership now lives on
data.transportation.gov with a Socrata API
(Complete Monthly Ridership, dataset 8bui-9xvu;
NTD annual metrics by agency).
Ridership weighting (ADR 0021) has been gated on a hand-committed CSV; a
weekly fetch keyed on ntd_id removes that gate and keeps the numbers
current. Ridership context also improves the compare page and the brief
("a 2M-trips-a-year agency" reads differently than "an agency").
Statistics Canada consolidates GTFS from about 150 Canadian agencies in the Canadian Public Transit Network Database, and the federal geospatial platform maintains canada-gtfs tooling. The Canada pilot runs three agencies; the discovery pathway for fifty-plus more exists, with provincial licenses already documented by Metrolinx, BC Transit, and TransLink. Growing Canada is now a registry-curation task, not an engineering one.
The OpenSidewalks schema and the OpenThePaths 2026 program are building statewide pedestrian-path data with transit providers, with an interoperability plan for trip planners. The scorecard measures whether a feed publishes accessibility data; OpenSidewalks measures whether the sidewalk to the stop exists. A future stop-area walkability lens would join the two. Not buildable yet; worth citing in the access methodology and watching the interoperability plan.
The pipeline already carries a gbfs.py module, MobilityData ships a
canonical GBFS validator with data-quality
reports, and GBFS is the de facto standard for shared micromobility in North
America. A "shared mobility" lens (does the city's bikeshare publish valid,
fresh GBFS?) reuses the whole scorecard pattern on a new corpus. Scope it
only after Canada scales; it is a second product surface, not a feature.
National RTAP
runs the free GTFS Builder used by rural and tribal agencies, holds weekly
GTFS office hours explicitly in support of the NTD requirement, and runs a
tribal transit program;
DOT is standing up seven new
TTAP centers.
These are the support desks for exactly the agencies the scorecard serves.
The concrete asks are small: /check/ and /try.html as office-hours tools, the
scorecard as the after-you-publish step in their GTFS Builder guide, and the
ntd_note field for waivered or shared-feed reporters they support. One
introduction email each; the product is already shaped for them.
WSDOT builds GTFS for any Washington agency that lacks it and is launching a shared data archive with ODOT; ODOT maintains statewide feeds through a single vendor and the GTFS-ride archive. These programs are the liaison persona at state scale, and the per-state rollup pages are already their portfolio view. The ask: show two state programs their own /program/ page and the one-fix-from-ready NTD table, and learn what a state data manager needs that a Cal-ITP-style CSM does not. Their archives are also candidate partners for the RT-history bet, hosting the storage the cost guardrail will not.
Local data journalism increasingly runs on shared national datasets: Big Local News partners national data with local newsrooms, and the UK's RADAR model generates thousands of localized stories from one dataset. The monthly dataset releases plus the by-state rollups are that shape already. A "story-ready" cut per state (plain-language summary, the covered-set caveat baked in, a reporter-facing methodology note) is one render away, and the press-explainer item (E3 in the research roadmap) becomes its cover page. The no-shaming framing must ride along or this audience is a net risk.
The Optibus/Trillium and Swiftly/Hopthru consolidations confirm vendors buy and sell on data quality and NTD reporting. The vendor worklist stays gated on one vendor interview per the research roadmap, but the interview target list is now obvious: the post-acquisition data teams whose marketing depends on quality claims an independent scorecard can confirm.
The official MCP Registry is the canonical, open feed of MCP servers and accepts listings without an enterprise account. The Claude Connectors Directory (511 connectors as of June 2026) requires a remote server, a privacy policy, read-only tool annotations, and a Team/Enterprise submission, so it is a later step that would require hosting the server remotely. Sequence: registry listing now; connectors directory only if remote hosting ever earns its keep against the cost guardrail.
The Transitland Atlas is
explicitly "open to use as a crosswalk within other transportation data
systems," and MobilityData's
awesome-transit list is
the discovery page the ecosystem actually reads. Two small acts of
ecosystem citizenship with distribution upside: a PR adding the scorecard to
awesome-transit, and publishing the scorecard's own id-to-mdb-to-onestop
crosswalk (the catalog already carries mdb_id) so consumers can join the
grades to either registry. The
Transitland v2 REST API
is also a second discovery source for feed moves, complementing the
Mobility Database in discover.
Same integration as the ridership item above, named here because it is an
integration pattern, not just a dataset: data.transportation.gov exposes the
NTD tables the crosswalk and ridership features need, keyed by the ntd_id
the registry already carries. One fetcher, several features fed.
- Now, small: cEMV detection on What-feeds-publish; MCP Registry listing; awesome-transit PR; publish the id crosswalk; NTD ridership fetcher unblocking ADR 0021.
- Next: Canada registry growth from the StatCan corpus; the story-ready state cut with the press explainer.
- Outreach (human): one email each to National RTAP and a state DOT data program; the vendor interview.
- Bets, unchanged but de-risked: RT prediction accuracy (adopt the published metrics; court a state archive as host); stop-area walkability (watch OpenThePaths); GBFS lens (after Canada scales).
The cost guardrail and the no-shaming principle still bind every item. The GBFS lens and remote MCP hosting are the two temptations most likely to breach the single-digit-dollars budget; neither proceeds without a named user. Nothing here makes the scorecard a feed editor or a vendor ranking.