howbadisthetrain.com scores every NYC subway line from 0 to 10 using the MTA's own published performance data — the share of rider trips that actually finish within five minutes of schedule. It's built for riders who want a quick honest read, and for anyone who wants to see a year of reliability trend without downloading a spreadsheet.
Why
The MTA publishes a great deal of performance data, and almost none of it is readable. It arrives as monthly tables, split by division, by day type, by time period, and by stop path. The numbers are public and genuinely good — they are just unusable if you want to know whether your line is getting better or worse.
There is also a narrower reason this site exists. Most transit information is published per-disruption: a notice that one stretch of track is rerouted, an alert about a signal problem somewhere in the Bronx. That format answers "what is wrong right now" and leaves the more useful question — "is my line having a normal month, or a bad one?" — for you to assemble yourself from a pile of notices. A score is a compression. So is a ranking. The reliability page is the fuller version of that idea: every line, side by side, with a year of trend and the delay causes behind each one.
How to read a score
The score is the MTA's Customer Journey Time Performance for August 2026, divided by ten. That is the entire transformation. 83.0% becomes an 8.3. A 10 would mean every trip finished within five minutes of schedule, which does not happen; a 4 would mean more than half of trips were badly late, which also does not happen on this system.
Read it relatively. An 8.3 next to a 7.8 tells you where to put your attention. It does not tell you whether a train will physically arrive in the next ten minutes — for that, the MTA's own status pages are the right tool, and they are the right tool because they are live and this is not.
What we get wrong
The score inherits every limitation of the underlying data, and the underlying data has real ones. It is monthly, so a bad week inside a good month disappears. It is published about two months behind, so the page describes a period that has already passed. It averages across the whole line, so a problem concentrated on one end of a route gets diluted. It says nothing at all about crowding, because the MTA does not publish per-car crowding data.
Where the MTA publishes no value, this site shows no value. Two lines are reported jointly (J and Z) and get identical figures rather than an invented split. One line (the Staten Island Railway) is not in the subway datasets at all and gets an explicit "no data" page rather than a borrowed number. The methodology page lists every one of these limits in full.
What you get
- Every NYC subway line, ranked worst-to-best, on a single page.
- A per-line page with the score, the raw percentage behind it, terminal on-time performance, wait assessment, and the line's biggest delay cause.
- A reliability rankingwith 60 months of history per line, year-over-year movement, and the network's delay causes broken out — the part you cannot get by reading the MTA's raw tables.
- A methodology page that names the datasets, the definitions and the limits. Read it before you treat any number as gospel.
What we don't do
- We don't require an account. There is no signup, no login, no profile.
- We don't sell or share any data. The site sets no first-party cookies. Cloudflare's privacy-first analytics may be enabled server-side at some point; nothing user-identifying ever leaves your browser.
- We don't invent numbers to fill a gap. That is a recent change and a deliberate one: earlier versions of this site shipped a hand-written table of scores and service advisories that described conditions that were never measured. It looked finished. It was fiction. Everything on the site now traces to a published MTA figure, and gaps are shown as gaps.
- We don't run sponsored posts or paid rankings. Ads come from Google AdSense only, and they only appear in the slots between content blocks, never on top of the score itself.
Who
One person, working on this in the open. The site runs on a small VPS, reads MTA open data (snapshot generated 2026-10-05), and earns revenue through Google AdSense. There is no team, no funding, no growth targets, and no roadmap beyond making the numbers honest and the page fast.
How it's built
- Next.js (app router) on Node 20.
- TypeScript end-to-end. Tailwind for styling, with a small custom palette that borrows from the MTA line colors but stays distinct from any single roundel.
- Docker + Caddy + Cloudflare for the serving stack.
- A Python script pulls the MTA datasets, computes the history, and writes a committed artifact the site renders. No runtime API calls, no database, no cache to go stale.
- No client-side analytics beyond Cloudflare's privacy-friendly web analytics (when enabled). No third-party tracking scripts beyond what Google AdSense sets.
Line colors
The colored ring around each line badge is the MTA-published line color, used here in a thin outline and matching text so it stays visually distinct from the official MTA roundel (which is filled solid). This site is not affiliated with, endorsed by, or sponsored by the MTA — the colors are publicly documented reference values, used to help riders identify their line at a glance.
What's coming
- Per-station reliability, which the MTA's stop-path data makes possible but the current build does not yet use.
- Saved lines — bookmark the lines you ride, get a shorter list on the home page.
- Service-change lookahead for planned weekend work. This is the one place a live feed would genuinely add something the monthly data cannot, and it would be labelled as live when it is.
Open-source pieces
The site itself isn't open source (yet). The build script that turns the MTA's published datasets into the reliability history — scripts/build-line-history.py — is the piece most likely to be useful to other transit-data people, and it is plain readable Python in the repository.
Get in touch
Bug reports, scoring disagreements, line-color corrections, and methodology questions all welcome. Email [email protected] — we read every email.
See also: Reliability, Methodology, Privacy, Terms.