FOI: Discrepancies in the API

As it turns out, Toronto’s Bike Share system uses a standardized API¹ format: General Bikeshare Feed Specification, or GBFS.

This is great, because apart from being clearly documented and fairly easy to work with (still looking at you, TTC), we can find some interesting tidbits of information where the API deviates from the standard. Of particular interest is the system information endpoint.

Let’s check it out:

curl "https://tor.publicbikesystem.net/customer/gbfs/v2/en/system_information" | jq .

{
  "last_updated": 1701754523,
  "ttl": 3,
  "data": {
    "system_id": "bike_share_toronto",
    "language": "en",
    "name": "bike_share_toronto",
    "timezone": "America/Toronto",
    "build_version": "2023.1",
    "build_label": "2023-11-17",
    "build_hash": "2a5b9d6",
    "build_number": "267",
    "mobile_head_version": "2",
    "mobile_minimum_supported_version": "1",
    "_vehicle_count": {
      "_mechanical_count": 8126,
      "_ebike_count": 825
    },
    "_station_count": 764
  },
  "version": "2.3"
}

Cool! Some fairly mundane version information, the number of stations, and - ooh, the number of each bike type!

The station count looks right based on the data I’ve gathered.

8126 mechanical bikes passes the sniff test too - or at least it seems within an order of magnitude of what I would expect (we’ll come back to this).

825 e-bikes though? No. No way. Strangely, that’s both way too high and way too low.

• Way too low, since Bike Share Toronto advertises over 1800 e-bikes in the system.
• Way too high, since (as of mid-Sepember) I’ve been querying Bike Share Toronto’s API for the number of bikes at each station, and I’ve rarely seen more than 250 e-bikes available across the entire system.

I have however, seen a ton of disabled bikes at various stations (and frustratingly often at our limited number of charging stations)².

It’s unclear to me what the API is reporting, since if I sum up all the mechanical bikes at every station I get 5698 (at the time of writing). Likewise, if I sum all the e-bikes, I get… 109³.

But, in the immortal words of Anchorman, “60% 70% of the time, it works every time”.

And by that I mean:

Hm. Well, 70% of 8126 (mechanical bikes) is 5688. So 5698 is, uhh, oddly close to their precise minimum contractual requirement.

Look, I’ve been staring at the occupancy graph of Wellesley Station for the last three months, and I walk past it multiple times a day. It’s always full of dead e-bikes, and I can see from the graph that they don’t ever charge. Someone in a truck shows up once every day of two, dumps a bunch of dead e-bikes there (which stay there, since they never charge so they’re reported as disabled bikes), and then a couple of days later someone in a truck shows up and takes them away (until the next night, when someone in a truck shows up…).

So forgive me for thinking that almost every disabled bike is a dead e-bike. I’m sure a small number of them are mechanical bikes, but humour me - the mechanical bikes are pretty bomb-proof.

If we add the number of disabled bikes (468) to the number of available e-bikes, we get: 578 e-bikes. Or, exactly 70% of the 825 e-bikes reported by the API.

Hm.

curl "https://tor.publicbikesystem.net/customer/gbfs/v2/en/station_status" | \
    jq --raw-output '.data.stations[] | [.vehicle_types_available[].count] | @tsv' | \
    awk '{j[1]="Boost"; j[2]="Iconic"; j[3]="E-Fit"; j[4]="E-Fit G5"; for (i=1; i<=NF; i++) sum[i]+=$i} END {for (i in sum) {print j[i] ": " sum[i]}}'