How it works
Saltery Bay – Earls Cove
The short version
A record, not a forecast.
The front page asks what a sailing did on days like the one you picked. It answers by finding comparable sailings that have already happened and showing what became of them — nothing is predicted, and no model is asked to guess about your sailing. Similarity search is the whole method.
Where the record comes from
Nobody publishes whether a sailing ran out of room. Five sources each know part of it, and none of them knows all of it.
The departures board
The operator's own conditions page, read on a schedule. Where a board carries a deck-space percentage, the first reading at zero is the minute the vessel ran out of room. Plenty of boards carry no percentage at all — including this route's — and those still publish the time each sailing actually left, and a note beside the ones that loaded to the maximum number of vehicles. Both are read.
The vessel tracker
The ship's own position and heading. It can say a sailing went, and roughly when, and nothing else — it never sees the deck or the compound. For a terminal whose page carries no board it is the only source of a departure time at all.
The terminal camera
A still frame every 5 minutes, archived. Where a camera's lanes have been calibrated, every frame is read geometrically against what that lane looks like empty: free, on every capture, and unable to report a lane it cannot see. Where they have not, a vision model can be asked how full the compound is on a five-level band — that one costs money, so it runs when somebody asks rather than nightly. A terminal can have more than one view, and the second is often the decisive one: the terminal camera watches the marshalling lanes, while a highway camera down the approach road sees the overflow that only starts once those lanes are full.
People in the line
A check-in from the queue, or a report afterwards. This is the only evidence that says outright whether somebody got on; every other source is inferring it from the outside.
The forecasts
Marine and shore forecasts from Environment Canada, shown beside the sailings. They never feed an outcome. Wind is the one condition that cancels a crossing outright and the only one FerryCast cannot observe for itself.
What this install can actually see
The two ends of a crossing are not alike, and neither are their cameras. This is read from the running configuration rather than written here, so it cannot describe a deployment other than this one.
Saltery Bay
- Conditions page scraped
- Camera archiving
- Lane reading calibrated
- Camera sees the berth no
- Departures from tracking yes
- Shore forecast yes
Earls Cove
- Conditions page scraped
- Camera archiving
- Lane reading calibrated
- Camera sees the berth no
- Departures from tracking yes
- Shore forecast yes
- Hwy 101 at Egmont Road overflow
A camera pointed up the approach road can watch a queue without ever seeing a vessel, which is why "no ferry in any frame" only counts as a cancellation where the berth is in shot.
How a sailing gets its word
Six words, and each one is a different claim. They are kept apart on purpose: blurring "filled up" into "waited a sailing" would assert something none of the evidence can support.
-
Space the whole time
Room right up to departure, or a compound the camera watched empty.
any source -
Filled up before departure
Ran out of room, and often the minute it did. How many were left behind is not known — deck space describes the vessel, not the road outside it.
deck space · the board's own note · a report -
Waited 1 sailing
Overloaded, and what was left over fits the next boat.
camera frames -
Waited 2+ sailings
What was left over is more than one vessel can carry.
camera frames -
Cancelled
The feed said so, or no vessel ever appeared at a camera that can see the berth.
the feed · camera frames -
No usable record
Not enough evidence to say. Never quietly dropped: the answer tells you how many records it had to leave out.
—
Which past sailings count
Comparable means the same sailing on the same kind of day. The search starts tight and widens a rung at a time, stopping as soon as it has 5 sailings to answer from:
- the same time (±15 min), day type and season
- the same time and day type, any season
- the same day type, ±60 min
- days outside a long weekend as well, which behave differently
- similar day types — a Saturday standing in for a Sunday
Every widening is said out loud on the answer, so a distribution never looks tighter than the search that produced it. Only the 5 most recent matches count: without a cap, a sailing from two timetables and another vessel ago would weigh as much as last week's.
What it can't tell you
- Deck space is not the queue. It measures room aboard the vessel, and says nothing about the vehicles still standing on the approach road — which is the number a traveller actually wants.
- A fullness band is not a count. Vehicles were counted until measuring against hand-read frames showed the counts were both unstable and the wrong unit: an RV and a hatchback are not interchangeable. How full the compound is holds up; how many cars are in it does not.
- A join time is not an arrive-by time. "Everyone who got on had joined by 11:15" is a fact about those people, not a target — whether it worked depended on how many were ahead of them.
- A share is a history, not a promise. Below 5 comparable sailings the page says the sample is thin and declines to characterise it.
- Silence is not an empty compound. A dark frame is marked unusable rather than read as a queue that cleared.