Accuracy
Benchmark results, with the scope made clear.
Parky’s internal benchmark measures whether the product reached the correct final parking conclusion from the visible rules in each scenario.
Share of scenarios where Parky reached the correct final parking conclusion from visible rules in a 544-photo benchmark using signs from Canada, the U.S., the UK, and Australia.
Parky internal benchmark resultsWhat the two categories mean
Two-rule scenarios
Parky reached the correct final conclusion in 96% of internal benchmark scenarios involving two parking rules.
Three-or-more-rule scenarios
Parky reached the correct final conclusion in 93% of internal benchmark scenarios involving three or more parking rules.
What these results do not mean
They are not a claim of overall worldwide accuracy or a guarantee for every language, photograph, or street. Real-world results depend on image quality, signs visible in the frame, temporary notices, road markings, local conditions, and ambiguity.
How the benchmark was constructed
The combined benchmark used 544photographs containing two or more visible parking rules. It included difficult and complex sign combinations rather than only simple, isolated signs.
The photographs were evaluated across different times of day, calendar dates, and months. This tested whether Parky’s conclusion changed appropriately as the visible day, date, month, and time restrictions became active or inactive.
The parking-sign photographs came from Canada, the United States, the United Kingdom, and Australia. This describes the benchmark dataset, not Parky’s product availability. Parky works worldwide.
The 544 photographs are the combined sample across the two-rule and three-or-more-rule categories. A category-level sample split, benchmark date, country-level distribution, and evaluator protocol have not yet been supplied and are not inferred here.
Benchmark history
This page will preserve versioned benchmark records as the evaluation changes. The current entry is the initial published summary and awaits the remaining methodology metadata.