Match the sailing and date first. A page can show two separate facts: the status reported in an İDO record, and Sefer İptal’s model estimate of cancellation risk when one is available. An estimate is not the operator’s decision.
Status and estimate are different facts
Sailing status comes from İDO source records and should be read with its source-check time and, when available, a source link. “Cancelled” means a cancellation is reported in the operator record; “scheduled” means listed at the time checked. Scheduled is not a guarantee of future operation. A missing record alone does not establish cancellation or absence of service.
A cancellation estimate is produced by the Sefer İptal model. It uses historical sailing outcomes and, depending on the model version, weather and marine forecast inputs. See sources and attribution for provider and scope details; external weather data do not explain İDO’s operational decision.
What the percentage means
In machine-readable results, current_prediction.probability ranges from 0 to 1; multiply by 100 to display a percentage. It is the model’s estimate for that sailing at its calculation time. It is not the share of archive records that were cancelled, and it does not mean İDO has cancelled the sailing. Independent future evidence has not established calibration, so do not read the percentage as a guaranteed frequency or outcome.
Status, prediction time and model version
The source-check time (checked_at or retrieved_at) and model calculation time (calculated_at) are separate clocks. Preserve model_name and model_version when a result provides them. These can vary by sailing and version; this page’s publication or update date is not the model’s update time. Use the timezone carried by each ISO timestamp; sailing dates and times shown by the site use Istanbul time. If no calculation time is supplied, do not infer one.
recorded_prediction is an estimate saved before departure; retrospective_prediction was calculated after departure. Do not present either as a current forecast. A missing, stale or unavailable prediction is not zero risk. For current information, use date-specific sailing results or the sailing tracker.
Evaluation and error
Historical replay evaluates past records using archived sailing outcomes and weather data. Reanalysis or weather data reconstructed later is not the forecast a passenger would have seen before departure. A historical score alone therefore cannot establish performance on real-time forecasts.
Prospective evaluation matches probabilities saved before departure with source-backed outcomes learned later. The project keeps notice/record time, forecast publication and calculation time, and outcome-availability time separate. A weather-cancellation positive requires a supported cause and eligible notice timing; an operated control can use a sailing that was confirmed completed after departure. Unknown-cause, nonweather and unresolved cancellations are not counted as cancellations or operated controls in this comparison.
What error and coverage checks are used?
Probability errors such as Brier score and log loss are compared on matching sailings and the same time before departure. Whether an estimate was available is tracked separately as coverage: a missing or old estimate is not scored as zero error or zero risk. Several sailings may be affected by one storm, so sailing counts do not equal independent event counts.
Why is this not yet an accuracy claim?
The frozen evaluation reviewed on 3 November 2026 compared the serving model with route-specific research models on selected route scopes at fixed time leads. Results varied by model and time to departure, and this scope does not represent every supported route. The matched records do not identify independent storm episodes, and multiple sailings affected by the same weather event are not independent evidence. Unless the provider supplies its model issue time, the time we received a forecast does not prove when its forecast model was initialized. This evidence is not enough to establish accuracy or probability calibration on independent storm groups. We do not claim independent validation or a particular success rate.
Coverage and limitations
The service covers supported İDO sailings and available source data only. A current estimate may not be available for every sailing or date. The model estimates overall cancellation risk; it does not establish that weather will cause a cancellation or predict when an announcement will be made. Verify current sailings and ticket decisions through official İDO information.
Sources, use and corrections
Sources and attribution · Automated fetching guide · Results by date · Open-Meteo · Official İDO website · Publisher contact and corrections