Live status and predictions are separate
CurrentJabe distinguishes live community status, historical outage patterns and a next-24-hour forecast. A historical pattern never fabricates a live observation, and a forecast is never described as an official schedule.
Live community status
An area appears out only after at least ten independent private contributor identities report an outage within a rolling 24-hour window. The state expires after one hour unless newer qualifying reports refresh it. Ten recent positive confirmations can clear it earlier. To make cookie-reset spam less useful, no more than three contributors from the same daily rotating, one-way network token count toward either threshold. Raw IP addresses are never stored. Without qualifying evidence, the status is unknown, not “on.”
Geographic precision
Electricity networks do not follow administrative borders cleanly. Reports are grouped in the selected catalog area: a sourced finer locality where one is available, otherwise a thana or upazila. A locality is its own community evidence bucket, but it is not called a feeder and its approximate map highlight is only for orientation. Feeder-level grouping is used only when a utility reference supports it. District-wide observations are not combined into one prediction, and low-precision results are labelled accordingly.
Reported times and forecast strength
Every usable completed outage time appears in the public aggregated history, including a window supported by one person. Similar times are grouped, and no contributor identity or individual history is published. Sparse evidence may produce an early pattern, but CurrentJabe does not call it a forecast.
A pattern receives the stronger community forecast label only when it has all of the following:
- Ten independent contributors.
- At least ten usable timed outage events.
- Reports across at least three distinct days.
- Evidence within the previous seven days.
- No single contributor dominating the evidence.
- No more than three contributors counted from one daily network token.
Time windows
Exact timestamps are retained, while the predictor analyses overlap in half-hour time-of-day bins. This is forecast resolution, separate from the 24-hour live-confirmation window. Recent, independent evidence receives more weight. The interface presents the strongest contiguous patterns as simple one-hour windows. Missing or forgotten times are never invented.
Accuracy
Report counts and evidence ranges are factual. Numeric accuracy appears only after forecasts have been made in advance and compared with later observations. Until then, CurrentJabe uses evidence strength rather than unsupported probability percentages.