How BlindSpot counts
The method behind every number on this site, and how often each part is right. Every figure below is read from the BlindSpot API as this page loads.
The blind hours
NASA's polar fire satellites look at Punjab in a few short windows a day. GK-2A, a geostationary weather satellite, scans every ten minutes. Its record shows when the fields burn, and how much of that no polar satellite sees.
Pass times, checked
The pass clock on the live map is computed in your browser from CelesTrak orbital elements. The API checks the same prediction against the time NASA stamped on every fire detection.
Reading a burn scar
A burnt field stays dark in Sentinel-2 pictures for days, whatever hour it was lit. The rule compares every clear pass with the clear pass before it, and then asks whether any fire alert explains what changed.
How often it is right
Every candidate unseen burn rests on that rule, so the rule was audited: scars drawn at random, each judged by eye from its pictures before and after.
Smoke against the air
The smoke model carries particles from every fire detection and candidate unseen burn on the forecast winds. To check it, the API replays it over the last days and compares the smoke at each city with the PM2.5 its stations measured.
Ranking the next burn
Fields that are harvested but not yet burnt are ranked by how likely they are to burn next, so a seeder or a visit can reach them first. The formula is open, and so is how well it has done.
How the investigator is checked
Gemini writes the investigator's answers from BlindSpot's own tools. Before an answer is final, code checks each of its figures against what those tools returned and holds it to four honesty rules. The counts below are BlindSpot's own record of every stored run.
What it cannot do
The limits, in the words the API publishes beside the numbers they qualify.
Where the data comes from
Every figure on this site comes from one of these, fetched by a scheduled job and stored with the moment it was fetched.
How the parts connect
Jobs fetch and compute on a schedule and write to one database. The website and the investigator only ever read the API, so a slow upstream never slows a page; it only makes its numbers older, and the page says how old.