The epoch is the exact moment a TLE describes; the further you predict from it, the less accurate the satellite's estimated position becomes.
A TLE is a photograph of an orbit, not a live feed. The epoch is the moment that photograph was taken. CelesTrak's TLE guide puts it simply: the epoch is the time to which all the changing quantities in the element set refer.
The epoch lives in line 1, right after the international designator, as a two-digit year followed by the day of the year with a decimal fraction. Days start at midnight Universal Time (UT).
The ISS TLE in What is a TLE? has the epoch `26308.88559265`.
India is 5 h 30 min ahead of UT, so this ISS snapshot describes the station at about 02:45 IST on 5 November 2026 (see UTC and IST).
Between snapshots, the satellite's real orbit changes. Air drag slowly lowers satellites in low Earth orbit, and a satellite that fires its engines moves to a new orbit the old TLE knows nothing about. The model can only carry the snapshot forward, so the estimate drifts the further you go from the epoch.
That is why this site keeps its data fresh. Its server collects new TLEs from CelesTrak on a schedule, and your browser reuses a downloaded set for at most six hours before asking again. Satellite pages show an Orbital Data Fetched row so you can see how recent the data is; landing pages show an Updated time under the raw TLE. Reading the epoch itself is the quickest way to judge how old a TLE is.
In Raw TLE Data, the number after 98067A in line 1 is the epoch — the first two digits are the year and the next three the day of the year. Open the ISS tracker
The two-digit year is ambiguous, and CelesTrak's FAQ notes that its interpretation had to change after 1999. In practice software, including satellite.js which this site uses, reads 57–99 as 1957–1999 and 00–56 as 2000–2056.
Converting an epoch to a Julian date is the first thing SGP4 does. For 26308.88559265: the Julian date of 0h UT on 1 January 2026 is 2,461,041.5; add 308.88559265 − 1 = 307.88559265 days to get JD 2,461,349.38559. SGP4 then measures time since epoch, tsince, in minutes: a prediction for 06:00 UT on 6 November 2026 has tsince = (2,461,350.75 − 2,461,349.38559) × 1440 ≈ 1,965 minutes.
Every secular change in the model grows with tsince. The drag terms grow fastest: the mean motion is corrected by terms proportional to tsince and its higher powers, so an error in drag shows up mostly as the satellite running early or late along its track, and that along-track error grows with time. That is why trackers prefer the newest TLE available and why a stale TLE is most noticeable for low, drag-affected objects.
The mean-motion derivative fields in line 1 (.00005352 for the ISS) are, according to CelesTrak's TLE FAQ, not used by SGP4/SDP4 at all — only by the older, simpler SGP model. SGP4 models drag through the BSTAR term instead.
Day 32 of 2026 is 1 February, and 0.5 of a day is 12:00, so it describes 12:00 UT on 1 February 2026.
IST is UT + 5 h 30 min, so it is 02:45 IST on 5 November.
The model carries the snapshot forward, but drag and manoeuvres change the real orbit. The longer the gap from the epoch, the more those unmodelled changes add up.