Method version 1.1, 2026-09-29. The same rule applies to every provider, vehicle and launch site.
Launch Library 2 (The Space Devs, a volunteer nonprofit you can support on Patreon; their paper on it) publishes the world's launch manifest but keeps no history of it: when a date moves, the old date is overwritten. Since 2026-08-29 SpaceNexus has recorded every change of more than a minute to a launch's date that our sync sees, with the date before and the date after. Nobody can rebuild this ledger for earlier dates, including us, so every figure states when recording began.
What the sync reads. Launch Library 2's upcoming-launch feed, orbital and suborbital launches alike (no filter is applied). The next 100 launches are refreshed every 15 minutes. Since September 24, 2026 the launches ranked 101 to 400 are refreshed too, one page of 100 every two hours, so each about every six hours; a change to one of them can be recorded up to about six hours after Launch Library made it. Before September 24 a launch further out than the next 100 was not refreshed at all until it came within that window.
A change is only dated by the day we see it if we were watching the launch when it changed. Every move therefore carries an observation. It is seen live when our sync had refreshed the launch within the previous 24 hours and Launch Library 2's own “last updated” time for it is no older than that. Otherwise it is caught up: the change happened while we were not refreshing the launch, so the day we recorded it is not the day it happened.
Caught-up moves are kept, listed and labelled on every page and in the dataset, and are never counted in a weekly, monthly or quarterly figure, a slip count or an on-time score. Since method 1.1 the recorder stores this evidence with every move. Moves recorded before it were classified from Launch Library 2's last-updated time for each launch (where the launch's date has not changed since), the September 29 fact check of the ledger, and the first sync of the deeper feed pages on September 24. Only the last of those rests on an inference (a launch dated more than 45 days out, first seen in that September 24 sync, is taken to be one the sync had not been refreshing), and the dataset names the reason on every move it decided. A backlog move we found no evidence for stays counted, so “seen live” means we found no evidence the change was made more than a day before we recorded it; it is not proof.
A launch with no announced date is still stored with one. “NET 2026”, “NET Q4 2026” and “NET December 2026” are all kept as December 31, 2026 at midnight UTC, with a precision field saying how much of that date is real. When the year turns, “2026” becomes “2027” and a naive ledger records a 365-day slip that nobody announced. We read the precision, and a move only counts as a slip when both dates were known to the day or better and the launch moved at least 24 hours later.
Only slips, pull-ins and same-day changes are “real” moves. Firmings, lost dates and placeholder rolls are published in their own tables and never added to slip counts.
The dataset gives each end's precision and the basis used for it (from_precision, from_precision_basis, to_precision, to_precision_basis: recorded, chained or inferred).
Moves of one launch observed on the same UTC day are netted into one: the earliest “from” and the latest “to”, classified again as a single move. A launch nudged twice in an afternoon is one change, and a move that returns to where it started is dropped.
For every launch that flew (success, partial failure or failure alike), the baseline is the first firm date we have on record for it: the date it held when we started watching, or the first date it was given to the day. The delay is liftoff minus that baseline. A launch is on time within one day, or within seven days, when it flew no more than that long after its baseline; flying early counts as on time. Averages and medians are of the delay in days.
A launch that went from a placeholder period straight to liftoff, with no firm date on record before it flew, is listed but not scored. Rates are shown only for a provider, vehicle or site with 3 or more scored launches.
Known limit. We cannot yet tell how far ahead a never-moved launch's date was set, so a launch added to the manifest a few days before it flew counts as on time. This affects every provider the same way, and weighs most on providers whose launches appear late in the public manifest. From September 26, 2026 we record when each launch first appears, and a later version of this method will add a minimum-notice rule once that history exists.
A frozen edition is stored once and never recomputed, so a figure you cite reads the same next year. If an edition turns out to be wrong, the correction is logged on /corrections with what it said and what it should have said. The live Slip Explorer and on-time record keep moving; the editions do not.
Method changes. Version 1.1 (September 29, 2026) stopped counting caught-up moves. Editions frozen under version 1.0 are not recomputed; where one counted caught-up moves, a correction beside it gives the corrected figures.
It is not a judgment of any company's engineering. Weather, range conflicts, customer readiness and regulators move dates too; the ledger says how often and how far dates moved, not why. It covers launches listed in Launch Library 2's public feed only. No sponsor or advertiser sees, shapes or pays to change any figure (advertising policy).
Data is licensed CC BY 4.0. Cite as: “SpaceNexus Launch Ledger (spacenexus.us/launch-slips), method v1.1” and link the edition you used. The full move-by-move ledger, with each move's kind, precision basis and observation (seen live or caught up), is a CSV on the datasets page.
Week in Space Data, every Friday.