Launch evidence guides
How to Verify a Satellite-Based Launch Claim
Published (UTC)
Two pages repeat the same launch claim. One embeds an infrared animation; the other cites the first page. How much corroboration do you have?
The answer depends on the claim and on where each page obtained its evidence. Counting links is a poor substitute for tracing them. An official statement can establish that a launch occurred, while a separate imagery analysis can support a particular satellite interpretation. A dozen copies of one analysis still share that analysis as their origin.
This article proposes a practical verification worksheet. Its labels are editorial tools, not calibrated probabilities, existing LaunchDetect API statuses, or the results of a validation study. Use them to make decisions inspectable, then adapt the requirements with the people responsible for your publication.
Write one proposition at a time
Begin with a sentence that can be checked. “A launch occurred” is different from “this satellite frame shows that launch.” Both differ from “the detection was available to users within a particular delay.”
Split a compound claim into separate rows before reviewing it:
- Event: a named mission launched from a specified site on a specified date.
- Observation: a recoverable scene contains a documented feature at a stated acquisition time.
- Attribution: that feature is consistent with the named launch after plausible alternatives are considered.
- Timing: a specified record became available by an evidenced time.
- Modeled context: a displayed path or area was produced by a named method and inputs.
This prevents a strong source for the event from lending unsupported certainty to the rest. It also gives a reviewer something precise to challenge.
Use the source that can answer the question
There is no single best source for every row. An operator’s completed-launch announcement is well placed to report its mission. Instrument documentation explains what a band measures and what complicates interpretation. A data file supplies acquisition metadata. A published scientific analysis describes the author’s interpretation. Your own receipt log describes what your system received.
For example, NOAA’s ABI Band 7 guide explains that the 3.9-micrometre band is sensitive to hot parts of a pixel and includes a reflected-solar contribution during daytime. That supports asking about illumination and competing explanations. It does not identify the cause of a particular bright feature for you.
Preserve the underlying source and the relationship between sources. If a newspaper copied a mission release, list the release as the origin of that factual assertion. If two visualizations use the same satellite frames, note the shared data even when their processing or interpretation differs. Independence can exist at one stage and be absent at another.
A proposed three-part verification rubric
Use these descriptions per claim, with a written decision. They are not a universal ladder that automatically turns an image into a confirmed event.
| Proposed label | Evidence to document | Reason to hold |
|---|---|---|
| Candidate observation | Recoverable scene or clearly attributed published analysis; time, instrument, band or product; the feature being discussed | Missing provenance, ambiguous feature, or unexamined quality issue |
| Corroborated interpretation | Evidence consistent with the attribution; source relationships; relevant alternatives and remaining uncertainty | Repeated claims share one unexamined source, or an alternative remains unresolved |
| Externally confirmed launch | A primary account of the completed event, matched to mission, site and date | Only a schedule, planned window, or livestream title has been found |
Keep distinct conclusions visible. A launch may be externally confirmed while your interpretation of an image remains on hold. Conversely, a genuine scene feature may be documented without enough information to assign a mission identity.
The downloadable worksheet should include: claim ID, exact proposition, source URL, source author or organization, access time, relevant passage or frame identifier, source relationship, competing explanation, decision, rationale, and reviewer. Leave the reviewer field empty until someone has actually reviewed it.
A source-reading example from Artemis I
NASA’s November 16, 2022 release states that SLS carrying Orion lifted off at 1:47 a.m. EST from Launch Pad 39B at Kennedy Space Center. This supports a bounded completed-launch statement.
CIMSS’s same-day analysis discusses GOES-16 ABI imagery of the rocket booster and launch condensation cloud. It reports that overlapping mesoscale sectors supplied 30-second images. This is a separate published imagery interpretation with its own identified material and authorship.
The two sources answer related questions through different routes. NASA supplies an official event account; CIMSS supplies an analysis of satellite imagery. Neither their agreement nor this article constitutes a new pixel-level reanalysis, a complete independence audit of their information sources, or validation of LaunchDetect’s detector.
Here are two worksheet decisions from that source-reading exercise:
| Proposition | Decision and basis | Remaining limit |
|---|---|---|
| NASA reported the completed Artemis I launch from Pad 39B on November 16, 2022 | Advance as an attributed event statement, using the NASA release | This alone says nothing about a particular detector’s accuracy |
| LaunchDetect issued a satellite detection within 30 seconds of Artemis I liftoff | Hold: neither cited source supplies the required LaunchDetect publication and receipt history | The necessary timing evidence has not been established |
The held proposition is an intentionally unsupported exercise, not an allegation about a product result. It demonstrates why familiar facts about an event cannot fill a missing evidence chain.
Test the alternatives before strengthening the wording
For a new imagery claim, record what else could explain the feature. Compare appropriate neighboring frames and bands, check available quality information, and identify whether a feature is stationary, transient, or affected by display choices. Those are proposed review steps; no new controls or radiometric measurements were executed for this article.
Select comparison windows deliberately. A quiet interval at the same location can be useful, but it is not automatically a representative negative control. Different cloud cover, illumination, viewing geometry, or data availability can change the comparison. State why each control is relevant and what it fails to test.
Likewise, a missing signature does not by itself show that no launch occurred. To make a negative claim, you would need to establish that the relevant data and method could have supported it under those conditions.
Preserve a decision that another person can revisit
Record why a claim advanced, stayed qualified, or was held. Keep the prior version when a source changes, and identify which published sentences need correction. Avoid replacing a specific uncertainty with an overall “high confidence” badge that hides what remains unsupported.
A useful final sentence might be: “The operator reported the launch, and the cited satellite analysis describes a consistent signature; this brief has not independently reprocessed the source frames.” It lets readers understand both the evidence and the work actually performed.
To practice, open a LaunchDetect evidence record and separate its observed imagery, published attribution, and modeled context into different worksheet rows. The page itself states that imagery alone does not establish a detection. Trace each assertion to its supporting source before deciding what you can responsibly repeat.