OSINT desk · Methodology · Corrections

Everything on this desk can be checked by the reader.

The OSINT desk works only from publicly available material — released satellite and aerial imagery, official statements, published tenders and parliamentary records, open shipping and flight data, and photographs shared publicly by the people who took them. Nothing here rests on a source a reader cannot go and look at. Where a claim cannot be supported from that material, we say so rather than fill the gap.

Read the method → How confidence is assessed →

The desk in figures

Recent threads →

412Threads published

1,286Sources cited

268Imagery comparisons

14Corrections issued

How we assess confidence

The six-step method →

One image can support three different statements, and they are not equally reliable. The desk keeps them apart on the page so a reader can see exactly where the evidence ends and the reasoning begins. Confidence falls at every step down that ladder, and the published label falls with it.

Tier ladder · Confidence index

Observation, inference, interpretation

Observation is what a second person would see in the same frame without argument: shapes, counts, shadows, markings, the date stamped on the file. Inference is the pattern those observations form when compared with an earlier frame. Interpretation is meaning — intent, planning, consequence — and it is the tier the desk is least able to defend.

Observation — visible in the frameHigh

Inference — what the pattern suggestsModerate

Interpretation — what it might meanLow

Bars show the confidence the desk will carry into print at each tier, not a measurement of the subject.

Worked example · Unnamed site

One frame, read three ways

Observation

Four of twelve hardstands in the frame are occupied. In the comparison frame, eleven weeks earlier, one was. Both frames are dated, and both dates are published in the thread.

Inference

Occupancy rose across the interval. The desk can say the pattern changed; it cannot say the change was steady, because there is no imagery between the two dates.

Interpretation

Why occupancy rose is not stated. Rotation, maintenance, an exercise cycle and a run of clear weather all produce the same visible change, and imagery alone cannot separate them. The thread stops here, and says that it stops here.

Published labels

What each label commits us to

Every claim carries one of four labels, set at publication and revised in the open if the sourcing changes.

High: Two independent sources
Moderate: One source, partly corroborated
Low: Single source, uncorroborated
Not assessed: Observation published alone

A low label is a publishable outcome. It is not a draft of a higher one.

Degradation · Where confidence is lost

What weakens a reading

Most confidence is lost long before interpretation — in provenance, in dating, and in the absence of a second pair of eyes on the same raw material.

Two independent sources, imagery datedHigh

Single source, imagery datedModerate

Single source, date or provenance unclearLow

Material that cannot clear the bottom row is held, not published.

Method

Sourcing standards →

Six steps, run in order. A thread that cannot complete a step does not skip it — it is published at the confidence the completed steps support, or it is held.

01

Source capture

The original material is archived before any work begins: full-resolution file, the address it came from, the account or body that published it, and the time of retrieval. Working copies are annotated; the archived original is never edited.

Output — archived original, retrieval log, source line
02

Timestamp and geolocation

When and where, established independently of what the source says. Shadow direction and length against sun position, terrain and coastline matching, published weather records, and any date carried in the imagery itself. If the frame cannot be fixed in time or place, that failure is recorded and travels with the thread.

Output — date range, location confidence, method notes
03

Corroboration against a second source

A second source, independent of the first, is sought for the same observation — a different imagery provider, an official statement, a published record. Two accounts that trace back to one origin are one source, and are counted as one.

Output — second source, or a single-source declaration
04

Peer check

A second member of the desk repeats the geolocation and the counts from the raw material alone, without seeing the first conclusion. Where the two readings disagree, the lower of the two is the one that goes forward.

Output — independent second reading, reconciled figures
05

Publication with confidence label

The thread is published with observation, inference and interpretation separated, imagery dates stated in the text, the source list attached, and a confidence label on each claim rather than on the thread as a whole.

Output — labelled thread, source list, stated imagery dates
06

Correction and update

Corrections are published as new posts on the original thread, with what changed and why. The original stays visible and carries a correction note. Where later material raises or lowers a label, the change is dated and recorded on the same thread.

Output — dated correction, amended label, unedited original

Sourcing standards

From the desk →

01

Open material only

Nothing is published that a reader cannot reach through the same public route.

02

No speculation as fact

Interpretation is marked as interpretation, or it is left out of the thread.

03

Every claim labelled

Confidence sits on the individual claim, not on the thread as a whole.

04

Imagery dates stated

Capture dates and the interval between compared frames appear in the text.

05

Corrections published

Errors are corrected in the open on the original thread, never edited away.

06

No operational detail

Nothing is published that could put a person, a crew or a civilian at risk.

From the OSINT desk

Open X feed →

OSINT · 6h

Imagery comparison: two frames, eleven weeks apart

Counts, shadow-derived timing and the interval between captures, with interpretation left open and the reason stated.

↻ 98♡ 740Imagery

Verification · 1d

How a widely shared clip was dated

Weather records, shadow direction and an earlier upload of the same footage placed the clip well before the event it was attached to.

↻ 316♡ 2.1KMethod

OSINT · 2d

Single-source thread, published as single-source

No second source was found in four days of checking. The thread runs with a low label and the gap named in the first post.

↻ 64♡ 430Thread

Corrections · 3d

Correction: a count revised down

A higher-resolution frame showed two objects previously read as one. The original post stays up and carries the correction.

↻ 41♡ 288Correction

Method · 5d

Reading shadows: a short working note

Working through sun position, shadow length and the error margins that follow, using a publicly released frame as the example.

↻ 187♡ 1.3KNote

Sourcing · 1w

When two sources are really one

Tracing three separate accounts of the same claim back to a single origin, and what that does to the confidence label.

↻ 129♡ 905Sourcing

All figures, threads, labels and worked examples on this page are illustrative demo content produced for a design build. They are not sourced reporting or real intelligence assessment, and they describe no real location, facility, unit or person.