One source is one source
Five outlets carrying the same wire story is one source, not five. Before
counting corroboration we trace each report to its origin. For the explosions
in Addis Ababa on 1 October, nine reports we located resolve to four
independent origins; the rest is syndication.
We say what a blank result means
Every instrument has a floor. A satellite that resolves ten metres per
pixel cannot see a single munition impact, and a thermal sensor that needs
sustained heat cannot see an explosion that lasts seconds. Reporting
“no damage visible” without saying that is misleading, so each
check records what it could have detected and whether it was looking.
Every assessment names its own refutation
Each one states the observation that would change our mind. An assessment
that cannot be overturned is advocacy, and our own tooling rejects it.
Confidence uses a published scale
We do not invent our own words for certainty. Terms come from a standard
analytic scale with fixed probability bands, so “likely” means
55 to 80 per cent every time it appears, and the bands are drawn on the page
rather than left to interpretation.
We archive before we cite
Posts disappear and wire stories are rewritten. The Reuters report on the
Addis explosions was revised after publication: the sentence saying no damage
was visible is absent from the copy published at 02:37 and present the next
day, under a different headline. We keep both.
We correct loudly
Every record carries its revision history. When we are wrong we mark it on
the record. We do not quietly edit a published claim.