Observations
What the evidence shows
What we see
Comparing the 29 September and 1 October true-colour crops of the same 5.0 x 5.0 km extent, no large-scale change is apparent in the built-up fabric, the open ground south of the junction, the road network or the vegetated areas. No burn scar, no plume, no collapsed or removed structure is visible at this scale. The two images differ slightly in overall brightness and colour balance, which is consistent with their having been acquired by different satellites of the constellation (S2C and S2A) under different atmospheric conditions.
What it is consistent with
The scene is consistent with an absence of large-area physical change in this 5 km box between the two acquisitions. It is equally consistent with a small strike that this sensor cannot resolve. The observation does not discriminate between those two cases.
What we see
A catalogue query over the area for the window 21 September to 6 October returns five Sentinel-1 GRD acquisitions, all of them before the reported event time, on two descending relative orbits (79 and 152) with a 12-day repeat. No post-event acquisition existed at the time of this query.
What it is consistent with
Radar change detection cannot be attempted for this incident yet. On relative orbit 79 the next acquisition falls due around 3 October, which would pair with 21 September; orbit 152 falls due around 8 October. Both are GRD amplitude products; interferometric coherence would additionally require SLC products, which this catalogue does not carry.
What we see
Reuters reports that witnesses said there was no damage visible near the military base, and that three diplomatic sources speaking anonymously confirmed the blasts.
What it is consistent with
An independent line of evidence pointing the same way as the Sentinel-2 comparison, by an entirely different method: ground observers saw no visible damage, and 10 m imagery shows no visible change. Neither discriminates between a small strike and no strike, and the two are not fully independent in what they would miss — both are insensitive to damage inside a compound or under cover. The blasts themselves are separately attested.
What we see
Both primary sources report that the Ethiopian national intelligence service restricted or banned drone flights over Addis Ababa. ACLED places the announcement a few hours before the blasts.
What it is consistent with
Establishes official concern about drones over the capital in the hours before the event, a contemporaneous official action rather than a post-hoc account. It does not establish that a drone strike took place, and an equally available reading is that the restriction was precautionary and the later blasts had another origin.
What we see
Of the contemporaneous reports located, four carry original material — Reuters, AP, ACLED and Ethiopia Observer — while CNBC Africa, US News, WHBL and others carry Reuters wire copy under the original Reuters headline. The syndicated copies reproduce the same sentences as the Reuters original.
What it is consistent with
The apparent breadth of coverage is mostly republication. Counting outlets would overstate corroboration several-fold. The number that matters for confidence is four, not nine, and two of those four (ACLED and Ethiopia Observer) may themselves draw on reporting rather than on direct knowledge.
What we see
The Reuters article has been revised since first publication. The syndicated copy frozen at 2:37 AM carries the headline 'Explosions heard overnight in Ethiopia's capital, diplomatic sources say' and contains no sentence about visible damage. The version served from reuters.com on 2026-10-02 carries the headline 'Eritrea cuts ties with Ethiopia over Asmara embassy closure' and adds 'Witnesses said there was no damage visible near the military base.'
What it is consistent with
The 'no visible damage' line is a later addition, not part of the first report. Anything citing Reuters for it must cite the revised version. This is the concrete case for archiving on sight: without the syndicated copy the revision would be invisible, and the original wording unrecoverable.
What we see
The sources do not agree on how many blasts there were, when, or where. ACLED says blasts around 1 a.m. EAT at Torhailoch and Meshalokiya. Ethiopia Observer says shortly after midnight, in Gotera and Bethel. AP reports an intelligence official confirming three blasts in three locations overnight, a diplomatic source hearing one at 7 p.m. near the defence headquarters, and residents hearing one past midnight near a factory. Reuters says only 'overnight'. Up to five distinct place names appear across the set.
What it is consistent with
Consistent with several separate detonations across the evening and night rather than a single event at 1 a.m., which is how this record currently models it. The 7 p.m. report is the clearest outlier and is carried by one source only. The occurred_at precision has been widened from 'hour' to 'day' to stop asserting a time the sources do not support; the record may need to be split once the sequence is better established.
What we see
No FIRMS thermal detection within 10 km of the reported area on the night of 30 September / 1 October. The nearest detection anywhere in the window was 8.1 km away on 30 September, flagged low confidence at 2.96 MW.
What it is consistent with
This result is close to meaningless for this incident and is recorded only so the check is not run again. VIIRS resolves 375 m and needs sustained heat; a munition detonation lasts seconds and no fire was reported by any source. The claimed event is below the instrument's floor by a wide margin, so its silence is not evidence. This is the opposite of the Kombolcha case, where a fuel depot fire was the claimed consequence and a null therefore carried weight.