Why an aircraft count fell: a measured SkyWatch case
A dated comparison of aggregate map records shows how time of day, incomplete source coverage, and a blocked feed limit what a lower count can mean.
Open the live mapThe question behind the count
On 26 September 2026, a SkyWatch reader noticed that the aircraft layer showed far fewer records than expected across the map. We preserved aggregate cache snapshots from that day and compared them with earlier retained windows. This study reports only totals and source health. It contains no aircraft identities, positions, or routes.
A displayed total is the number of records that survived the available feeds, position checks, filters, and duplicate merging at that moment. It is not a census of aircraft in the sky. If one provider fails, its missing contribution cannot be counted as zero aircraft; it is an unknown part of the picture.
What the snapshots actually showed
The free global map layer showed 31–34 records in retained morning samples, 54–56 in the afternoon, and 79–104 across several evening samples on 26 September. One cache-only sample at 18:18 UTC held 98 records. These are separate snapshots, not a continuous measurement of every intervening minute.
At approximately the same evening time on the preceding Saturday, two retained minutes held 88–89 records. That older window is short and incomplete, so it does not establish a weekly baseline. The preceding Friday's 18:00 UTC hour showed 247–304 records, but a weekday comparison mixes different traffic patterns, provider coverage, and possible enhanced-feed contribution.
The supplier check was separate
At 18:47 UTC on 26 September, ADSB.lol and ADS-B.fi reported healthy responses in our source check. Airplanes.live returned HTTP 403 for the aircraft feed, and its separate United States candidate request also failed. A failed request is recorded as unavailable rather than an empty result. SkyWatch had removed OpenSky from ingestion by product decision and did not count it as an active supplier.
That health check happened 29 minutes after the 18:18 count, so it cannot prove the exact supplier state of that earlier snapshot. We used existing caches and did not trigger a paid FR24 refresh or an individual-aircraft lookup. The blocked feed establishes a coverage limitation, but this sample alone cannot measure how many additional records it would have supplied.
How to compare two map counts fairly
Record the observation window and time zone, the selected region, layer and filters, the age of the server cache, and which providers succeeded. Compare the same hour and weekday across several dates. Keep enhanced paid data separate from the free-source count, and state how many minutes of each comparison window were actually retained.
Then distinguish the raw supplier totals from records with valid positions and from the total after filtering and duplicate merging. A provider can return many reports that never become map markers. If a step is unavailable, label it unavailable; do not fill the gap with a guessed value or turn a transport failure into a zero.
What we could and could not conclude
The within-day increase and the similar, limited Saturday evening samples show that the low morning display was not enough to prove a sustained collapse in map records. The Friday comparison is too different to settle the question. A blocked source is a real software coverage problem, but the available snapshots do not isolate its numerical effect.
Most importantly, none of these totals proves a change in actual aircraft activity. They describe what SkyWatch received and displayed under known gaps. The useful follow-up is a series of matched, source-aware aggregate samples; until then the honest result is a partial observation, not an operational claim.