A load figure needs things the event duration monitoring system does not collect. How much left the pipe, what was in it, whether rainfall diluted it, and what state the receiving water was in [3]. EDM records only the moment an overflow starts and the moment it stops [2]. The account published on phys.org makes the failure mode concrete: an hour into a small river in dry weather and an hour into a river running high after heavy rain are the same row in the dataset [4]. Worse for anyone ranking outfalls, several short spills can carry more pollution than one long one [5], so the metric does not even rise reliably with the thing it is used to proxy.
Then there is the arithmetic of validation. Across 91 sites on the tidal Thames between December 2025 and January 2026, the researcher recorded 1,988 published spill hours that were later cut by 82.9% to 339 [6]. That is 1,649 hours deleted [1], and it means the live feed was reading about 5.9 times the eventually validated duration [2]. Per site, the first-published average over the period was roughly 21.8 hours; the validated average was about 3.7 [3]. Anyone who pulled a snapshot for a dashboard, a study or a league table in that window captured the 21.8.
The error also concentrates. At one site near Southend-on-Sea, an apparent spill of more than 850 hours was revised to nine minutes [8], a correction of at least 5,700 to one [5]. If that outfall is among the 91 the source does not say so, but on those figures it would account for roughly half the entire correction on its own [6]. Anglian Water's reported live durations over the same period fell 93%, from more than 1,551 hours to 108 [7], a removal of about 1,443 hours [4]. Aggregate spill-hour totals, in other words, are sensitive to individual telemetry faults, because the revisions come from staff reviewing automated sensors after equipment failures, telemetry problems and precautionary alerts [9].
You cannot fix this with a discount factor, because the error has no fixed sign. The same monitoring can also miss a spill that is actively happening [10]. The researcher is explicit that the original figures were not deliberately misleading, and that live numbers are provisional rather than confirmed environmental records [11]. That is the gap the January 2025 mandate left open [1]: a feed with the update cadence of an instrument and the evidentiary status of a draft, published to an audience that includes swimmers using the maps to decide whether to get in [12].
It is worth noting how this surfaced. The monitoring began in December 2025 as an attempt to explain microplastic pollution found in the Thames, and the researcher concluded the data did not answer the questions environmental impact requires [13]. That is the honest test of a transparency dataset: someone tried to use it for its stated purpose and it did not carry the weight.
