Data snapshot: October 4, 2026. The Atlas country index was read at 20:14 UTC; its source data was as of 15:21 UTC. Incident evidence was read at 20:15 UTC. All figures below describe those responses, not a current access test.
Myanmar's Atlas country row had an anomaly score of 22 from 1,137 OONI samples in a rolling 30-day window. The row calls that signal “low.” It would be easy to read the label as a verdict on internet freedom. It is a label for one measurement calculation. It does not count the people who could connect, the networks that were shut down, or the sites that were reachable throughout Myanmar.
The distinction is especially consequential here. Freedom House's 2025 Myanmar assessment, covering June 2024 through May 2025, documented local shutdowns, blocks on Signal and major VPNs, and damage to telecommunications infrastructure. Access Now and the #KeepItOn coalition documented at least 95 Myanmar shutdown incidents in 2025. Earlier OONI and iMAP research analyzed January–June 2022 measurements and reported confirmed blocking of 112 of 2,130 tested websites across six networks. These are historical findings from different methods and periods. They do not supply a correction factor for an October 2026 Atlas score. They show why a narrow anomaly score should never be introduced as a broad measure of freedom or continuity.
What the country score can say
Atlas says its score starts with the share of available OONI measurements marked anomalous and reduces the effect of small samples before rounding to a 0–100 value. Its published scoring field describes a 30-day window and an empirical-Bayes adjustment. On this read, Myanmar's raw anomaly rate was about 22.6%; the adjusted rate was about 22.4%, which rounded to 22. These percentages describe submitted tests in the index's window. “About 22% of Myanmar was censored” would claim a population estimate that the index did not make.
The sample count belongs beside the score. The same October 4 response gave Venezuela a score of 22 from 17,833 OONI samples. That is more than fifteen times Myanmar's 1,137. Equal rounded scores can rest on very different volumes, test mixes and network coverage. The denominator helps a reader judge the strength of a comparison, but even a large denominator would not make volunteer measurements a census of a country.
OONI's interpretation guide warns that an anomaly can be a false positive. A failed connection needs examination of the test, its network, repeated observations and possible ordinary failures. In its analysis of a separate shutdown, OONI explains that people need connectivity to run and submit its tests. An absence of measurements during a disruption is therefore a measurement gap, not evidence that access was normal.
Follow one dated evidence trail
A Voidly incident record dated September 30, 2026 links 24 evidence items: 17 attributed to OONI and seven to Censored Planet. Its corroboration field counts two source families. The associated incident row lists social and circumvention services, including Signal, WhatsApp, Tor and Psiphon, as subjects of the record. The evidence response is a useful route into what the system associated with that date. Twenty-four links are not 24 independent organizations or a count of people unable to connect.
The OONI items point to dated Myanmar Explorer searches, including a Signal search for September 30. Censored Planet items point to its data portal. Several evidence items share a destination, so the link total should not be mistaken for 24 distinct public measurement pages. To claim that a particular app was blocked on a particular network, a researcher should open the underlying measurements and identify the network and test result.
Censored Planet's data guidance also cautions that its remote observations need processing to avoid false inferences and cannot identify who caused interference or why. Two source families make the trail broader than one, but the combination alone does not establish a nationwide block, an actor, or a continuing condition. Voidly's October 2 Myanmar record, for comparison, linked 13 OONI items and counted only one source family. The number and identity of sources can change from one dated record to another.
Keep the windows separate
The 30-day country score, a one-day incident record, and annual shutdown reports answer different questions. The Atlas index summarizes a rolling set of OONI tests. The September 30 record organizes links related to a particular date. Access Now's 2025 count concerns documented shutdown incidents during an earlier calendar year. None of these quantities can be substituted for another.
That separation helps when writing a sentence a reader can check. “Atlas scored Myanmar at 22 from 1,137 OONI samples in its October 4 response” states the method, denominator and date. “A September 30 Voidly record linked 17 OONI and seven Censored Planet items” states a different, narrower fact. A claim about who ordered a shutdown, how many people were affected, or whether a named service works today needs fresh, specific evidence beyond either response.
The practical reading is to begin with the data-as-of time, carry the sample count with the score, then follow a dated record to its upstream tests. Myanmar's row can direct attention to possible interference. The evidence trail can sharpen the next question. Neither should be made to stand in for the whole country's experience online.