Editorial standards

Corrections & clarifications

Corrections are part of the public record. This page lists changes to a factual claim, number, date, attribution, title, or the meaning of a figure. Each entry keeps the original wording beside the correction and links the source. Minor copy edits, formatting fixes and dead-link repairs are not listed. Some entries correct a factual error; others clarify an inference, scope, or attribution.

Updated October 25, 2026 Original wording preserved Sources linked
Correction and clarification

Starlink outage: headline, scope and attribution

Three related changes to the same July 24, 2025 postmortem.

Headline and outage scope

Original
“Starlink Core-Network Outage: How 8,000 Satellites Went Dark for 2.5 Hours.”
Corrected
“Starlink’s 2.5-Hour Global Outage: The Core Network Failed, Not the Satellites.”
Why it changed
The article and its sources establish a ground core-network failure; they do not report a satellite failure. The old title implied the satellites went dark. The new title names the core network as the failed layer, not the satellites.
Source

Subscriber count, satellite status and figure wording

Original
"6M users across 140 countries dropped"; "8,000 satellites up"; "8,000 satellites shared the core's fate"; "healthy data plane"; "The trigger is unknown."
Corrected
The service had 6M+ users across roughly 140 countries; tens of thousands were reported offline. The figure says "8,000+ launched," and fate-sharing now describes the service's dependency on one control plane, not satellite failure. ThousandEyes described a control-plane failure with a data plane that was sporadically functional but unreliable. The component-level trigger is not publicly documented; ThousandEyes relays a secondhand reseller report of an upgrade procedure.
Why it changed
The 6M figure is Starlink's subscriber footprint, not an affected-user count. "8,000 satellites up" is not supported; sources say more than 8,000 launched since 2020. The sources do not report satellite failure, so the glossary and figure were narrowed. The trigger wording now distinguishes a secondhand report from a SpaceX publication.
Source
Guardian, CNET, Straits Times, ThousandEyes and The Verge; see the article's Sources and Method.

Source attribution and glossary precision

Original
The July 24 section included a Navy drone-boat test; the report geography named Asia, Africa and Colombia; the 58,000-report row cited Reuters, Guardian and CNET; glossary rows said "all gateways," "everything downstream" and "millions of sessions"; the figure said "16% of normal traffic"; the queue card said "Recovery math."
Corrected
The Navy drone-boat test is removed because the Reuters-bylined report dates it to August 2025, not July 24. The geography list is narrowed to North America, Europe, Australia and other served regions. The 58,000-report row is attributed to TelecomsTechNews and USA Today. Glossary rows now say gateways on multiple continents failed at once, downstream service failed globally with some links partially active, and the reattach surge is modeled. The figure says "16% of ordinary connectivity"; the caption ends at the public-record limit; the queue card labels recovery math as modeled. The section heading now says gateways across regions, and the social alt and SVG descriptions match the figure's blast, public-record-limit and ordinary-connectivity wording.
Why it changed
Each change brings the wording back to the fetched sources and removes claims the July 24 record does not support. The August 2025 USV test is a separate incident.
Source
ThousandEyes, The Verge, TelecomsTechNews, USA Today, and the Military Times/Reuters syndication for the August date; see the article's Sources and Method.
Correction

GCP us-central1 status window

Original
The status-page header started at 07:44 and recorded a 4-hour-8-minute window.
Corrected
The status-page header starts at 07:42 and records 4 hours 10 minutes. The detailed customer-impact window remains 07:41–11:52, or 4 hours 11 minutes.
Why it changed
Google’s incident page header says the incident began at 07:42 and ended at 11:52. The detailed addendum uses 07:41. The article now reports both without reconciling them.
Clarification

PHP2Shell warning window

Original
“roughly eighteen-hour public defender window.”
Corrected
Three warning intervals: CVE identifiers roughly 22.5 to 23 hours before release, the Calif video roughly 18 hours before release, and Chainguard’s early images 6 hours 19 minutes before the 14:00 UTC release.
Why it changed
Different defenders first learned from different public artifacts. The article now reports the interval attached to each start time instead of treating them as one universal countdown.
Correction

Discord voice outage recovery attempts

Original
“two unsuccessful recovery attempts.”
Corrected
“multiple recovery attempts.”
Why it changed
Discord’s postmortem describes multiple recovery attempts; it does not support an exact count of two.
Clarification

OpenAI monitor diagram attribution

Original
Figure labels said “cause: no-messages instruction generalized” and “cause: ‘as appropriate’ to reference PR.”
Corrected
The labels say “OpenAI account: no-messages instruction generalized” and “OpenAI account: ‘as appropriate’ to reference PR.”
Why it changed
The causes are OpenAI’s account of the incidents, not independently established findings. The diagram now attributes them to the company.
Correction

Supabase fleet size

Original
“2M Postgres databases.”
Corrected
Supabase’s General Availability page reports 44M+ databases managed and 200K+ launched daily. The article uses those figures with their fetch date.
Why it changed
The earlier 2M figure was not supported by the cited sources. The replacement uses Supabase’s own published counters.

Corrections are dated when published. The article's dateModified field and the sitemap use the same date. Read the editorial policy.