The account is back. What about the time lost?

On 20 October 2026, the Oversight Board again criticised Meta's election-related enforcement errors, citing the No Kings Instagram account in the US and Lula's advertising accounts in Brazil. The issue extends beyond politicians. It raises an awkward question for any appeals system: what exactly does it repair when access returns too late?^[1]

Consider a hypothetical community organisation whose account is wrongly restricted after it posts an invitation to a meeting. Access returns after the meeting has ended. A technical report can mark the error as corrected. The organisers face a different result: being able to speak tomorrow is not the same as being able to invite people yesterday.

A correct decision can still arrive late

The opposite error has a timing problem too. In its decision of 22 June 2023, the Board described a Facebook video calling for people to storm government buildings. It was reported on 3 January and removed on 20 January, after the 8 January attack on Brazil's government buildings. That earlier case shows what a removal count misses: the final decision matters, but so does when it takes effect.^[2]

That does not mean political accounts should be untouchable, or allowed to publish threats until a review ends. Urgency should change how a review is organised, not what the rules say. The system should establish whether a violation occurred more quickly, rather than agree more quickly with a famous account holder. Access to that review should depend on the risk of losing an opportunity that cannot be recovered, not just on name recognition.

Fewer mistakes matter. They answer a different question

Meta reported roughly halving enforcement mistakes in the US between the fourth quarter of 2024 and the first quarter of 2025. That company-reported improvement should not simply be dismissed. But the overall reduction in mistakes in that period cannot tell us whether a particular mistake was corrected before an important event in 2026.^[3]

One useful measure would be the share of confirmed wrongful restrictions linked to a scheduled event that were lifted before it began. The calculation would need to include all identified errors in that group, including those still unresolved at the deadline. This is our proposal for evaluating appeals, not a published statistic or a promised review time.

It would not explain everything: some harm occurs immediately, and an event's date is not always known. It would, however, keep two different achievements separate. Correcting a decision helps. Correcting it while someone can still use the access being returned helps more. An appeal needs not only the right answer, but an answer that arrives in time.