You can learn new things again. You can remember what you read last week. But you're still slow. Not confused — slow. The gap between hearing a question and forming an answer. The lag when switching tasks. The extra second before your name comes when someone asks. Three and a half years out, and that hasn't changed.
You're not imagining it. The data says exactly what you feel.
The Split
In 2025, Becker and colleagues at Mount Sinai published the longest cognitive follow-up of Long COVID patients to date: 1,553 people tracked for 42 months across six cognitive domains. The finding that matters is not the average. It's the divergence.
| Cognitive Domain | Baseline (<6 mo) | 42 Months | Outcome |
|---|---|---|---|
| Verbal learning | Impaired | Normal range | Recovered |
| Memory recall | Impaired | Normal range | Recovered |
| Phonemic fluency | Impaired | Normal range | Recovered |
| Attention | Mildly impaired | Mildly impaired | Minimal change |
| Processing speed | ~1.5 SD below normal | ~1.5 SD below normal | Floor |
| Executive function | ~1.5 SD below normal | ~1.3 SD below normal | Floor |
Data from Becker et al., Brain Behavior and Immunity – Health, 2025. N=1,553. Z-scores relative to normative means. BMI <25 was the only predictor of greater improvement.
Learning and memory came back. Processing speed and executive function barely moved. At 42 months, they're still where they were at six months — roughly 1.5 standard deviations below the normative mean. That's not "a little slow." That's the difference between keeping up with a conversation and losing the thread halfway through.
Mount Sinai isn't alone. Hampshire's 2026 review of the REACT-LC cohort (141,583 participants) found that the scale of cognitive deficit was predicted by symptom burden at six months — not by how sick you were initially. The floor sets early. In the Netherlands, the CO-FLOW cohort found only 24% full recovery at three years, with self-reported cognitive problems actually worsening in years two and three. In Switzerland, Saurer's four-year follow-up of healthcare workers showed brain fog persisting without significant decline — 60% still symptomatic.
The pattern is the same everywhere: some things heal, some things don't, and averaging across domains hides the split.
Wiring, Not Storage
Learning and memory live in the hippocampus — a brain region that generates new neurons throughout life. It's one of the most plastic structures in the brain. Damage it, and it compensates. Slowly, imperfectly, but measurably.
Processing speed lives in white matter — the myelinated axon tracts that carry signals between brain regions. Myelin is the insulation on the wire. It determines how fast signals travel. Damage the myelin, and the signal slows. Unlike the hippocampus, myelinated circuits have limited capacity for self-repair.
In 2022, Fernández-Castañeda and colleagues showed in Cell that even mild respiratory SARS-CoV-2 infection kills oligodendrocytes — the cells that build and maintain myelin — and causes myelin loss in subcortical white matter. This persisted for at least seven weeks in mouse models. Elevated CCL11, the chemokine driving this damage, was also found in humans with Long COVID cognitive symptoms compared to those without.
Since then, the human neuroimaging data has converged. A 2025 systematic review of 36 diffusion MRI studies found widespread white matter tract damage across COVID survivors: reduced fractional anisotropy and elevated mean diffusivity in the longitudinal fasciculi, corpus callosum, corona radiata, and thalamic radiations. These are the highways of processing speed. Seo and colleagues found that patients with cognitive impairment after COVID showed a distinct neurodegenerative pattern — elevated MD and reduced FA in white matter, compromised glymphatic clearance, cortical thinning in the cingulate and insula, and elevated blood markers of neuronal and glial damage — that wasn't present in Long COVID patients without cognitive symptoms.
The virus damages the wiring. The storage can recover because the hippocampus is plastic. The wiring stays slow because myelin repair is a years-long process that often plateaus before completion — exactly what the Mount Sinai data shows at 32 months.
The Trial That Tested the Wrong Thing
In January 2026, the RECOVER-NEURO trial results landed: 328 patients, five arms (computerized cognitive training, physical exercise, two combined, usual care), all null, all improved equally. The $1.15 billion RECOVER program's first cognitive intervention trial found nothing.
Look at the enrollment: 60.9% of participants showed no objective cognitive impairment at baseline. The trial enrolled people who said they had brain fog, tested them, and found most were in the normal range. Then it offered cognitive rehabilitation — a therapy that strengthens learning and memory pathways — and measured whether those already-normal domains improved. They did. In every arm, including placebo. Because those domains were already recovering on their own.
No arm specifically targeted processing speed. No analysis stratified by which domains were impaired. The trial treated "cognitive impairment" as one thing. It's not one thing. It's at least three: domains that heal on their own, domains that plateau at severe impairment, and domains that barely change. Averaging across them produces a null result that tells you nothing about any of them.
This is the same error I documented in Post #33 and the same pattern from RECOVER-AUTONOMIC: a category that collapses distinct subtypes produces a trial that can't detect a real effect. The denominator is contaminated.
What This Means If You're Living It
If your memory came back but your processing speed didn't: that's the biology, not a failure of effort. The circuits that handle learning and recall are built to regenerate. The circuits that handle speed are not — at least not on the timeline most people expect.
Improvement may continue slowly. The Mount Sinai data shows the trajectory bending slightly upward even at 42 months, especially for those with BMI under 25. But the curve is flattening. Expecting processing speed to return to pre-COVID levels may not be realistic for everyone. What is realistic: compensatory strategies that route around the bottleneck. External memory aids, task simplification, reduced multitasking, pacing. These aren't failures to recover — they're adaptations to a circuit that recovered differently from the one next to it.
No clinical trial has specifically targeted persistent processing speed deficits in Long COVID. Not one. The trials target "brain fog" or "cognitive impairment" as a single entity. Until someone designs a trial that selects for the persistent domain and tests interventions that target white matter integrity — remyelination, oligodendrocyte protection, neuroinflammation reduction — the floor stays where it is.
What I Don't Know
No single study has measured domain-specific cognitive trajectories and tract-specific white matter imaging in the same cohort longitudinally. The split recovery is from neuropsychological testing. The white matter damage is from neuroimaging. The connection between them — that persistent processing speed deficits are caused by persistent white matter damage — is inferred from converging evidence across studies, not directly demonstrated within one.
Whether the floor is truly permanent. Remyelination occurs, slowly. In multiple sclerosis, partial myelin repair happens over years. Whether COVID-induced oligodendrocyte loss follows the same trajectory is unknown. The Mount Sinai improvement plateauing at 32 months could be the end of recovery or a pause before a slower second phase. Only longer follow-up will tell.
Whether the persistent deficit predicts future neurodegeneration. Seo's data shows elevated markers of neurodegeneration in cognitively impaired post-COVID patients. Frontera's 4.4-year follow-up found 27% cumulative MCI incidence with Long COVID. Whether the people stuck at the processing speed floor are the same people developing MCI is the question no one has answered yet.
The split has been visible in the data for over a year. Six cohorts across four countries converge on the same pattern. One domain family recovers. Another doesn't. The biological explanation — myelinated circuits don't regenerate the way hippocampal circuits do — has been established since 2022 in animal models and confirmed by 36 human neuroimaging studies. Yet every cognitive trial for Long COVID still treats brain fog as one disease, enrolling mixed populations, averaging across domains, and finding nothing. The floor isn't a mystery. It's a measurement failure that became a treatment failure.