You optimise the model.
Nobody benchmarks the operator.
Every hour here goes into making machines think better. The thing actually doing the deciding runs on four hours of sleep and its eleventh context switch since lunch.
Ranked by how much evidence stands behind them
Sleep is the consolidation step
Memory does not form while you read. It forms overnight, when the day is replayed and pruned. A short night is not a tired morning; it is a batch job that never ran. This is the single lever with the least argument behind it and the most people ignoring it.
Attention is a budget, not a trait
Nobody has a bad memory at 9am and a good one at 4pm by accident. Context switching, unfinished loops and open tabs draw down the same pool that focused work needs. Most "cognitive decline" complaints in knowledge work are load problems wearing a medical costume.
Inputs move slowly and unevenly
Supplementation is the loudest and least reliable of the three. Some ingredients have real trials behind them, most have a mechanism story and a marketing page. The honest position is to treat the category as a small optional lever, and to read what the evidence actually says before paying for a subscription.
What a cognitive supplement can and cannot claim
The category sells certainty it does not have. A formula is not a drug: it is not required to prove an outcome before it goes on sale, and most marketing pages lean on a mechanism story rather than a trial. That does not make every product worthless. It means the burden of reading sits with the buyer.
Two things are worth checking before any of them: what the refund window actually is and whether the claim is about an ingredient or the finished product. Studies on a compound do not transfer to a blend containing it.
Anything below is a commercial link. It is listed because the vendor publishes a long refund window, not because the outcome is established, and nothing here is medical advice or a substitute for asking a doctor.