companion · the receipts

The Science Behind the Window

"Our Life Context Window" is opinion — my lived observation of time seeming to speed up, told in the vocabulary of the machines I work with. But I'm a technologist, and the logic-driven part of me wanted to know: has anyone said this before, and does any of it hold up? So I sent research agents out to check every claim against the published literature. This is what came back — the support, and honestly, the places where my metaphor bends.

Does time really speed up as we age?

Yes — but more modestly than the folklore says. Studies asking people of all ages how fast time feels find the effect is real but small, and it shows up mainly in one place: how fast the last decade felt. Week to week, a 70-year-old and a 20-year-old report time passing at about the same speed; how busy you are predicts felt speed better than your age does (Friedman & Janssen, 2010; Janssen, Naka & Friedman, 2013; Wittmann & Lehnhoff, 2005). So the acceleration is real — it's just retrospective. It lives where the essay says it lives: in the playback, not in the moment.

The popular explanation — each year is a smaller fraction of your life — dates to 1877 (Paul Janet) and turns out to fail when tested directly. The account researchers now favor is the one the essay's compression metaphor maps onto: remembered time tracks how many new, distinct events got encoded. Childhood summers are packed with firsts, so playback runs long; routine adult years compress toward nothing (Hammond's "holiday paradox"). And a 2026 study sharpened this in a way I didn't expect: it's not how full the window is that predicts the feeling of acceleration — it's how much capacity you have left to encode new things. The intake slows; the disk doesn't just fill. I'd have written that into the essay if I'd known it going in.

Does emotion decide what gets kept?

This is the best-supported claim in the essay. Four decades of work, much of it from James McGaugh's lab, shows that the stress chemicals released at an emotional peak — adrenaline, cortisol, acting through the amygdala — strengthen the consolidation of whatever just happened. The causal proof is elegant: give people a beta-blocker that mutes the adrenaline signal, and the memory advantage for emotional events disappears (Cahill et al., Nature, 1994). The chemistry really is doing the choosing. One nuance: emotion biases what's kept rather than strictly determining it, and it preserves the vivid center of an event at the cost of the edges.

The slow-motion thing is real — but it's memory, not perception

David Eagleman's group dropped volunteers 150 feet into a net wearing a wrist display flickering too fast to read. If fear actually slowed perception, they'd have read it. They couldn't — but afterward they remembered their own fall as ~36% longer than it was (Stetson, Fiesta & Eagleman, 2007). Terror doesn't slow time; it writes memory at a higher bitrate, so the playback runs long. Which is to say: a traumatic event is an uncompressed segment in the window. I used the phrase "recited as happening in slow motion" in the essay before I knew this experiment existed — the recitation, not the moment, is exactly where the science says the slow motion lives.

The honest caveat: what persists in traumatic memory is the vividness and the felt certainty — not the accuracy. Even flashbulb memories drift like ordinary ones while confidence stays high (Talarico & Rubin, 2003).

Is forgetting really index loss, not data loss?

Substantially, yes — with one concession. Psychology has long distinguished a memory being gone from being unreachable (Tulving's cue-dependent forgetting). Smells and songs really are privileged keys: odor-cued memories are measurably older, more emotional, and more "relived" than word-cued ones. And the most remarkable support comes from neuroscience: in mice with induced amnesia, the physical memory trace — the engram — still exists, and stimulating it directly brings the full memory back even though every natural cue fails (Ryan et al., Science, 2015). Intact data, broken index, literally.

The concession: the field no longer believes nothing is ever erased. There is an active decay process — some of it running during sleep — that really does destroy traces (Hardt, Nader & Nadel, 2013; Ryan & Frankland, 2022). Most of the drive may still be there behind the lost boot sector. Not all of it. Some files really do rot.

Copy of a copy of a copy

The essay's claim that retelling a memory reseeds a blurrier facsimile is not metaphor — it's mechanism. Retrieving a memory returns it to a chemically unstable state from which it must be re-stored; block the re-storage and the memory is gone (Nader, Schafe & LeDoux, Nature, 2000). Every playback reopens the file for writing. Elizabeth Loftus spent thirty years showing what gets written: post-event information blends in, details migrate, whole scenes can be implanted. In the ten-year study of 9/11 memories, recollections were only about two-thirds consistent after the first year — while confidence stayed sky-high (Hirst et al., 2015). The copies degrade, and the owner of the copies is the last to know.

Do other people really carry your memories?

Half yes, half humbling. Transactive memory is a real, well-studied phenomenon — couples and close groups genuinely divide the labor of remembering and can retrieve more as a system than either member alone (Wegner, 1991 onward). Cross-cuing works; other people's recall does correct yours. But the same literature carries a warning the essay only gestures at: groups also converge on an edited, trimmed, shared version of events, and what a storyteller leaves out of a retelling fades from the listeners' memories too (Cuc, Koppel & Hirst, 2007). Seeding your memories to others preserves a version of them. And the version the group keeps becomes the version.

The back burner

Handing a lost name to the subconscious and having it surface at 2 a.m. is a documented phenomenon: a meta-analysis of 117 studies confirms stepping away from a retrieval or problem genuinely helps (Sio & Ormerod, 2009), and diary studies record memories popping up unbidden days later. What science hasn't settled is my favorite part of the metaphor — whether anything is actually searching down there. Competing explanations include simple release from a wrong fixation, and dormant problems waiting for an environmental cue. The sub-agent scouring the drive is my picture of it, not the field's consensus. The break works; nobody agrees on why.

Who got here first

I checked, because I didn't want to rebuild something already built. The pieces of this idea exist separately: technologists routinely compare working memory to context windows, and "why time speeds up as we age" fills books — Draaisma's Why Life Speeds Up As You Get Older is the scholarly one, Hammond's Time Warped the accessible one. But the full mapping — a life as a context window that fills, autocompresses, loses its index, and ends in a handoff of compressed context to the next generation — doesn't appear to have been published. The closest piece I found runs the metaphor the other direction, using human aging to explain AI sessions. So the essay stands on well-trodden ground and, as far as diligent searching can tell, builds something of its own on it.

I'll also name the strongest objection, because it deserves the light: every era casts memory in its newest storage technology — Plato used a wax tablet, the Victorians a palimpsest, the last century a tape recorder, and critics (fairly) warn that we keep mistaking the era's furniture for the mind's mechanism (Epstein, "The Empty Brain," 2016; Draaisma, Metaphors of Memory, 2000). Plato, it turns out, also compared memories to birds in an aviary — you own them, but you can't always catch the one you want. Index loss, four hundred years before the codex. My defense is that the context window earns its keep precisely where the tape recorder failed: it assumes compression, loss, and reconstruction. The tape recorder promised playback. The window only promises what it kept.