A Day, Well Written
“If we had a keen vision and feeling of all ordinary human life, it would be like hearing the grass grow and the squirrel’s heart beat, and we should die of that roar which lies on the other side of silence. As it is, the quickest of us walk about well wadded with stupidity.” — George Eliot, Middlemarch
Perception requires selection. To attend to everything — every sound, every passing face, every thought half-formed and dropped — would not enrich you; it would undo you. Eliot’s “wadding” is not a defect of the mind but the condition of having one: we focus, and we forget, and the forgetting is as human as the focus.
But the same wadding that keeps us sane also keeps us in the dark. We miss what we never thought to notice; we repeat what we never caught ourselves doing. This is the problem Virtues was built around — what we call Personal Intelligence: the slow work of discernment and introspection, of seeing the facts you missed, the blind spots you kept, the virtue or the vice you were quietly habituating all along.
To help with that, Virtues writes your day, and it writes it twice:
- The Event Timeline: a granular, factual account of what happened.
- The Biography: a written paragraph that aims not at completeness but at salience: what, of everything that happened, stood out and is worth keeping.
One tells you what happened, and how unusual it was. The other tells it back to you in language you’d want to read — and that difference is the rest of this essay.
The Event Timeline
Your life has measurable signals: a heart-rate sample, a location ping, a calendar block, a transaction — thousands of them, each sorted the moment it lands into an ontology the typed category system every signal is filed under as it arrives — health_heart_rate, finance_transactions, location_visit. The raw data is never a meaningless heap; it is pre-classified by domain before a single event is drawn from it.
Event Synthesis
Resolving those signals into events is the first pass. The model reads a span of the day the way a detective reads a room, asking not what each signal is but what they add up to, and returns, for each event one of the 8–16 distinct blocks of time in a day over which the context holds steady, ending at the next meaningful shift in that base pattern, two things: a structured W5H (who, what, where, when, why, how) and one plain sentence that states it.
An unhurried early coffee with Maya at the Watch House, caught before the South Bank crowds.
- whoMaya Bell
- whatan early coffee
- wherethe Watch House
- when07:42
- whybefore she moves
- howunhurried, warm
Most of those events are unremarkable, because most of a day repeats every other day: the commute, the lunch, the evening scroll. A timeline of fourteen accurate events still won’t tell you which one mattered.
We could have gone a step further and let the machine decide which events were good ones. We deliberately don’t. A system that grades your days — this one well spent, that one wasted — is quietly deciding how you ought to live. We want a mirror, not a judge: it records what happened and surfaces what was unusual, and leaves up to you what any of it was worth.
Event Novelty: Global & Local
So what the timeline scores is novelty, not value — not whether an event was good, but whether it was unusual for you.
Measuring that starts by turning each event into numbers. An embedding model reads the event’s W5H and summary and returns a vector — a point in a space where similar events fall close together, so your morning runs gather in one region and your commutes in another. To score it, the timeline measures how far this point sits from the crowd of everything you’ve logged before — the recent days weighing most and older ones decaying away on a smooth curve, with no hard cutoff, so the baseline is always your present life and not the person you were years ago. That distance comes back as a z-score: how far from your own average, in your own units, because a heart rate that’s unusual for you is ordinary for someone else.
And it measures distance two ways, because there are two ways to be unusual. The first is global: a new kind of event has entered your life, like a first kayak, a city you’ve never been to, a name that has never come up. Its point lands where you have no others; nothing you’ve logged sits near it. The second is local, and it’s usually the one you actually remember: an odd instance of an ordinary thing. You lift every morning, but today you ran. You take the same route to work, but today your heart sat far above its usual. Against the whole of your life the event is unremarkable, just another workout, but against its own kind it sits at the very edge. A wearable logs both days as “workout” and notices neither.
Each dot is one of your recent events; events in similar contexts fall close together, and the soft clusters are the routines your life keeps.
Global — a new kind of event, far from every cluster. Local — an ordinary kind at the edge of its own cluster: the unusual instance you actually remember.
Entity & Topic Resolution
The same passes that resolve events also pull out the day’s entities (the people, places, things, and organizations it touched) along with its topics. Each one is parsed by the model as it reads and matched against what you already have: a confident match links to the existing record (the Maya in this morning’s coffee is the Maya from a hundred days back), and anything it can’t place is kept as plain text rather than forced into a wrong link. Resolved this way, entities and topics become the threads that run between events, the ones the biography later uses to write one day against all the others.
So the first pass ends with a full, scored account of the day — what happened, and which parts of it were unusual. It is accurate, and it is everything a machine needs for information retrieval. It is also not something anyone would want to read. That is the second pass.
The Biography
The biography is the day written for you — not by you. There’s no page to face at the end of a tired evening, nothing to keep up, no streak to break. It writes itself each night from what the day already left behind, and asks nothing of you in return. That is the intent everywhere in Virtues: it should run the way memory itself does, in the background, without maintenance.
It does not start from scratch. The timeline has already done the arithmetic — every event scored for novelty, the routine marked routine, the unusual flagged and ranked — and the biography inherits all of it. Before it writes a word it knows which three or four of the day’s fourteen events carried weight, and which were the rest. You never see a z-score. You read a paragraph that lingers exactly where the day did. The craft is mostly in what it leaves out: fourteen become three sentences, the morning on the water gets its clause, and the trip to the pharmacy does not.
And because the same night resolved the day’s people, places, and topics (Maya, the South Bank, the kayaks, each recognized once and then wherever it returns), the biography can write a day against all the others. It knows Maya is not a stranger but a thread running back through your weeks; that the kayaks were the first in months; that before you move is the kind of line that turns out to matter. A day read by itself is just a day. Read against the arc, it has motifs: the things that keep recurring, the one that is quietly ending. Those are most of what you’d want kept.
Today began on the South Bank, just after seven, where you met Maya1 for an early coffee before the crowds. You walked the river east, rented kayaks below Tower Bridge, and spent the better part of the morning on the water. By noon the weather had turned, so you caught the early train home — quiet, sun-tired, content.
- where South Bank
- where Tower Bridge
- what kayaking
- topic a coffee date
- 07:42 Coffee with Maya Watch House · South Bank
- 09:15 Rented kayaks below Tower Bridge
- 12:10 Train home Northern line · sun-tired
- … +11 more events
This is where the writing has to know its place. A machine can measure what stood out; it can’t measure what it meant — and one that pretended to would only be flattering you. So it tells you the morning held a goodbye, and stops there. That judgment is yours: the machine keeps the day so that you can keep the meaning.
And if you want to supply what the day couldn’t see (why you really left early, what the coffee was actually about), there’s a journal feature for that. It’s the one corner of Virtues you keep by hand, and the whole point of everything around it is that it’s the only one.
What the biography does not choose is how it sounds. Some people want their days in plain report; others want them in the register of a novel — Austen’s dry attention, Lewis’s warmth, whatever voice makes a life read like a life. That is yours to set; like everything at Virtues, the system prompt is open. But voice is the only thing you choose: honesty isn’t a setting underneath it. Whatever register you pick, it stays inside what the day can actually support.
A Day, Well Written
So the night ends with two accounts of one day, written in two hands that want opposite things.
The events are for the machine: terse, complete, W5H-faithful, built to be embedded and retrieved, indifferent to beauty because beauty would only get in the search’s way. They are the fidelity layer, the part that makes sure that in ten years the day can still be found, exactly, down to the 07:42.
The biography is for you: selective where the events are complete, warm where they are flat, and careful never to claim more than the day will bear. It is the insight layer, the part that makes sure that when you come back to the day, it is worth reading. The events guarantee the Tuesday can be recovered. The biography is the reason you’d want to.
Neither could do the other’s job. A biography indexes terribly; you cannot run a clean search against prose that romances its own contents. A timeline reads terribly; no one recognizes their life in a filled-out form. The mistake is to build only one of them — a quantified-self log that remembers everything and means nothing, or a fond account of your life you could never find your way back through. The day needs both hands, because the day has both readers.
“Well written,” then, is a smaller and stricter thing than it sounds. It does not mean ornate, and it does not mean flattering. It means the day is held accurately enough to be found and told truthfully enough to be trusted, stopping exactly where the evidence does. That last restraint is the whole safety of it. A machine that will write your life is a short step from a machine that will decide it: one that tells you what your day was worth, how you ought to have felt, which version of you to believe. The danger was never that the inference would be too weak; it is that it would be too willing, eager to find meaning you never put there, to round a dull Tuesday up into a good one, to do your discernment so smoothly that you slowly stop doing it yourself. Everything here is built to keep the narrator on its leash: novelty instead of judgment, the score you never see, the verdict it refuses to render.
Eliot’s warning was that a true vision of ordinary life would be a roar on the other side of silence, and that we live only by going well wadded, attending to almost nothing. Salience is that wadding made deliberate — not numbness, but chosen attention, and chosen by you. A day well written is one that attends to what you attended to, says no more than it knows, and hands the meaning back. A life is a great many days. It is worth having them well written.