By the time Tuesday's meltdown happened, the signs had been there for nine days. The night log showed Maya waking at 3 a.m. four nights running. The lunch DSP noted she'd left most of her food twice that week. Her BCBA had logged a new behavior — chewing her sleeve raw — and filed it as minor. Her mom had moved one medication to mornings on a doctor's advice. Each note was reasonable. Each writer did their job. No one saw all five at once — so no one saw the shape they made together.
That's the thing about a crisis: it almost never arrives out of nowhere. It announces itself, quietly, in the daily notes. The question is whether anyone is positioned to read across them in time.
The signals that tend to come first
Hard episodes have a vocabulary, and it's surprisingly consistent. Sleep is usually the loudest — two or three short or broken nights in a row is one of the most reliable precursors there is, and it's almost always logged by a different person than the one who'll handle the fallout. Appetite shifts the same way: eating less, or suddenly refusing textures that were fine last week.
Then there's behavior that's new or escalating — not one bad moment, but the same small thing creeping up in frequency or intensity across several days. A recent medication change belongs here too; a dose moved, added, or stopped can take a week to show its edges. And sensory load that stacks — a noisy classroom, a schedule change, a dentist visit, a houseguest — each survivable alone, dangerous in a pile.
None of these is a crisis. Every one of them is a sentence in the paragraph that leads to one.
Why the pattern hides
Here's what makes this genuinely hard, and why it isn't anyone's fault: the signals are scattered across three axes at once.
They're spread across different people — the night caregiver, the day DSP, the therapist, the parent — and no two of them read each other's notes. They're spread across different days, so even the same person, looking only at today, sees one data point and not a trend. And they're spread across different domains — sleep in one place, food in another, behavior in a third, meds in a fourth — so the link between "barely slept" and "chewed her sleeve" never gets drawn, because the two facts never sit on the same page.
Each person holds one corner of the picture. It only exists when the corners are laid side by side — and in most care setups, they simply never are.
What catching one looks like
Catching a pattern in time is undramatic, which is why it's easy to undervalue. It looks like someone noticing on day three — not day nine — that the sleep notes and the appetite notes are both trending the wrong way, and that a med change landed just before both. It looks like a short message to the team: something's building, let's ease the schedule this week and watch closely.
And then, often, nothing happens. The hard week doesn't come, because the houseguest was rescheduled, the dose was revisited, the demands came down for a few days. The crisis you prevent is invisible — there's no incident report for the meltdown that didn't occur. That's the strange grief of doing this well: the wins don't announce themselves. You just get a quieter month, and rarely know which catch bought it.
The discipline: read across, not just down
So the skill worth building isn't sharper instincts in the moment. It's the habit of reading across — across time and across domains — instead of reading today's note only for today's plan.
Concretely: whenever something feels subtly off, scan the last seven to ten days as a block — sleep, food, behavior, meds, and anything new in the environment, all at once. You're not looking for the dramatic entry. You're looking for two or three quiet ones pointing the same direction — that's the signature. Ask whether anything changed about a week ago, because consequences lag causes. And treat a note from someone else's shift as part of your picture, not just theirs.
The warning is almost always already written down. The work is making sure someone can read across all of it in time.
A hard episode is rarely a surprise to the data. It's a surprise to the people, because the data was split among them and never assembled. Sleep slipping, appetite changing, a behavior creeping up, a med that moved, sensory load stacking — read alone, each is forgettable. Read together, across a week, they're a sentence finishing itself. The whole job is to read them together before it ends.
This is exactly the seeing Trellis is built to give a whole care team — patterns surfaced across notes, goals, sleep, behavior, and medications at once, with the actual entries behind them so you can check the evidence yourself, not just trust an alert. It doesn't replace your judgment; it hands every person the corners of the picture they were always missing. See how Trellis surfaces patterns →