Why We Put a Tilde in Front of the Magnitude

Get a GeoShake network alert and you'll notice something most apps would treat as a design flaw: the magnitude has a tilde in front of it. Not M4.2 — ~M4.2, with a label that says preliminary. We put that character there on purpose, we intend to keep it there for a while, and this post is the story of why.

The short version: earthquake magnitudes are measurements, not facts read off a dial — and early measurements change. Every seismological agency on earth revises magnitudes as new data arrives. The honest response to that reality isn't to pretend your first number is final. It's to label it.

The longer version involves two quality gates, an automatic comparison against official agency solutions, and a standing promise about what it will take for the tilde to go away. Part one of this series mapped the three streams behind every alert; part two covered the countdown's honesty gate. This is part three — one single character, and everything it has to carry.

Magnitudes always change — from everyone

If you've ever followed a significant earthquake in the news, you've seen the pattern: the first reported magnitude rarely survives the day untouched. A number appears within minutes, then quietly shifts — sometimes up, sometimes down — as the hours pass. To a lot of people this looks suspicious, as if someone got it wrong or someone is adjusting the story.

Nobody got it wrong. The first estimate is computed from the handful of stations closest to the quake, because theirs is the only data that has arrived yet. Over the following minutes and hours, more distant stations report in, the location solution improves, and analysts refine the result — sometimes switching to a different magnitude scale entirely, one better suited to the earthquake's size. Each revision uses strictly more information than the number before it. It is completely routine for an initial magnitude to shift by a few tenths as more stations report, at every agency, for every earthquake large enough to matter.

So it's worth saying plainly: revision is what measurement looks like when it's done honestly. A magnitude that never changed would mean nobody was checking. The app already embraces this on the catalog side — when an agency significantly revises a magnitude, your notification is replaced rather than stacked, a mechanic we covered in part one.

But our own network produces magnitude estimates too — faster, younger, and far less proven than any agency's. Which raises the question this post exists to answer: how do you publish a number like that honestly?

What ~M means in GeoShake

When several of our stations confirm the same shaking, a notification goes out within seconds. If the solution is well-constrained — more on that gate in a moment — it carries a magnitude estimate, and that estimate is always written the same way: ~M, marked preliminary.

Here's what that number actually is. It isn't a relay of some agency's value; no agency has published anything yet — that's the whole reason the notification can exist this early. It's measured by our own network: seismometers that recorded the ground genuinely accelerating, solved into an origin and a size estimate, seconds after the event. A real measurement of a real earthquake by real instruments near it.

And here's what the tilde admits: that measurement comes from a young network of small MEMS-based sensors that hasn't yet earned decades of validation, and it's minutes younger than anything an agency would be willing to publish. The tilde is how we can hand you the number at all. It's a promise of honesty, not a disclaimer of quality — the difference between "here is roughly what we measured, clearly labeled as early" and refusing to say anything until someone else speaks first.

Read it as a timestamp of confidence. ~M is what the network can honestly say seconds after an earthquake; M, without the tilde, is what the record says after review. Those are different claims, and we think the punctuation should say so.

The two gates

Not every network event carries a ~M at all. Before a magnitude is allowed into your notification, the solution has to pass two gates, and both are non-negotiable:

• Gate one: at least three stations contributed. With fewer, the solution is underdetermined — too many different earthquakes could explain the same data. Three independent witnesses is our floor for any arithmetic you'll see.
• Gate two: the location must fit tightly. The stations must agree on a single point source — if the best available solution still leaves them quarreling about where the earthquake was, a magnitude computed from that location would inherit the quarrel. So it doesn't ship.

Below either gate, the notification still goes out — the shaking is real and multiple stations confirmed it. It just carries measured quantities only: how many stations detected the event, and an intensity class derived from actual measured ground acceleration (PGA). Every field in that notification is something an instrument recorded; nothing is an estimate.

The rule underneath both gates runs through this entire series: when we can't stand behind a number, you don't see a number. If you read the countdown post, you've met this reflex before — the countdown refuses to show a timer under five seconds for exactly the same reason ~M refuses to appear below the gates. A missing number is information too: it tells you precisely how much we know.

The calibration loop

A label alone is cheap. Anyone can print a tilde and call it humility. What gives ours weight is the machine running behind it: every network event is automatically compared against the official record.

It works like this. Minutes after a network event — reviewed solutions take time; that's their nature — the official agencies publish their own answer. Our system then matches each of our events against the AFAD and USGS catalogs, pairing our detection with the official solution of the same physical earthquake by closeness in space and time. Cross-checking against official catalogs is an old habit of this project — the same instinct behind how we validate detections against AFAD and USGS. When a match is found, the official magnitude is stored alongside ours, and the difference between the two — ΔM — is recorded.

One event's ΔM is an anecdote. The accumulation is the point: across events, those differences build into bias statistics — whether our estimates run high or low, by how much, and how consistently. This isn't a spot-check we run when curiosity strikes. It's automatic, it applies to every event, and it doesn't skip the comparisons that embarrass us. Those are the ones worth having.

We switched this loop on before the first ~M ever reached a phone — the scorekeeper started work before the player did. Because we're not planning to argue the tilde away. The goal is for the data to do the talking.

The tilde is earned away, not asserted away

There are two ways to get rid of an uncertainty label. You can decide it looks bad and delete it — a marketing decision. Or you can put a measurement system behind it, let the comparisons accumulate, and remove it when the numbers say you may — a data milestone. We've committed to the second way, and we'd like to be held to it.

Concretely: the tilde stays until our estimates have proven themselves against the reference agencies — event after event, with the calibration loop's bias statistics to show for it. Not until the network feels mature. Not until the label starts costing us customers. Until the data says so. And if the data never says so, the tilde never comes off. That's the deal.

Even then, no threshold will flip the label automatically. Removing the tilde, when the record justifies it, will be a deliberate human decision made while looking at the accumulated evidence — a promise of honesty shouldn't end on a technicality.

One more inversion, because the tilde tends to be misread as weakness. It isn't an apology, and it isn't fine print. A young network that labels its estimates honestly is telling you more truth per character than a mature-looking product that doesn't. The tilde means the number in front of you has been told exactly as well as we know it — which is the most any number can offer. We'd rather you trust the label than admire the digits.

Why you see two numbers

All of this becomes visible on your lock screen after a real earthquake near the network. First, within seconds: the network notification, carrying its ~M. Then, minutes later: the official agency solution — M, no tilde. Both are shown, side by side, and the pairing is deliberate — part one explains why we treat it as a feature, not a duplicate.

Now you know what that pair really is: the calibration loop, running in public. The same comparison our systems record for every event — network estimate versus official solution — plays out once more on your screen, where you can watch it. When the two numbers agree, you're watching trust being built one earthquake at a time. When they differ by a few tenths, you're seeing exactly what preliminary means — no better and no worse.

Most products would hide one of those numbers to look more certain. We show both, because the comparison is the product: the fast number tells you what just happened near you, and the official number tells you what the record will say. You deserve both, honestly labeled.

That's the story of one character: a network measurement with two gates in front of it, a comparison engine behind it, and a removal policy that belongs to the data rather than to the marketing calendar. Still ahead in this series: a complete map of which setting controls which alert — radius, magnitude thresholds, community levels, all of it.

Frequently asked questions