How Quell thinks

We simulate your future, we don’t predict it

There is no single “what will happen.” So instead of one forecast, Quell plays out your life many thousands of times — a wide range of plausible futures where markets, your job, your costs, and unexpected events unfold differently each time. We then count what happened across all of those runs. A “15% chance of running out of buffer in five years” literally means it happened in fifteen out of every hundred simulated futures. This is a well-established approach to risk; we apply it to your whole financial life rather than just a portfolio.

We take bad years seriously

Real life is not a gentle bell curve. Crises cluster: markets fall while jobs disappear and credit tightens, all at once. Quell’s simulations are built to reflect that — rare, severe, correlated bad stretches are represented, not smoothed away — because those are exactly the situations a risk tool exists to warn you about. The result is a more honest picture of the downside than a simple average would give.

Every number comes with a range

A point estimate pretending to be precise is a lie. Quell always shows a band around each number — a range that reflects how much we genuinely know. When your inputs are thin, or the underlying evidence is young, the band is deliberately wider. A wide band is not a bug; it is us being honest about uncertainty instead of hiding it.

Grounded in real, public evidence

The world Quell simulates is calibrated to real, reputable, public data — for example, public economic data from the U.S. Federal Reserve’s FRED database, government labor statistics, and peer-reviewed research on how automation and AI affect different occupations. Where you see a number, you can open a panel that names the sources behind it. We update this evidence over time as new public research is released.

The numbers come from the engine — never from a chatbot

Quell uses AI to read facts and to explain results in plain English, but every figure you see is produced by our simulation engine, not by a language model. The AI may gather public facts (such as an economic figure or a public signal) and turn an engine-computed result into a readable sentence — but it is never allowed to invent or alter a number. We verify that the words match the math. This boundary is the heart of how we keep Quell trustworthy.

You stay in control of the inputs

Quell does not connect to your bank. You decide what to share, in coarse terms, and you can edit it and re-run at any time to see how your risk moves. Better inputs make for a sharper reading; the tool is transparent about what it does and does not know.

One band, two kinds of doubt

A simulation is uncertain for two distinct reasons: finite sampling (we ran thousands of futures, not all of them) and parameter doubt (the world’s dials — how fat the tails are, how exposed your job really is — are themselves estimated). Our headline bands carry both: every full reading re-runs the simulation under many plausible parameter settings and folds the per-run sampling error into the spread, reported at a consistent 90% level. When a band is wide, that width is itself information about what we don’t know.

People differ in ways we can’t see

Two households with identical spreadsheets do not face identical layoff risk. The engine models that unseen spread directly: each simulated future draws its own job-loss propensity around your profile’s estimate, so the tails reflect the realistic case where your personal hazard turns out worse — or better — than the averages suggest. When live evidence arrives (a confirmed layoff round at your employer, a sector drawdown), it updates that belief the way evidence should: gradually, with corroboration counting for more than a single headline.

Questions answered from the same futures

When you ask something like “what if I lose my job?”, Quell doesn’t guess — it looks inside the futures it already simulated for you, finds the ones where that event happened, and reports what followed, along with how common that event was. The answer arrives instantly because the work was already done, and it is consistent with your reading because it is your reading.

We test ourselves in public

Every full reading files a falsifiable one-year prediction at the moment it is computed. Once those twelve months pass, we score the prediction against what actually happened and publish the running scorecard — see backtesting. We publish the absence of evidence too: until enough predictions have matured, that page says so instead of dressing noise up as validation.

Honest about limits

Models are simplifications. They rest on assumptions and historical patterns that can be incomplete or wrong, and the future can always surprise us — your actual outcome will differ from any estimate. Quell is a lens for thinking about risk and trade-offs, not a crystal ball and not advice. For decisions, talk to a qualified, licensed professional.

We keep the specific mathematics, parameters, and implementation that power the engine proprietary. This page describes our approach, not the recipe.