Infinto · Method
We tell people their skills decay, so we owe them the workings. Infinto's score rests on a number we forecast rather than observe — that deserves a page rather than a footnote, including the parts where the method is weakest.
Every task in a logged week is scored into one of three columns. The test is not difficulty — plenty of hard work is automatable and plenty of easy work isn't.
Half-life is the time until half of today's task list has moved one column to the left. The term is borrowed from physics, where it describes constant proportional decay: a fixed fraction of what remains is lost each period, never a fixed amount.
We borrow it because skill erosion behaves the same way. Work doesn't disappear at a steady number of tasks per year — it loses a share of whatever is left, and the residue is always the harder part. One figure describes the whole curve, and the curve never quite reaches zero.
Radioactive half-life is a physical constant. Skill half-life is a forecast, and the rate itself changes the week a new model ships — a figure of 2.4 years today could be 1.4 next year. It also can't be observed directly; it's inferred from task-level scoring, which means it inherits every error in that scoring. Treat it as a planning number with a wide error bar, not a measurement.
Two different things on this site carry numbers, and they are not the same kind of number.
| Where | What it is | Confidence |
|---|---|---|
| Half-life index Homepage instrument |
An illustrative model of six common roles, built to show the shape of the argument. It is not survey data and no employer's actual tasks were scored to produce it. | Illustrative |
| Short audit Eight questions |
Derived entirely from your own description of your week. Self-report, so it reflects how you see the job rather than what a task log would show. | Indicative |
| Full audit Three weeks |
A real fortnight of logged tasks, scored against current model capability by a practitioner, with the scoring shown to you so you can dispute it. | Evidence-based |
The first two exist to make the argument legible. Only the third should be used to make a decision about someone's career or a team's budget, and we say so on the results screen rather than in a disclaimer.
Partly, yes — that's the honest finding for some roles, and softening it wouldn't help anyone. But the audit's output is a direction, not a verdict: every band comes with the nearest adjacent compounding work and a route into it. A diagnosis with no treatment would be cruelty dressed as insight.
Tool skills sit in the supervised column, which is the one with the shortest remaining life. They're worth having and they date fast. We teach the column that appreciates instead — which happens to be harder to teach and harder to sell, hence the diagnostic first.
Practitioners in each domain test representative tasks against current frontier models on a rolling basis, and the scoring rubric is re-cut when capability moves. This is the weakest joint in the method: it is expert judgment, not benchmark data, and it lags releases by weeks.
The full audit requires real work data, which is the most sensitive thing we handle. Logs stay within your own tenancy, are scored under agreement, and are not used to train anything. The short audit on this site never leaves your browser at all.
An audit that becomes a redundancy instrument stops producing honest task logs within one cycle, which destroys the thing that makes it work. Organisation engagements are scoped so individual bands are not returned to management — the function lead receives coverage, not names.
Start
Eight questions, about four minutes, nothing stored.