Scoring methodology

Every number on this site is a claim about someone's livelihood. If we cannot explain how we got it, we should not display it. This page is that explanation.

Principles

Deterministic core. All scores are computed by declared arithmetic over stored data. No language model produces a number here.

Every score returns its breakdown. A score is not a single figure — it carries each component, its weight, its contribution and its source. The explanation you see is generated by the same arithmetic that produced the score, so the two cannot drift apart.

Confidence is shown, never hidden. A thin profile produces a low-confidence result and we say so. Below a threshold we show no number at all, because false precision is worse than an honest gap.

Versioned weights. Weights are pinned to a model version. Tuning means publishing a new version, not quietly editing a live one.

Weights

Career Risk (career_risk@1.0.0)

Factor	Weight	Direction
Role automation exposure	0.22	Raises risk
Technology disruption	0.15	Raises risk
Skill currency	0.18	Raises risk
Market demand	0.15	Protective
Industry disruption	0.10	Raises risk
Employer stability	0.08	Raises risk
Internal mobility	0.12	Protective

On direction. Market demand and internal mobility protect you rather than threaten you. Showing them in the same column as risk factors would misrepresent them, so they push the other way on the axis and are coloured differently. This is a correctness requirement, not a styling choice.

Exposure is not risk. A role can be highly exposed to automation and still be safe if that exposure augments the work rather than replacing it. We separate the two and dampen exposure by the augmentation share before it enters the score.

Data sources

• O*NET — occupations, tasks and skill requirements. Also the substrate for published task-based automation exposure research.
• ESCO — multilingual skills and occupations taxonomy, mapped to ISCO-08.
• ISCO-08 — the international occupation spine that makes cross-country comparison possible.
• National labour statistics — wages and employment projections.
• Live job postings — current demand signal.

Known limitations

We publish these rather than waiting to be asked.

• Exposure is occupation-level, not individual. Two people in the same role with very different day-to-day work receive the same input.
• India salary coverage is thin. Open data sources are US and EU weighted. Because Indian professionals are a primary audience, this is our largest known gap, and it is why we show ranges rather than point estimates where the sample is small.
• Transition suggestions are currently derived, not observed. They are computed from skill overlap and are hypotheses about what is reachable, not records of what people actually did.
• Employer stability is a weak proxy. It is inferred from company size alone. We show it anyway, labelled as weak, because hiding an input that moves the number would be worse.
• Certification lift is self-reported. Published survey figures come from people who chose to respond, and people who benefited are likelier to answer. We discount reported lift by 30% before using it.

Your rights

These scores are advisory to you, about you. We do not sell individual risk scores to employers. Any future workforce product operates on aggregated, de-identified cohorts only.

Every result is contestable: the full breakdown is always available, and you can correct the inputs and recompute. Nothing here is a decision about you, and nothing here should be treated as one.

The calculators on this site run entirely in your browser. Your inputs are not transmitted to us.