Evidence first. Numbers only with support.

The method is openly described; each value still requires its own evidence

Method definition, calculation, calibration and publication are separate steps. A technically computable value is therefore not yet a public claim.

• 01Task-based work methodologyActive
• 02Methodology statusActive contract
• 03Public impact valuesCurrently withheld
• 04Cross-occupation rankingNot used
• 05Methodology LabAccess controlled
• 06Historical calculationsRemain documented for comparison but do not determine a current public metric.Archive only

Five areas, clearly bounded subjects

The analysis area determines which identity, relationships and comparison data are relevant. A catalogue record identifies a subject; it does not demonstrate the impact of AI on that subject. Evidence must fit the question, version, region and time.

• 01WorkRoles, occupations, tasks and concrete workflows. Occupation catalogues provide identity and task context in this area.Distinct analysis context
• 02EducationStudy programmes, learning offers, curricula and skill requirements. No admission decisions about individuals.Distinct analysis context
• 03OrganisationsCompanies, organisational units, processes and dependencies. No assessment of individual employees.Distinct analysis context
• 04States and public systemsPublic services, institutions, infrastructure and policy context with explicit geographic and temporal scope.Distinct analysis context
• 05Products and technologiesProducts, software and technical systems with a specific version, function and deployment context.Distinct analysis context

If required details such as country, sector, current process, system configuration or control model are missing, Kasp does not estimate a neutral replacement value. The affected claim remains withheld and names the missing context.

For work, eight impact dimensions are assessed separately

There is no universal combined value. Each impact dimension requires suitable data, sources and approval conditions. The following eight dimensions belong to the work methodology. Their criteria do not automatically transfer to other areas; qualitative reports must identify their own subject, relevant evidence and open questions.

• 01Technical feasibilityWhether the concrete system configuration can complete a defined workflow step under specified conditions.Separate subject of review
• 02Workflow state distributionSeparate shares for no material change, assistance, partial automation and full automation of a step.Separate subject of review
• 03Change in human working timeHuman time before and after the AI use, including setup, monitoring, review, correction and rework.Separate subject of review
• 04Review and control burdenExpected time for control, correction, escalation and approval per output unit.Separate subject of review
• 05Quality changeDifference between current quality and quality with the concrete AI use on a suitable metric.Separate subject of review
• 06Augmentation impactAdditional benefit while a relevant human role remains.Separate subject of review
• 07Task substitution potentialShare of human execution time that could be removed without breaching quality, safety, legal or accountability thresholds.Separate subject of review
• 08Residual risk after controlsRemaining risk by hazard category after required and demonstrably effective controls.Separate subject of review

The basis comes before every number

A completely filled analysis does not yet prove an empirical effect. Supported result dossiers can separately show stored sources, source status, assumptions, uncertainty and gaps.

• 01Claim statusKeep evidence, user input, assessment and withheld claims distinguishable
• 02Source statusShow status and evidence grade when stored
• 03Source familiesFit matters more than the raw number of references
• 04Assumptions and gapsKeep visible where the dossier contains them
• 05Uncertainty and data dateShow when supplied; otherwise do not hide the gap
• 06Source detailTitle, publisher, date, link and other stored source details

Current process, concrete AI use and controls remain separate

Kasp is designed to assess the additional effect of a concrete AI use. Existing software, automation or robotics effects are not silently attributed to AI.

• 01Current processHuman work, existing software, controls, time, quality, errors, cost and accountability.Baseline
• 02Development without the new AI useWhat would likely have happened over the same period without the planned introduction.Counterfactual
• 03Concrete AI useProduct version, tools, permissions, autonomy, human oversight, fallback, country and data flow.Clearly bounded
• 04Law and effective controlsTechnical capability, legal permissibility and residual risk are assessed separately. Human oversight counts only when it is concrete and effective.Separate assessment

A computed value is not yet a published value

A value may appear only when the context is sufficiently complete, the sources support the concrete impact dimension, uncertainty is known and the required checks have passed.

• 01Reproducible calculationSame inputs, same result
• 02Evidence requirementFor every published impact value
• 03Calibration and independent reviewPer impact dimension
• 04Unmet conditionWithhold the value with a reason
• 05Current public stateNo approved impact or comparison values

Automated checking remains clearly labelled

Kasp can structure, map, check and summarise inputs. Multiple models are not independent proof of truth. Results may contain errors and are not human-expert reviewed unless this is explicitly and verifiably stated.

• 01Automated AI estimateClearly labelled
• 02Human expert reviewClaimed only when evidenced
• 03Legal assessmentIssue screening, not legal advice
• 04Unsupported decisionsNo assessment, ranking, or decision about individuals in employment, education admission, credit, insurance, benefits, migration, law enforcement, justice, or biometrics. Aggregate analysis and human review do not widen this boundary.Technically block