28 principles for recognizing positive impact

Eligible improvements and relevant relationships receive recognition. The score does not calculate harm or the net balance of all activity.

A profile with eligible improvements can receive a score even if its activity has other effects. The score does not certify that the whole profile is sustainable.

An AI reader assigns reach levels from L0 (symbolic) to L5 (transformative), following shared rules. The scoring engine converts those assessments into points. L4 and L5 count as L1 until human confirmation; new improvements or those whose text has changed also count as L1 until reassessed.

Reducing consumption can be described as an improvement and assessed by its reach. There is no automatic comparison of lifestyles, purchases or mobility alternatives.

There are no separate scoring components for improvement and current performance. Measurements and their averages are not yet included in the algorithm.

A conservation activity can be assessed by its reach. No automatic additional points are awarded for each year of maintenance or continuity.

There is no separate assessment of unsuccessful effort and the result that remains. The model uses published improvements and their reach assessments.

Recording a goal or promise adds no points. A published plan that qualifies as an eligible improvement is assessed by its reach; ambition receives no automatic bonus.

The AI assessment can identify the same work and its phases or components. The engine groups those improvements within the profile and counts the highest applied level once. A plan and its measures remain separate. Municipal profiles also limit overlap between actions; detection depends on the available information.

Consumption and outcome measurements can appear on the profile, but do not yet directly change the score. A published improvement on that topic is assessed separately.

Environmental and social improvements across different topics are accepted. Their reach is assessed in the profile’s context; points do not convert those topics into a common physical unit.

The algorithm does not perform a lifecycle assessment or automatically calculate manufacturing, use and end-of-life impacts. That analysis remains a development goal.

AI interprets reach relative to the profile’s own scale: the municipality’s population and territory; the organisation’s operations and reach; the person’s life, household and activity; or the product and its users. Size enters through that interpretation of the described context. There is not yet automatic numerical normalisation by population, employees or area, or scoring based on position against comparable profiles.

An AI-assisted reach assessment does not verify that the facts are true. Impact uses that assessment; Reliability considers evidence, sources and relationships. Confirming a high reach level reviews that assessment and does not replace documentary support.

Eligible improvements can receive points without an audit. Reliability distinguishes coverage, own sources, independence and corroboration through bands and a numerical score.

With profile completion enabled, AI can identify improvements from saved text, readable PDFs and photo or video captions. It does not analyse pixels or audio, and extraction does not verify the facts. Reach assessment for Impact is a separate process. For Reliability, confirmed links and source families determine support; uploading a file is not enough.

A recognition with an identifiable issuer can support Reliability according to its scope and links. Linking an SDG or recording a VSME report adds no points by itself.

The assessment instructions require using stated facts, inventing no quantities and choosing the lower level when uncertain. AI can make mistakes: assessments are stored against the text and can be reviewed. Without a current assessment, a conservative provisional level applies; without eligible improvements, the profile remains unscored.

Relevant relationships and the reach of improvements can recognise these contributions. The algorithm does not assign an exact causal share of the outcome to each participant.

The network transmits limited influence according to eligible relationships and their trust level. The actual intensity and duration of each relationship are not weighted automatically.

Recognition received from the network does not subtract points from the source profile. It remains subject to the limits and rules of each scoring component.

Network influence decreases and has limits on traversal and points. The calculation avoids counting profiles again within a path; disputed relationships transmit no influence.

The model groups the same work within a profile, but does not maintain a global record of physical outcomes shared across profiles. Points from different profiles cannot be added as tonnes avoided.

Details show the applied reach level and, where relevant, an assessment awaiting confirmation or grouping of the same work. They also show versions and the effect of removing an item and recalculating. Those individual effects cannot be added together.

There is no automatic depreciation due to age or improvements by other profiles. Scores can change when data, reach assessments or the model version are updated.

The history preserves snapshots with the version that calculated them and marks model changes. There is no streak counter that adds or removes points.

Strategies and examples can be explored. Saving or linking a strategy adds no points and does not mean an improvement has been carried out; personalised recommendations remain in development.

The methodology explains reach levels, human review, calculation rules and their limits. Assessments store a fingerprint of the interpreted text, and historical snapshots retain their version. Correcting an assessment or changing the model can change the score; clarifying this page does not change the algorithm.