Responsible AI Use

Our approach

Lumimetrica uses curated statistical methods and machine-learning models to turn learning activity into learner statistics: estimates of progress, growth, engagement, trajectories, cohort patterns and risk. These outputs can influence decisions about real learners, so the way they are produced and used matters.

This page sets out the principles that guide how Graph25 Informatics Limited designs Lumimetrica, and what we ask of the organizations that use it. It applies to all statistical and machine-learning features of Lumimetrica.

Human judgment remains central

Lumimetrica is designed to inform educators, not to replace them. Statistical outputs are one source of evidence among many. Educators know their learners, their context and their constraints in ways no dataset can capture, and decisions about learners remain theirs.

Outputs are interpreted in context

A learner statistic describes a pattern in data. It does not, by itself, explain why the pattern exists. Lumimetrica presents statistics alongside their context — the learner’s history, relevant peers and the underlying trend — so that they can be read in proportion and not as isolated verdicts.

Models are not infallible

Every statistical model is a simplification. Estimates carry uncertainty, models can be wrong, and patterns that held in the past may not continue. Lumimetrica reports estimates with an indication of their uncertainty and avoids presenting probabilities as certainties or predictions as facts.

Appropriate human oversight

Signals raised by Lumimetrica, such as emerging risk, are surfaced for review by people with the authority and knowledge to act on them. Organizations decide who reviews signals and how, and can adjust how Lumimetrica is used within their own processes.

Avoiding inappropriate automated decisions

Lumimetrica is not designed to make decisions that have legal or similarly significant effects on learners — such as admission, progression, sanctions or eligibility for support — without meaningful human review. We ask organizations not to use Lumimetrica’s outputs as the sole basis for such decisions.

Transparency

We aim to explain, in plain language, what each statistic means, what data it is based on and what its limitations are. Organizations can ask how a given output was produced, and we will explain the methods involved at a level appropriate to the question.

Data quality

Statistics are only as reliable as the data behind them. Missing, inconsistent or biased data can produce misleading results. Lumimetrica is designed to make data coverage visible, and we work with organizations to understand the quality and limitations of the data they provide.

Fairness

Patterns in learning data can reflect existing inequalities. We pay attention to whether statistics and signals behave differently across groups of learners, and we encourage organizations to review outputs with that possibility in mind rather than treating differences between groups as inevitable.

Privacy

Learner data is used to provide learner statistics to the organization that supplied it, following the principles described in Enterprise Privacy. We favour approaches that use no more personal information than the statistics require.

Security

The models and the data they rely on are protected by the same security measures that protect the rest of Lumimetrica, described in the Security section of the Lumimetrica website.

Monitoring

Statistical models can drift as learning environments change. We monitor the behaviour of Lumimetrica’s models over time and review them when their outputs no longer reflect the data they describe.

Responsible intervention

A signal is an invitation to look more closely, not an instruction to act. When educators intervene, we encourage them to communicate with learners respectfully, to avoid labelling learners on the basis of a statistic, and to use Lumimetrica’s longitudinal view to check whether the intervention is helping.

What we ask of organizations

Organizations using Lumimetrica are asked to:

• make sure the people who use learner statistics understand what they mean and what they do not;
• keep a person responsible for every decision that significantly affects a learner;
• provide data that is as accurate and complete as practical;
• tell us when outputs appear wrong, unfair or unhelpful.

Raising a concern

If you believe Lumimetrica has produced an output that is inaccurate, unfair or harmful, please tell us using the contact details below. We take these reports seriously and use them to improve the system.

Contact

Questions about this page can be sent to Graph25 Informatics Limited:

[email protected]
+256 756 494 518
3rd Floor, Kanjokya House
Plot 90–92 Kanjokya Street
Kampala, Uganda

← Back to Lumimetrica