A completion rate appears to be a clear number: it shows how many people reached the last step of a course, form, or digital sequence. It can also lead to a mistaken decision when used as a substitute for learning. Reaching the end does not demonstrate understanding, transfer, judgement, or an ability to correct an error.
An exception is not noise to be hidden so a dashboard looks orderly. A person unable to sign in, an answer that does not fit closed options, excellent work submitted after a deadline, or a recommendation that harms a minority group tests the actual design. If a system works only for its expected path, its completion rate describes administrative convenience rather than educational quality.
A robust digital learning system treats out-of-pattern cases as evidence. It asks which assumption failed, who was left without support, which automated decision can be reviewed, and which equivalent alternative preserves the learning aim. Equity does not mean lowering a criterion in response to difficulty; it means separating the criterion from accidental barriers that prevent it from being demonstrated. The exception should therefore leave a trace, receive human response, and improve the shared process.
Progress systems turn visible actions into percentages, streaks, badges, and ranks. Those data can signal difficulty or a barrier, but they do not explain on their own what happened. A responsible number opens a question; it does not close an assessment of effort, capability, or commitment.
The risk grows when a metric is easy to compare and a conversation about work takes time. Activity that leaves a platform trace is then rewarded, while slow reading, peer support, offline work, or the revision that changed a mind become invisible.
Interpreting progress requires separating measurement from meaning. A percentage may record that a unit was opened or that an answer was reached after several attempts. It does not automatically record the quality of a question, the source used, the reasoning, or the material conditions for study.
The alternative is not to stop measuring. It is to use limited, explainable indicators alongside evidence of process. When learning matters more than completion, data guide support and review; they never turn a varied path into an automatic verdict.
What a metric sees and what it leaves out#
A responsible technology decision begins by describing the capability to be developed. If the aim is to interpret a historical source, the indicator cannot be only a convincing-looking document. It must be possible to see whether authorship, date, context, intention, and missing information were distinguished. If the aim is to model a phenomenon, it is important to inspect assumptions, omitted variables, and what changes when a datum changes. In both cases, a useful tool makes reasoning visible rather than covering it with an immaculate presentation.
A responsible technology decision begins by describing the capability to be developed.
The initial question also limits the information an activity needs to collect. Asking for a detailed record of every click simply because a platform offers it can normalise unnecessary surveillance. In many situations, samples of work, short explanations, revised versions, and a conversation about decisions are enough. Privacy and assessment are not separate matters: measuring only what a platform captures narrows the very idea of learning. UNESCO’s media and information literacy work includes understanding how information circulates and evaluating it critically.
Material conditions deserve advance attention too. A resource requiring an account, a recent device, extensive connectivity, or personal data can widen existing inequalities. Design should provide equivalent alternatives, clear instructions, and time to become familiar with the resource. This is not a rejection of digital work. It prevents skill with an interface from being confused with the knowledge being built. Accessibility belongs in the opening question: who can participate, with which support, and how exclusion will be checked.
Case one: many summaries, little checked evidence#
A social-science class was asked to investigate causes of a recent rent increase. The first instruction appeared efficient: collect articles, use a summarising tool, and submit an infographic. Within hours, visually ordered pieces appeared, with charts and confident sentences. On reading them, the teaching team found incompatible explanations presented as facts, percentages without sources, and repeated confusion between advertised prices and signed contracts. The activity had produced complete documents before agreeing which question available sources could answer.
The revision did not ban the generator or punish technology use. Submission was paused and a more precise question was proposed: which indicators describe the price change in this municipality, over what period, and for what kind of housing? Each group made an evidence card with author, date, method, scope, and possible interests of the source. It then compared a public statistic, a sector report, and press testimony. The automatic summary became an object of analysis: verifiable claims, omissions, and words suggesting unproven causation were marked.
The final infographics included a section called “what cannot be concluded.” Some groups learned that their data did not distinguish neighbourhoods; others that a national figure did not describe the city. The limitation did not lessen the work’s value. It made the explanation more honest. The OECD notes that digital education should strengthen judgement and participation, not merely access to devices. The case showed that a tool can support comparison when the question and sources retain priority over the appearance of a finished answer.
Case two: the progress dashboard turned into a verdict#
A school introduced a platform with dashboards showing connection time, completed exercises, first-attempt correct answers, and rank in a table. Meetings began to revolve around these numbers because they were immediate and easy to compare. One group appeared inconsistent even though several people downloaded material to work offline. Others appeared advanced because they repeated exercises until correct, yet could not explain the procedure in conversation. The dashboard offered useful signals, but it was being used as a verdict on capacities it could not observe.
The team then defined three assessment questions: which concepts are understood, how they are applied in a new situation, and which strategies are used after an error is noticed. Platform data became secondary evidence. To answer the other questions, the team added varied problems, reasoning notes, and peer review guided by public criteria. Each learner could attach a brief explanation of a decision the system had not recorded. Rankings were no longer projected; individual progress and opportunities to revise were shown instead.
This change altered the conversation. Low connection time was no longer automatically read as low effort; access barriers, care work, and study preferences were investigated. An incorrect answer ceased to be a lost point and became diagnostic material. The European Framework for the Digital Competence of Educators stresses technology for better assessment and inclusion. Measurement is not understanding. Indicators gain meaning when a pedagogical question defines what they can say and, especially, what they cannot say.
Case three: finished code, understanding still pending#
A programming workshop proposed an application for air-quality information. Code templates and automatic assistants made a functional screen possible during the first session. There were buttons, maps, and colours changing with values. Yet nobody had decided where data would come from, how often it would update, what a high index meant, or what should happen if the service failed. Some teams copied access keys into public files, while others showed old figures as current readings. Completing the interface had displaced the problem that gave the application meaning.
The next session started with thirty minutes away from computers. Participants drew the path of a person checking information before leaving home and formed questions: what datum is needed, which source has authority to publish it, which warning is clear without causing alarm, and how should a missing reading be shown? Each group then wrote acceptance criteria and a risk list. Only then were services, libraries, and components chosen. The code assistant remained available, but every added fragment required a note explaining its function, dependency, input data, and test performed.
The final applications had fewer decorative features and more comprehensible messages. They included an update date, a source link, a notice for unavailable data, and a test preventing an old value from appearing current. Spain’s National Cybersecurity Institute highlights the need to incorporate security from design rather than at the end of a project. Here, learning programming meant connecting a technical decision to a consequence for someone receiving information. Fast code generation became valuable when it served a previously discussed criterion.
Assess processes that leave room for correction#
When learning matters more than completion, assessment needs traces of process. Every gesture need not become a metric. A reasonable selection can include the initial question, consulted sources, a first hypothesis, changes made, an explanation of errors, and a final version. This sequence permits improvement to be valued without requiring every attempt to be correct. It also reduces the temptation to present an incomprehensible result as one’s own, because it invites concrete decisions to be explained rather than perfect appearance to be defended.
Rubrics can help when they describe observable actions: defining an issue, comparing evidence, recognising limits, selecting an appropriate method, and communicating clearly. They are less useful when they reward mechanical obedience to a template. A good rubric leaves room for two reasonable routes to reach different conclusions. The Council of Europe connects digital citizenship education with critical, responsible, and safe participation. That participation requires the ability to disagree with an automatic suggestion and explain why a source, rule, or result deserves review.
Feedback should arrive while something can still change. A comment at the end of a closed task teaches little; a question during the process opens an opportunity to learn. “What evidence would support this claim?”, “what if this datum were incomplete?”, and “who is not represented?” bring attention back to reasoning. Tools can organise drafts, enable comments, and provide alternative access, but they should not replace the conversation that makes a doubt interpretable.
Technology that helps learning continue to ask#
The first question is not a formality archived at a project’s start. It should return when unexpected results appear, when a source contradicts a hypothesis, or when a tool recommends an opaque option. Asking again permits a change of method without abandoning the aim. It also protects against a thin idea of efficiency: saved time is valuable only when invested in better understanding, support for someone who needs it, verification of a claim, or review of a possible harm.
An educational organisation can make this approach habitual through straightforward decisions: state purpose before adopting a platform, test the resource with diverse groups, publish which data are collected and for how long, provide a non-digital alternative where appropriate, and review the tool after use. Decisions should not rest only on commercial promises or novelty. They should observe what learning became possible, which barriers appeared, and what was lost by automating part of the activity.
Completion remains necessary: projects need submissions, services must work, and people need answers. Yet an outcome gains value when it remains explainable, open to questioning, and correctable. A well-chosen tool does not silence the first question. It makes it more productive by allowing better resources for investigation without confusing an available answer with understanding of a problem.
Institutional sources: UNESCO, media and information literacy; OECD, digital education; European Commission, DigCompEdu; INCIBE, cybersecurity and digital citizenship; Council of Europe, digital citizenship education.
