Flashcard review: reveal, judge and recover

Ask about recalled knowledge rather than visual familiarity

Cogni labels the visible solution Answer and asks how well the person recalls it. AlgoApp keeps the question and answer together above Again, Hard, Good and Easy. Both preserve enough context to compare the attempted answer with the solution instead of forcing a judgment on an isolated rating screen.

Phrase the question in a way that refers to the attempt before reveal. A generic Did you understand this? can blur comprehension, recognition and recall into one signal. If the task includes a typed response, retain it beside the solution when practical. Do not overwrite a partially correct response just to display a visually clean answer card.

Keep the rating targets stable while someone reads. A delayed image or a long answer should not push one button into the location of another just as it is tapped. Make the full solution reachable without requiring the person to submit a rating first. Content inspection and committing a review are different interactions.

Show the scheduling consequence without turning it into a shortcut

Mopiq explains its four ratings with example review intervals and a separate Start Studying action. Flashka displays interval labels beside Again, Hard, OK and Easy within the active review. The first teaches the model; the second brings a consequence into the actual decision.

Show estimates calculated for the current card when the scheduler supports that information. Do not copy one fixed set of times into every state merely because those values look familiar. New, learning and established cards can have different rules. If the displayed time is approximate, say so rather than implying an exact reminder appointment.

Keep the primary meaning of the grade clear. If people choose Easy only because it postpones work, the interface can corrupt the signal it asks them to provide. Explain the judgment first and the scheduling effect second. Product-specific documentation, such as Anki’s answer-button guidance, is useful context but is not a universal algorithm for every app.

Make every point on the grading scale interpretable

Retain Cards explains Wrong, Hard, Good and Easy in a modal, while the underlying buttons use Incorrect, Struggled, Confident and Easy. Brainscape uses a numbered five-point scale with endpoint labels. Both raise a language question: can someone map the explanation to the exact control they must tap?

Keep help vocabulary aligned with button vocabulary. If the explanation says Hard but the actual control says Struggled, show the connection rather than making the learner translate it. A scale with numbered intermediate values needs meaningful accessible names, not merely Button three. Color can support the ordering but should not be its only explanation.

Choose the scale around the distinction the product can use reliably. More grades do not automatically provide better information. If two neighboring choices are impossible for the intended audience to distinguish, simplify the decision or improve the explanation. Do not claim that a more detailed scale is scientifically superior based on its appearance.

Keep the unrevealed state a genuine opportunity to think

Eppika shows a word, translation and three grading choices. Reverso has an explicit Reveal Answer surface with Still Learning and I Know controls. Comparing these states shows why a review specification needs to identify whether the solution is currently visible, not simply whether a card is on screen.

When recall is the intended activity, make reveal an intentional action and avoid leaking the answer through an unlabelled preview or accessibility description. Audio may be a prompt, a hint or the solution itself. Define its role and name the control accordingly. A learner should not accidentally reveal information by trying to hear the question.

If the product allows grading before reveal, explain how that differs from the normal review path and verify that it does not record a successful response merely from moving forward. For recognition-oriented activities, an always-visible translation may be appropriate. The important point is to name the task honestly instead of treating every two-language card as the same assessment.

Separate leaving the card from submitting its result

Knowt provides Show Answer with numbered card navigation. Gizmo prompts the learner to think first and offers Reveal while previous and next arrows remain available. Navigation, reveal, hints and grading are separate events even when the controls fit around the same card.

Specify what Next means before attaching analytics or scheduling to it. Browsing another card may be an inspection rather than an assessed attempt. Do not count it as a correct answer by default. Preserve an ungraded card for the appropriate later path, and make returning to it understandable rather than silently adding a completed review.

Hints and assistance can change how the result should be interpreted. If an explanation or generated hint is available, keep that provenance with the review event when it matters to the product. Do not punish help-seeking with unexplained lost progress. Also keep source correction distinct from study grading: a flawed card needs an editing or reporting route, not just another Again tap.

Store one deliberate review event per decision

Model unrevealed, revealed, graded and skipped states explicitly for a stable card identity and content version. A grade should create one review event with the relevant context, not a new event on every rerender or repeated tap. If a card changes while the session is open, do not apply a rating to a different solution without resolving that mismatch.

Keep an undo policy for accidental ratings. Reverting the visible card without reverting its scheduling event leaves the interface and future queue inconsistent. If undo has a limit, explain it near the action. Synchronization should not replay a locally committed grade as a second review when the device reconnects.

Preserve session position and unsubmitted work across interruption. Keep scheduling data separate from content edits, favorites and generated assistance. A saved star is not a review, and an edited answer is not evidence of successful recall. Use privacy-conscious state metadata for diagnosis rather than collecting private study answers when event identity and timing are enough.

Review the mistakes people can make quickly

Ask a tester to attempt a card, reveal, inspect a long solution, grade and undo. Then ask them to browse forward without grading. Verify which actions change the future queue. Include a card with an image, a hint and an incorrect source answer so the session is not tested only with ideal one-line material.

Try rapid repeated taps, delayed media, an offline grade and reopening on another device. Compare the event history with the visible card and proposed next interval. Test screen-reader order so the solution is not announced in the hidden state and the rating buttons remain reachable after a long answer.

Measure accidental-grade reversals, duplicate review events and unexplained queue changes separately from cards viewed. More completed cards do not by themselves prove improved learning. If learning outcomes matter, evaluate them independently with an appropriate study design rather than attributing gains to the color or number of self-assessment buttons.

Review the learning interaction before the streak counter

Use these checks on a working build, not only on the polished reference state.

• ✓Reveal does not automatically record successful recall.
• ✓Question, attempted response and solution retain their distinct roles.
• ✓Grade explanations use the same terms as the actual controls.
• ✓Interval cues represent the current scheduling state honestly.
• ✓Browsing, hints, favorites and edits do not masquerade as reviews.
• ✓Undo restores the scheduling event as well as the visible card.
• ✓Interrupted and offline sessions cannot create duplicate grades.

Connect active review to the wider study flow

The language-learning onboarding article covers how a learner reaches a suitable plan. Active grading needs a clearer interaction contract once study begins.

Use learning re-engagement examples to reconnect a returning learner with the correct uncompleted work, not a misleading completion total.

For one scheduler’s specific terminology, see Anki’s study manual. Its documented rules should not be generalized to every recorded product.