By use case

A LEFT JOIN report can lose unmatched students when its status filter moves from ON to WHERE.

A key icon beside both enrollment columns can imply that each one is independently unique.

A 100-unit inventory becomes 300 when the root, groups, and leaves are added together.

Compare accepted-to-reviewed and accepted-to-eligible rates in a fictional funnel, with the chosen denominator and unresolved statuses kept visible.

Edit a six-node graph, retain an isolated node, and compare degree counts and shortest paths with the current edge list.

Inspect a nine-point cloth model, compare pinned points and links, and see how changes affect a small simulated sheet.

Run a fictional classroom proposal through discussion and voting while keeping a passed vote separate from confirmed implementation.

A cleanup removes a zero count and treats a label that looks like markup as HTML.

Adding a ready class by writing another class attribute can discard existing values or print duplicate attributes.

Use a shared classroom scenario to record an initial position and a later revision, with seated participation and room for uncertainty.

A sentence frame can give students the word “because” without giving them a reason to explain.

A lesson plan that lists “Collaboration: pairs” still leaves a practical question: what will the partners compare?

You rename label to displayLabel in your OpenAPI description, but the reference page still says label.

Pressing Next should not reveal a recommendation before the audience has seen its supporting facts.

Two setup.adoc files can contain different instructions without colliding.

“Wow!” and “Hooray!” can both fit a finished poster.

A neat route diagram can hide an unsupported arrow.

An invoice can show 24.00 USD while its XML lacks the currency attribute.

One respondent describes email requests; another mentions a paper board.

A team can prefer a simpler export and still lose a comma inside a field.

A spring can cross the center while an outdated label reports zero velocity.

“Send the records” leaves the period, source version, and checking criteria unresolved.

The largest joint score does not tell a player what to do against a fixed choice. A four-cell game lets learners compare those two questions.

Tuesday works for one role, while Basic setup is the only option the other can supply.

A retelling says “I saw the red stars,” but the chosen narrator never looked inside the box.

One successful truth-table row cannot establish validity. Look for a valuation that keeps every premise true while making the conclusion false.

A bold preview is not enough if the copied text is HTML or its closing tags cross.

A contact looks correct on a page, but the receiving app loses a note line or splits the name differently.

In “Annam videt Iulia,” the first word names the object in the supplied form bank. Cite case evidence before assigning the actor.

In “Heute Abend liest Mia ein Buch,” liest is the third written word.

Two people receive the same matching task but different label packets. Comparing only finished work hides a constraint that existed before the turn began.

Six placed legs can still make an incorrect model. Give each part an ID and check its body-region target as well as the count.

A strip starts at 3 cm and ends at 9 cm. Recording 9 cm confuses its endpoint with its length.

A listener guesses “envelope,” but the clue was “a paper pocket.” The target was identified and a declared word restriction was broken.

The HTML email can look correct while the plain-text body gives a different date.

A code email that shows 83417 has lost the first character of 083417.

Mia needs ten more minutes with shared markers, but promised to return them in five. A quick vote cannot explain how learners weighed that conflict.

“Know anyone who needs this?” leaves the connector to write the pitch and decide whose name it uses.

“Mina ne a pas de papier” preserves the verb and changes the article, but misses the elision.

Three pencils written as 三本 but read as さんほん need different feedback from three sheets labeled 三本.

A planner can name a concept while its assessment ignores that concept.

A recorder writes N-O-T-E-B-O-K, then silently inserts the missing O. The retry now looks like the first attempt.

A partner changes the recorded guess after the answer is revealed. The sheet now looks correct but no longer shows what happened.

A welcome draft should not promise a download or confirm a subscription that the setup has not established.

An XPath returning zero nodes may be using the wrong namespace rather than finding an empty document.

A cohort has a 100% error rate, but it does not contain every error.

Copying a child count into the parent creates two values that can disagree.

One Upload click can start two requests when two handlers subscribe to the same HttpClient observable.

Prototype rows look useful in a static template, but they must disappear from processed output.

Seeing home.html somewhere in the project does not mean the configured loader can find it by that name.

Applying the same acceleration to two unequal masses hides the distinction between force and acceleration.

A French route opens successfully, but its body is English. Trace the selected source before counting that page as translated.

Replacing “caused” with “related” is only the first step. Ask learners to name competing explanations and request evidence that could distinguish them.

A trailing space can be image data. Trimming it from the first ASCII row changes the width even when the visible symbols look correct.

Reordering the answer buttons should not change which region is correct.

A redirect can reach a page while losing repeated query values or trusting an unwanted target parameter.

An image can disappear when only Markdown text is copied into another cell.

A pruning trace returns 3 for a branch whose full minimum is -20.

A vacuum exercise skips the memory problem if its policy can inspect the whole world.

Query 3 can prefer train 0 while train 0 prefers a different query.