0/2 — Produce a polished five-minute script as quickly as possible.
This treats the deliverable as the goal and ignores the understanding the teacher will test.

2/2 — Understand and explain why the Salt March mattered, using evidence and my own reasoning.
Strong. It identifies the knowledge, reasoning and evidence the student must demonstrate.

1/2 — Collect as many historical facts as possible, even if I cannot explain how they connect.
Facts may help, but the assignment requires explanation and connected reasoning, not a list.

0/2 — Ask AI to write the complete script, memorise it and disclose that AI wrote it.
Disclosure is important, but it does not repair the loss of the learning task.

2/2 — Ask AI for planning questions and a possible structure, then verify sources and write the explanation myself.
Strong. AI supports the process while the student retains authorship, verification and understanding.

0/2 — Use an AI summary, change a few words and rely on the teacher not asking detailed questions.
Minor wording changes do not create genuine understanding or authorship.

Required elements: Ask me clarifying questions before suggesting an outline., Suggest a structure that connects causes, events and consequences., Mark claims or quotations that need independent verification.

2/2 — Use the planning prompt, verify all evidence, write my own presentation and record/disclose the assistance as required.
Strong. This combines appropriate use, verification, authorship and disclosure.

1/2 — Do not use any AI because all AI support is automatically cheating.
This is safe but overgeneralised. The school permits limited support that preserves the learning goal.

0/2 — Use the complete AI script because a disclosure statement makes any use acceptable.
Disclosure does not make substitution consistent with the assignment's learning goal.

Strong selection: The city or region being discussed., The dates and comparison period., Whether the claim concerns daily highs, averages, surface temperature or another measure., Local observations and relevant weather or land-use data.

0/2 — High confidence in System A because its explanation is more detailed.
Detail and fluency do not establish that the explanation fits the unnamed city and period.

2/2 — Low confidence in a specific cause; treat both as hypotheses requiring local evidence.
Strong. This separates plausibility from case-specific support.

0/2 — Combine the answers and say both were equally responsible.
Averaging unsupported claims does not create evidence about their relative contribution.

1/2 — Use System A but add the word 'probably'.
A qualifier helps, but it still privileges one explanation without a basis.

Strong selection: Compare local meteorological records across the relevant periods., Check reliable land-use and urban-development information for the city., Check reputable climate data explaining El Niño's regional effects during the period.

0/2 — System A is correct: cities are hotter because of buildings and roads.
This turns a general mechanism into an unsupported case-specific conclusion.

2/2 — Both mechanisms could matter, but the information provided is insufficient to identify the main cause; local data must be checked.
Strong. It communicates useful possibilities and calibrated uncertainty.

1/2 — AI cannot answer questions about weather, so reject every part of both outputs.
The outputs contain plausible concepts, but their application to this case is unverified.

Fabricated — Tagore wrote: “Education is the lighting of a fire, not the filling of a vessel.”
The quotation and attribution are presented as exact, but the cited source cannot be found and the attribution conflicts with available references.

Verified — India's literacy rate increased between the 2001 and 2011 censuses.
The direction of change is directly supported by the identified official data.

Unsupported — Seventy-three percent of Indian teenagers use AI every day (UNESCO 2023, p. 42).
The claim may or may not be true, but the supplied citation does not support it.

Partly supported — Generative AI can support brainstorming but may also produce inaccurate citations.
The general proposition is well supported, but its broad wording needs context and should not be treated as a quantified universal claim.

2/2 — Check the exact quotation and the 73% statistic first.
Strong. Both are precise claims carrying apparent authority and high credibility risk.

0/2 — Check only the spelling and grammar because the citations look professional.
Presentation quality does not establish factual support.

1/2 — Check the broad brainstorming claim first and leave the quotation for later.
The broad claim needs context, but the exact quotation and statistic carry a more immediate verification risk.

Strong selection: Remove the quotation unless a reliable primary or scholarly source is found., Remove or clearly mark the statistic as unverified until a supporting study is located., Cite the official census table for the literacy claim.

0/2 — Submit it unchanged because the draft includes references.
References that do not support the claims create false assurance.

2/2 — Revise it claim by claim, keep only supported content and record/disclose the AI assistance.
Strong. This preserves useful work while repairing evidence and authorship.

1/2 — Reject every sentence because one quotation was fabricated.
Caution is reasonable, but supported claims can be retained after independent checking.

Strong selection: A dated notice on the school's official website or verified school channel., A current pollution-control or education-authority notice for the relevant location.

Primary/current — A school notice dated today, signed by the principal and posted on the official domain.
It has direct authority over the school's operation and is current.

Independent corroboration — A local newspaper article that links to today's education-authority order and quotes the order accurately.
It independently reports and links the authoritative order.

Circular or derivative — A blog cites a news aggregator; the aggregator cites the same viral message.
Multiple pages repeat the same unsupported origin.

Outdated for this claim — A genuine school-closure order from a previous pollution episode two years ago.
Authenticity does not make an old order evidence of a current closure.

Required elements: Write the exact claim being checked., Record the date, time and location covered by each source., Link or identify the primary current notice., State any unresolved conflict or missing information.

0/2 — Forward it with “not sure if true” so classmates can decide.
Forwarding still amplifies an unverified alarm and may detach the warning from your qualifier.

2/2 — Do not forward it as fact; point classmates to official channels and alert a responsible adult if it is spreading.
Strong. This limits harm while preserving a route to authoritative confirmation.

1/2 — Announce that school will definitely remain open because no notice was found.
Lack of a notice is not definitive proof. The responsible response should preserve uncertainty.

1/2 — Choose whichever threshold selects the same number from each group; that proves the system is fair.
Equal counts can be informative, but they do not prove the features or individual decisions are fair.

2/2 — Threshold changes affect outcomes, but they cannot by themselves fix a score that may encode unequal access; the features and missed students need review.
Strong. You distinguish a decision rule from the quality and fairness of the inputs.

0/2 — Use the threshold that selects the most students because a higher selection rate is always fairer.
Overall selection rate alone does not address unequal errors, limited places or the relevance of the score.

Strong selection: Students selected and not selected, including those with fewer opportunities., Families who may need to understand or challenge the process., Teachers and the scholarship committee accountable for the decision process., The tool provider, which should explain design assumptions and limitations.

Strong selection: Review whether extracurricular participation is relevant and whether opportunity differences require removal or adjustment., Test selection and error patterns by relevant groups, with a clear purpose and safeguards., Require trained human review of borderline and adverse cases., Give students a clear explanation and a way to provide missing context or appeal.

0/2 — Use the score as the automatic shortlist because a human makes the final award later.
The shortlist already determines who receives consideration, so the automated gate is consequential.

2/2 — Pause automatic use, revise and test the model, then use it only as one input with human review, explanation and appeal.
Strong. This recognises both technical and procedural safeguards.

1/2 — Delete every use of data in scholarship decisions and select students randomly.
This avoids the tool's risks but does not create a defensible scholarship process. Relevant evidence may still be used responsibly.

Generally safe for this purpose — The public URL of a museum page used as a design reference.
A public reference URL is generally low risk, subject to normal source and copyright checks.

Permission or school check required — A class photograph showing identifiable students.
Identifiable images of other students require consent and compliance with school rules and tool terms.

Unsafe or high risk to enter — A student's live home location so the design can show a nearby landmark.
It is sensitive and unnecessary for the stated purpose.

Unsafe or high risk to enter — A classmate's health condition mentioned in project notes.
Health information is highly sensitive and should not be entered for this purpose.

Generally safe for this purpose — A teacher-provided anonymous list of colour preferences with no names or identifiers.
Purpose-limited anonymous data prepared by the teacher is generally appropriate.

Unsafe or high risk to enter — A spreadsheet containing student names, marks and attendance.
The data is sensitive, belongs to multiple people and is unnecessary.

0/2 — Upload it because the project is educational and the image already exists.
Educational purpose does not automatically authorise a new external upload or reuse.

2/2 — Use a lower-risk alternative; upload only if the school approves, the tool is appropriate and every affected person has the required consent.
Strong. This preserves the goal while respecting necessity, tool risk and affected people.

0/2 — Crop one student out and upload the rest without asking because fewer people are visible.
Reducing the number of people does not create consent for those who remain identifiable.

Required elements: Confirm the information is necessary for a clear learning purpose., Check permission, consent and school rules for information about other people., Use the least detailed or least identifiable information that will work., Sign out and avoid saving sensitive information on the shared device.

2/2 — Proceed using public references and anonymous teacher-prepared material; ask the teacher before any personal-data upload.
Strong. This is proportionate and keeps a clear escalation route.

1/2 — Avoid every digital tool because privacy can never be managed.
This is cautious but unnecessarily absolute. Low-risk, purpose-limited use can be appropriate.

0/2 — Upload all available material, then delete the final poster after submission.
Deleting the output does not control what the tool retained or reused.

0/2 — Issue the penalty immediately because 91% is above 90%.
A confidence score is not proof and the tool provides no explanation or validated threshold for this context.

2/2 — Pause the accusation, review the work and process, and give the student a fair opportunity to explain and provide drafts.
Strong. The tool triggers inquiry, not an automatic adverse decision.

1/2 — Ignore the flag completely because AI detectors are never useful.
The flag may justify a proportionate review, but not a penalty by itself.

Strong selection: The student's notes, drafts, revision history and ability to explain the argument., The assignment rules and what AI or accessibility support was permitted., The checker's documented limitations, validation evidence and error rates for comparable students and tasks., Relevant accessibility or language-support context, handled confidentially.

Strong selection: The teacher, who understands the assignment and student work., The student, with a chance to explain and provide evidence., An appropriate academic or school lead for serious or disputed findings., A relevant accessibility or safeguarding professional where necessary and authorised.

0/2 — Allow an appeal only if the student can prove the detector is technically defective.
This places an unrealistic burden on the student and ignores errors in context or process.

2/2 — Explain the concern and evidence, allow the student to respond, and provide review by an appropriate human who can change the decision.
Strong. The route is informed, participatory and capable of remedy.

0/2 — Let the same automatic checker run the essay again and call the second score an appeal.
Repeating the same mechanism is not independent human review.

Assistance — The group asks for possible topic angles, then chooses and develops one themselves.
The tool helps generate possibilities while the group makes and develops the decision.

Substitution — A complete analysis is generated and submitted unchanged as the group's reasoning.
The AI replaces the core work being assessed.

Assistance — After writing a draft, students use sentence-level editing suggestions and accept only changes they understand.
The students retain the ideas and make informed editing decisions.

Depends on checks, permission or disclosure — An AI image is added without checking licence, accuracy or teammate agreement.
Use may be acceptable after checks and group agreement; the current process is incomplete.

Substitution — The AI performs every calculation and no student checks or understands the method.
The tool replaces assessed reasoning and creates an accuracy risk.

0/2 — Say nothing because each student controls the tools used on their own section.
The final submission represents the group and may expose all members to academic-integrity or accuracy consequences.

2/2 — Tell the group, review the calculations and image, remove or correct problematic material, and agree on the disclosure.
Strong. This respects shared responsibility and allows correction before submission.

1/2 — Remove every AI-assisted sentence without discussion, even if it was permitted and verified.
This is cautious but bypasses group decision-making and may discard legitimate assistance unnecessarily.

Required elements: The purpose: brainstorming, outline support, editing, image generation or another specific use., What the AI actually contributed., What the students verified, changed or created themselves., That the students reviewed and remain responsible for the final submission.

1/2 — “AI was used.”
This discloses something but is too vague to show the extent of assistance or human work.

2/2 — “We used an AI tool to brainstorm topic angles, suggest an outline, edit selected sentences and generate an image. We checked the factual content and image, redid the calculations ourselves, revised the final work and remain responsible for the presentation.”
Strong. It is specific about contribution, checking, correction and responsibility.

0/2 — “No AI produced the final presentation because we edited the output.”
Editing does not erase the AI contribution and this wording conceals material assistance.