Choose context that changes the answer — Super Intelligence Academy

The possibility

Turn a crowded packet into a useful request for one enjoyable creative session, and explain what you left out.

A quick win starts with a task small enough to finish. Before writing a long prompt, name the thing you want to do today. Then choose the details that would change a sensible answer. A compact request can still be complete; a long one can still miss the one limit that matters.

Use the removal test: For each detail, ask: if I removed this, what might the assistant get wrong? Removing a 45-minute limit could produce a two-hour project. Removing yesterday’s weather changes nothing about an indoor drawing session. Give a reason for keeping or omitting a detail; do not choose by word count alone.
Separate limits from preferences: ‘No purchases’ rules out a plan that requires new supplies. ‘A plant motif would be nice’ guides the subject without forbidding alternatives. Label each honestly. Turning every preference into a hard rule can make the task unnecessarily difficult; treating a hard limit as optional can make the result unusable.
Keep enough evidence to check the result: Preserve source labels beside the time, materials and goal. Compression should remove repetition, not the reason a constraint exists. A good brief lets you check the proposed steps against the packet, including setup and cleanup, without reopening a pile of unrelated material.
Change one thing, then compare: Write a first request before viewing the authored comparison. Revise one missing boundary or distracting detail. Explain the expected improvement, then inspect whether a proposed plan actually fits. If you later try two live requests, differences in their answers alone do not prove which context choice caused the difference.

This is a prepared exercise with fictional sources and authored feedback. It does not call an AI service or evaluate your writing automatically. Try the separate Context Practice Lab in Build a context pack for the feedback-and-revision loop; live feedback is labeled there when available.