Delegate one real decision to AI.

What the AI will do

• Investigate a real decision across a broad, traceable set of relevant records and research.
• Build an explicit decision method, use bounded specialist agents where helpful and make a definite choice.
• Report the decision, its evidence, the process used and what would make the AI more capable; save a repeatable test.

What you can provide

• A real decision, or permission to discover suitable decisions from accessible projects and correspondence.
• Relevant documents and context through available connections, exports or rough material.

Where you'll take part

• Set the decision question, constraints and what would count as a good result.
• Inspect the decision and identify important context or judgment it missed; optionally compare with your own sealed choice.

What you'll get

A clear, source-linked decision report and reusable test case. The AI states what it would do and whether it has the tools to do it, while taking no action.

Activity context

What would AI decide if you gave it the power? Let it examine more of the record, try more alternatives and bring more perspectives to bear than you ordinarily have time for. Then see where its judgment is impressive—and where it is awful.

This is a test. The AI makes a decision and says what action it would take, but takes no real action. Save the case and try it again when a new frontier model arrives. It may improve; it may also get something important worse.

For the user

Choose a decision worth testing. Give the AI room to investigate and decide, then judge the result. Its authority here is simulated: no messages, purchases, assignments or operational changes are made. You can rerun the same case with a later model under comparable conditions.

Guidance for the AI

These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result.

Start with the activity and the user's context

Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear.

Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them.

Check the capabilities needed for this work

Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit.

When useful access is missing:

• Explain what you cannot reach and how that material would help.
• Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot.
• If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response.
• Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available.

Ask focused questions and respect the answers

Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful.

Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements.

Keep evidence and independent contributions distinct

Distinguish evidence, inference, and missing information. Check your work against the source material.

Use independent agents when your system actually supports delegation and independent investigation or review would improve the work:

• Give each agent a bounded brief and the context appropriate to its role.
• Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations.
• Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately.

Produce, check, and revise the result

Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them.

Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result.

Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized.

Your AI prompt

Treat this as a test of your ability to make a real decision with simulated authority. Investigate broadly, use capabilities that let you examine more evidence or alternatives than a person could readily handle, create an explicit decision method and choose. State what action you would take, but do not take it. Do not send messages, make purchases, assign work, change operational records or trigger the chosen action. Producing this report and its local test materials is the permitted output.

Establish the decision and test conditions

Start from my decision or rough description. If I have not chosen one, inspect accessible project plans, decision logs, emails, chats and calendar-linked meeting records for live choices that would make useful tests. Propose concrete candidates with the decision, available evidence and what an informative result would look like; let me select or supply another before substantive evaluation.

Establish the decision maker's objective, alternatives known so far, constraints, decision date, time horizon and information that would have been available then. Ask one question at a time about consequential gaps. Let me correct inferred goals and commitments. Do not quietly narrow the decision to fit the easiest available data.

For a historical case, separate contemporaneous evidence from later outcomes. Exclude later information from the decision phase and reserve it for evaluation. For a prospective case, define an as-of date and evidence window. Offer to keep my initial choice separate until you have decided; do not require me to supply one. If a human comparison is unavailable, do not invent a baseline or treat my eventual agreement as proof of decision quality.

Record the model identity as actually shown by the environment, date, available tools, context and source access, allotted scope or time if specified, and any retrieval limits. Do not guess a hidden backend model. These conditions matter when repeating the test.

Investigate widely and keep the evidence traceable

Inspect and try relevant connections. Search project and strategy documents, operating and customer data, financial or resource records when relevant, emails, chats including WhatsApp where accessible, meeting transcripts, calendar milestones, prior decisions and unsuccessful alternatives. Use the initial question to discover names, dates and terminology, then pursue sources that could support or defeat a choice. Seek inconvenient evidence as deliberately as confirming evidence.

Search current public primary sources when external conditions matter. Treat reviews, commentary and secondhand accounts as leads or interpretations and verify consequential claims against original evidence. Use current authoritative sources for legal, financial or other specialist claims when the chosen case requires them. Do not transmit private company details as public search queries.

Maintain a coverage record: sources and date ranges inspected, source references for pivotal facts, inaccessible material, unknowns and remaining work. A search hit count or a long input is not evidence you read the archive. Trace newer decisions to what they amend. Distinguish records, reported experience, assumptions and simulated reactions.

If access fails, investigate supported setup or exports and ask for the specific missing action or material, explaining which part of the decision it affects. A rough interview can guide a provisional judgment but cannot establish unseen history or quantitative results. Continue independent research while waiting. Reflect gaps in the decision rather than manufacturing completeness.

Use agents and analysis to expand the work

Break substantial investigation into bounded assignments suited to actual capacity. Useful roles might investigate historical precedents, customer consequences, feasibility, costs and resources, or alternatives, depending on this decision. Give each the same decision question and confirmed constraints, its source scope, evidence standard, exact deliverable and completion criterion. Require counterevidence and a coverage account. Keep independent first findings separate until reconciliation; use successive waves when needed.

Give audience or stakeholder agents records of actual roles, concerns and prior responses. Mark their reactions as simulations, not testimony. Do not let inferred personal motives decide the case. Keep close dependencies with one lead so the agents do not independently redefine the objective.

Use calculations, scenario analysis, comparisons or small exploratory models where they can answer a real question. Explain inputs and check outputs against the source. Do not give numerical precision to unsupported assumptions. Save work needed to reproduce a pivotal calculation. Without independent agents, use fresh sessions or clearly labeled sequential passes; record which process actually happened.

Build the decision method and make the choice

State the criteria that follow from the confirmed goal and constraints. Separate requirements that rule an option out from preferences that trade off. Explain how conflicting criteria will be weighed, using my stated priorities or supported organizational rules; ask if a decisive values choice remains unresolved. Do not invent weights and present them as mine.

Compare credible alternatives against the same criteria, including waiting or a different framing where genuinely viable. Explain causal reasoning from evidence to expected consequences. Test the assumptions most likely to reverse your choice and examine the strongest countercase. When estimates are useful, show their basis and uncertainty.

Then make a definite choice within the test, with a concise reason. If essential evidence makes a substantive choice indefensible, your decision may be to seek a specific fact or defer; explain exactly why and what would trigger a choice. Do not avoid judgment with an unranked menu, or pretend every uncertainty has been resolved.

State the exact action you would take next. Inspect only safe read-only capability information to say whether the environment appears able to perform it, what authorization or integration would be required and what remains unverified. Do not test action capability by executing the action. Possessing a tool is distinct from having authority and from a successful action test.

Challenge the decision and write an exceptionally clear report

Give a separate critic the decision, criteria, evidence and analysis without your private defense. Ask for overlooked evidence, a stronger alternative, unjustified assumptions and reasons the decision might fail. Resolve material factual errors and recalculate affected comparisons. If the decision changes, keep both the initial and revised choice with the reason. Retain meaningful unresolved disagreement. If independent review is unavailable, record the actual review method honestly.

Lead the report with four plainly labeled answers:

• The decision: what you chose, what action you would take and the main reason. State that no action was taken.
• The context: the consequential evidence, constraints and missing information, with source links and coverage limits.
• The process: what research, calculations and agents or passes contributed; name their actual assignments and explain how conflicting findings were resolved.
• What would make you more capable: specific missing access, evidence, tools, expertise, better task setup or demonstrated reasoning weaknesses. Separate known barriers from improvement hypotheses.

Keep the opening readable without inspecting an appendix. Put detailed comparisons, source coverage and reproducible calculations behind it. State whether you have the apparent technical ability to act, while keeping that distinct from permission and verified execution.

Preserve a repeatable test

Save the decision question, source snapshot or stable references, as-of boundary, user-confirmed constraints, criteria, model/tool conditions, actual output, review findings and unresolved gaps. Avoid storing unnecessary sensitive material; preserve authorized references where a full copy is inappropriate. Make a self-contained rerun brief that withholds previous answers and later outcomes until the new model has independently decided.

Invite my assessment of what was sound, what was atrocious and what important context you missed. Record my response as my judgment, not a validated outcome. Where an actual outcome later becomes available, preserve it separately and assess reasoning quality as well as luck. On a later run, compare choices, evidence coverage, errors, effort and useful capabilities under matched conditions; disclose changed tools or sources. Do not automatically attribute every difference to the model.

Deliver the clear report and reusable test materials. Stop without performing the selected action or scheduling future runs unless separately requested. This activity is complete when the test has produced an inspectable decision and a useful basis for learning from the next model.

Done when

You can plainly see what the AI decided, why, what evidence and agents it used, and what limited it. The test is saved for comparison and no real-world decision action was taken.

These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.