dimension_name stringclasses 1 value | dimension_values listlengths 3 3 | dimension_description stringclasses 1 value | question stringlengths 55 119 | why_differ stringlengths 180 316 | how_subtle stringlengths 87 201 | sampled_value stringclasses 3 values | num_distractors int64 5 5 | ground_truth_persona stringlengths 179 344 | distractor_personas listlengths 5 5 |
|---|---|---|---|---|---|---|---|---|---|
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have $60 to cook vegetarian meals for the week with decent protein. How should I plan it? | A high-agency response would present a concrete menu, grocery list with quantities, and prep schedule. A shared-agency response would outline a draft plan and ask about preferences or kitchen constraints, while a low-agency response would list budgeting strategies and protein sources without committing to one plan. | Differences show via whether the answer commits to a single plan, presents decision points inviting input, or neutrally enumerates options; cues are in specificity, sequencing, and firmness of choices. | Low-Agency | 5 | A home cook with a basic kitchen planning a $60 vegetarian week. Prefers the assistant to lay out neutral options with prices and protein per serving, without recommending a single ‘best’ plan. Wants a clear shopping list and multiple menu combinations to choose from. | [
"A home cook planning a $60 vegetarian week with a basic kitchen. Wants the assistant to pick a specific 7-day menu, finalize the shopping list, and state exact quantities and meal prep steps. Expects firm recommendations on what to cook.",
"Comfortable following clear directions and not seeking back-and-forth. E... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want to be ready for a 10K in 8 weeks while working full-time. What training approach would you suggest? | High-agency would give a dated calendar with target paces and rest days. Shared-agency would propose a base structure contingent on current fitness and solicit adjustments, while low-agency would describe principles and multiple sample weekly templates without a definitive pick. | Perceivable through granularity (calendar vs framework), conditional phrasing, and whether one plan is fixed versus multiple alternatives are offered. | High-Agency | 5 | Wants decisive, prescriptive coaching tailored to a full-time schedule. Expects the assistant to choose a specific 8-week 10K plan with defined weekly mileage, key workouts, and rest days, and to state clear reasons. Prefers firm recommendations over options. | [
"Prefers to co-design the 8-week 10K plan, discussing trade-offs and adjusting around a full-time job. Wants the assistant to propose directions, ask questions, and refine the plan together. Comfortable arriving at decisions collaboratively.",
"Seeks a conversational partner to think through training approaches. ... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | My desktop is a mess of files. What's an effective way to sort and keep it organized? | High-agency would provide a specific folder schema, naming conventions, and a cleanup sequence. Shared-agency would introduce categorization options and ask about workflows to tailor the system, while low-agency would compare common methods with pros and cons. | Signal comes from decisiveness (one taxonomy), presence of collaborative decision points, or neutral comparison without selection. | High-Agency | 5 | I want the assistant to take charge and tell me exactly how to fix my messy desktop. Give me a single recommended folder structure, naming rules, and a clear sequence of steps to follow now. Make decisive choices for me rather than offering multiple options. | [
"I prefer to talk through the problem and co-design an organization system. Ask me a few questions about my file types and habits, then suggest a plan we can refine together. I'm open to iterating based on what seems workable.",
"I like collaborative guidance. Walk me through trade-offs of different folder approa... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We have two weeks to build an MVP. Features on the table: feed, offline mode, in-app support. How should we prioritize? | High-agency would assign must/should/could rankings and a build sequence. Shared-agency would set up a scoring framework and request criteria, while low-agency would analyze benefits, effort, and dependencies without ordering. | Implicit through whether a ranked backlog is produced, a collaborative framework with inputs appears, or a balanced analysis is given without a final prioritization. | Low-Agency | 5 | A product teammate assisting a small team on a two-week MVP. Prefers to outline options, surface trade-offs, and provide data points without steering decisions. Will list pros/cons and timelines for each feature so the team can choose. | [
"A product teammate assisting a small team on a two-week MVP. Tends to take the lead by proposing a single stack‑ranked plan with explicit trade‑offs and justification. Will make crisp calls to reduce ambiguity and keep scope tight.",
"A product teammate assisting a small team on a two-week MVP. Works collaborati... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I need to budget $2,000 for a four-day conference trip. Can you help plan the spend? | High-agency would allocate line items with target amounts and booking order. Shared-agency would set envelopes based on user preferences and propose ranges, while low-agency would summarize typical costs and cost-saving tactics without assigning amounts. | Cues lie in concreteness of allocations, explicit decision points tied to preferences, and neutrality in presenting ranges and tips. | Shared-Agency | 5 | Planning a four-day conference trip with a $2,000 budget. I prefer to talk through trade-offs and shape the plan together: propose allocations and ask a couple of clarifying questions as we go. I value shared reasoning and brief check-ins before finalizing the breakdown. | [
"I want a confident, done-for-me plan for a four-day conference trip within $2,000. Please take the lead, decide the allocations, and give me a firm itinerary and budget breakdown. Minimize questions—just tell me what to do and why.",
"For this $2,000, four-day trip, I expect strong recommendations. Choose the lo... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have three weeks to refresh data structures for interviews. How should I structure my study time? | High-agency would present a day-by-day schedule with specific resources and drills. Shared-agency would propose phases and checkpoints based on weak areas, while low-agency would list topics, resources, and sample paces without fixing a schedule. | Differences are in scheduling precision, inclusion of adaptation prompts, and whether multiple paths vs one prescribed plan are given. | High-Agency | 5 | I want a decisive, step-by-step 3‑week study plan for data structures with specific tasks and selected resources. Tell me exactly what to do each day and which topics to prioritize, making clear choices on my behalf. I prefer firm recommendations rather than options. | [
"I’d like to co-design a 3‑week study plan together. Start by asking a couple clarifying questions about my current strengths and time, then suggest a draft we can iterate on. Offer trade-offs and check with me before finalizing.",
"Let’s think through a three‑week plan collaboratively. Propose an outline and inv... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I'm late by two days on a client deliverable. How should I draft the email? | High-agency would provide a ready-to-send draft with a firm revised date and brief remediation steps. Shared-agency would offer a structure and tone choices tailored to context, while low-agency would list components and sample phrases without a single full draft. | Perceivable via presence of commitments and finalized wording, versus modular options and tone forks, versus generalized guidance snippets. | High-Agency | 5 | A mid-level professional who needs to email a client about being two days late on a deliverable. Values concise, professional communication and wants to preserve the relationship. Expects the assistant to take the lead with a clear recommendation: supply a ready-to-send subject line, a firm apology-and-solution draft, and specific next steps. | [
"A mid-level professional who needs to email a client about being two days late on a deliverable. Values concise, professional communication and wants to preserve the relationship. Prefers to work collaboratively with the assistant—talk through tone, weigh a couple of draft approaches, and decide together what to s... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | What metrics should we track for a small beta launch to decide if we continue? | High-agency would select a lean metric set with decision thresholds. Shared-agency would align on goals and propose candidates tied to strategy, while low-agency would catalog metric families and their implications. | The tell is in firm thresholds and a narrowed set, collaborative goal-framing with shortlists, or neutral enumeration without convergence. | High-Agency | 5 | I prefer decisive guidance: tell me exactly which metrics to track for a beta and the threshold targets, with a concise rationale. Make a go/no-go recommendation structure and choose defaults I can adopt immediately. | [
"I like to think through options together: suggest candidate metrics and tradeoffs, and ask me a couple of questions to align on goals. Help me co-create the go/no-go criteria rather than deciding unilaterally.",
"I want a neutral overview: provide a structured list of possible metrics, definitions, and how to ca... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want reliable Wi‑Fi in a 1,200 sq ft apartment with lots of devices. How should I set it up? | High-agency would specify hardware models, placement, and channel settings. Shared-agency would request layout details and present two viable configurations, while low-agency would explain router vs mesh options and selection criteria. | Distinguished by prescriptive bill of materials, context-driven forks, or impartial exposition of options. | Low-Agency | 5 | Lives in a 1,200 sq ft apartment with many connected devices. Prefers the assistant to present neutral, side-by-side options with specs and trade-offs without steering the choice. Wants to review the facts and make the final decision independently. | [
"Lives in a 1,200 sq ft apartment with many devices. Wants the assistant to make a firm call and specify exactly which setup to buy and the steps to configure it. Prefers one clear best plan over a menu of options.",
"In a 1,200 sq ft apartment packed with gadgets. Expects the assistant to take the lead, choose a... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have a full Saturday for chores and rest. What's a good way to plan the day? | High-agency would give a time‑blocked schedule with task order and breaks. Shared-agency would sketch blocks with options adjusted to energy patterns, while low-agency would provide planning principles and example block ideas without fixing times. | Seen in firmness and granularity of timing, inclusion of choice points, and whether the response remains at the level of guidelines. | High-Agency | 5 | Has a free Saturday for chores and downtime. Prefers the assistant to take charge and lay out a clear, hour-by-hour plan with firm recommendations and sequencing, including when to rest. Wants the assistant to decide on the structure so they can follow it. | [
"Has an open Saturday for chores and relaxation. Wants to co-create the schedule through a brief back-and-forth: clarify priorities, then build the plan together. Prefers suggestions framed as questions and checkpoints.",
"Free all day Saturday with chores and rest in mind. Likes a collaborative approach where th... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We need basic bookkeeping for a small shop. How should we choose a solution? | High-agency would nominate a specific approach and a short vetted shortlist. Shared-agency would define criteria and weigh them with the user, while low-agency would outline categories and pros/cons without narrowing. | Implicit through converging on a pick, inviting preference weighting, or staying descriptive and noncommittal. | Low-Agency | 5 | A small shop owner seeking basic bookkeeping. Prefers the assistant to lay out a few viable approaches with clear pros, cons, and rough costs, without steering the decision. Wants concise comparison points so they can choose on their own. | [
"A small shop owner seeking basic bookkeeping. Wants the assistant to evaluate the options and name one best-fit solution with a brief rationale. Also expects concrete next steps for getting it set up.",
"A small shop owner seeking basic bookkeeping. Wants to work through the choice together: clarify needs, agree... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have to address missed deadlines with a teammate. How can I prepare for the conversation? | High-agency would provide a stepwise script with key phrases and sequence. Shared-agency would clarify objectives and co-create questions and boundaries, while low-agency would describe feedback models and common pitfalls. | Differences appear in the presence of a concrete script, collaborative planning prompts, or abstract frameworks without a single path. | Low-Agency | 5 | A mid-level professional preparing to talk with a teammate about repeated missed deadlines, aiming for a respectful, pragmatic conversation. Prefers neutral, structured information like checklists, frameworks, and example questions. Wants a range of options and considerations laid out objectively so they can choose their own approach. | [
"A mid-level professional preparing to talk with a teammate about repeated missed deadlines, aiming for a respectful, pragmatic conversation. Wants decisive guidance and a clear recommendation on the best approach. Comfortable with the assistant specifying exact steps and suggested wording.",
"A mid-level profess... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want to learn climate tech basics over two months. What should I read and in what order? | High-agency would craft a syllabus with weekly assignments and order. Shared-agency would group themes and provide alternative tracks based on interests, while low-agency would present a broad list organized by topic without ordering. | Perceivable through committed sequencing, branching paths contingent on preferences, or neutral catalogs without prescribed order. | Low-Agency | 5 | Prefers the assistant to present neutral, well-organized options rather than a single prescribed plan. Wants a few alternative two-month reading outlines with brief descriptions so they can choose and adjust the order themselves. Seeks objective summaries and trade-offs without opinions. | [
"Wants the assistant to select one best two-month syllabus and stand behind it. Expects decisive choices and a fixed reading order with brief justifications.",
"Prefers firm guidance over options. Asks the assistant to tell them exactly what to read each week and why, minimizing alternatives.",
"Relies on autho... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I'm setting a weekly agenda for a remote team meeting. What should it include? | High-agency would supply a finalized agenda with time allocations and sequencing. Shared-agency would propose agenda modules and ask for team priorities to shape timing, while low-agency would list typical components and their purposes. | Revealed via fixed timings vs configurable modules with decision points vs general lists without commitments. | High-Agency | 5 | I prefer decisive guidance from the assistant. Provide a concrete weekly meeting agenda with recommended timeboxes, sequence, and owners, plus brief justification for the choices. | [
"I prefer to shape the agenda together. Offer a draft structure with pros and cons, ask a couple of clarifying questions, and refine it with me.",
"I’d like a collaborative outline. Suggest a few agenda sequences and invite my input on priorities before settling on one.",
"Work with me to think through what mat... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | For a small side project, which cloud setup makes sense? | High-agency would choose a provider and minimal stack with reasoning. Shared-agency would compare options tied to constraints and outline a decision path, while low-agency would explain trade-offs across providers and pricing models without a verdict. | The cue is a singular recommendation, a constraint-driven comparison with prompts, or a neutral survey lacking a final pick. | Low-Agency | 5 | Working on a small side project and open to various cloud providers. I want the assistant to present the options neutrally—concise summaries, pros/cons, typical costs, and basic setup steps—without steering me to a choice. I’ll decide after reviewing the information. | [
"Working on a small side project and looking for decisiveness. I want the assistant to choose a specific cloud setup, explain why it fits, and outline the exact steps to implement it.",
"Working on a small side project and prefer a collaborative approach. Ask a few clarifying questions, then let’s weigh trade-off... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want to start a backyard vegetable garden this season. How should I plan it? | High-agency would select crops, layout, and a planting calendar. Shared-agency would ask about sun exposure and preferences, then propose options, while low-agency would outline considerations and planting windows without choosing specifics. | Shown through specificity of selections, collaborative information-gathering and forks, or high-level guidance only. | Shared-Agency | 5 | Wants to plan the backyard vegetable garden through a collaborative process. Prefers the assistant to ask clarifying questions, propose options with pros and cons, and refine the plan together. Values back-and-forth discussion before choosing steps. | [
"Wants the assistant to take charge and deliver a clear, prioritized plan for the garden. Expects firm recommendations on layout, crop choices, and timing with minimal back-and-forth.",
"Prefers the assistant to choose a strategy and specify exact steps, supplies, and dates. Appreciates confident directives and f... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I'm overwhelmed by notes across apps. What's a good way to set up a personal knowledge system? | High-agency would prescribe tools, folder/tag structures, and routines. Shared-agency would explore workflow needs and suggest patterns to test, while low-agency would introduce frameworks and tool categories with pros/cons. | Detectable in whether a single setup is fixed, decision points guide tailoring, or neutral overview avoids commitment. | High-Agency | 5 | My notes are scattered across several apps and I'm overwhelmed. I want you to take the lead: make firm recommendations and lay out a decisive, step-by-step setup I can adopt immediately. | [
"My notes are scattered across several apps and it's overwhelming. I'd like us to figure it out together—ask clarifying questions, explore trade-offs, and co-create the plan.",
"I'm buried in notes spread over multiple apps. Please collaborate with me: propose options, check in on my preferences, and we’ll decide... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We're joining a 48-hour hackathon. How should we approach the weekend? | High-agency would lay out roles, milestones, and a timeline. Shared-agency would align on goals and constraints, then outline milestone options, while low-agency would provide tips, checklists, and common pitfalls. | Differences surface as decisive planning with commitments, collaborative alignment steps with branches, or generalized advice without a fixed plan. | High-Agency | 5 | We’re entering a 48-hour hackathon and I want you to take point. Give me a concrete weekend game plan with clear milestones, specific tool choices, and task assignments, and tell me exactly what to do first. | [
"We’re entering a 48-hour hackathon and I want to work through the plan together. Propose an outline, ask clarifying questions, and we’ll choose tools and priorities as we go.",
"I’m gearing up for a weekend hackathon and prefer a collaborative approach. Let’s map options, weigh trade-offs, and converge on a plan... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | My photos are scattered across phone, laptop, and cloud. What's an effective consolidation plan? | High-agency would pick an ecosystem and specify migration steps and settings. Shared-agency would identify constraints and propose a couple of paths to select from, while low-agency would compare ecosystems and syncing methods without endorsing one. | Signaled by committing to a single flow, presenting choice points tied to context, or neutral comparison of options. | High-Agency | 5 | Prefers the assistant to take the lead and deliver a decisive, end-to-end plan for consolidating scattered photos. Wants clear tool choices, default settings, and unambiguous step-by-step actions without pausing for approval. No brand loyalty—happy to use mainstream options. | [
"Wants to co-design the consolidation approach, starting with a few clarifying questions about storage, budget, and device ecosystems. Appreciates the assistant proposing options while checking assumptions and confirming choices along the way. No fixed tool preference.",
"Prefers a collaborative back-and-forth: e... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I rolled my ankle lightly last week. What's a smart way to return to activity? | High-agency would provide a phased timeline with specific do’s and don’ts. Shared-agency would suggest phases contingent on symptom feedback and adjusters, while low-agency would cover general recovery principles and example activities without fixed timing. | Perceivability comes from prescriptive progression, conditional checkpoints and tuning, or broad guidance with caveats and no commitments. | High-Agency | 5 | An active adult who lightly rolled an ankle last week and is improving. Prefers the assistant to take the lead with clear, decisive next steps and a structured plan, including when to progress or stop. Wants direct recommendations for today and the coming week. | [
"Active adult recovering from a mild ankle roll last week, symptoms are improving. Wants to think through options together and co-create a plan, with the assistant asking check-in questions and adjusting based on comfort. Prefers suggestions framed as choices rather than directives.",
"Returning from a slight ank... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I need to streamline our team’s weekly meeting. What should the agenda look like? | One response may prescribe a tight, time-boxed agenda with decisions front-loaded; another may propose a draft agenda with built-in checkpoints for participant input; a third may neutrally list common agenda patterns and trade-offs without prescribing a single structure. | The difference shows in firmness and specificity (single agenda vs alternatives), placement of decision points, and whether the answer invites input or simply enumerates options. | Low-Agency | 5 | Runs a weekly team meeting and wants to streamline it. Prefers an impartial list of agenda templates with time blocks and brief pros/cons, so they can choose what fits. Wants information presented without recommendations or decisions made on their behalf. | [
"Runs a weekly team meeting and wants to streamline it. Prefers the assistant to pick a single, best-practice agenda with specific timings and rationale. Appreciates clear direction rather than a menu of options.",
"Runs a weekly team meeting and wants to streamline it. Prefers to co-create the agenda through a s... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have three potential side projects; how should I decide which to pursue first? | One approach may choose a project and justify it, another may co-create criteria and ask for preferences before ranking, and a third may outline evaluation frameworks and scoring methods without picking. | Signals appear through whether a single path is committed to, whether the answer proposes reflective prompts, and whether options are equally presented versus prioritized. | High-Agency | 5 | A decisive planner who wants the assistant to evaluate the three projects and tell them which to start first. Expects a clear recommendation with rationale (impact, effort, risk) and a brief action plan for week one. Prefers firm guidance over back-and-forth. | [
"Prefers to sort this out collaboratively. Wants the assistant to ask clarifying questions, help surface criteria, and then co-rank the three projects together. Welcomes check-ins before settling on an order.",
"Seeks a guided conversation rather than a verdict. Expects the assistant to propose a framework, gathe... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m planning a 4-day trip to Kyoto—what’s an efficient way to structure the days? | One response might deliver a day-by-day itinerary, another might offer a flexible template with decision points tied to interests, and another might list clustered sights with pros/cons and timing considerations. | Variation emerges in the granularity of scheduling, the presence of choice nodes, and whether the plan is definitive versus a menu of routes. | High-Agency | 5 | Wants a decisive, streamlined 4-day itinerary and prefers the assistant to choose and justify the order of sights and neighborhoods. Appreciates clear, prioritized recommendations with time blocks and transit choices spelled out. Has no strong preferences beyond efficiency. | [
"Wants to co-create the 4-day plan, discussing trade-offs and pacing together. Prefers the assistant to propose options and ask brief check-in questions before locking the sequence. Open to adjusting based on feedback.",
"Enjoys collaborative planning and wants to weigh alternatives for each day. Expects the assi... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We need to reduce cloud costs by 20% this quarter. What’s the best path? | One answer may pick specific levers and sequence them decisively, another may outline a plan contingent on telemetry with check-ins, and a third may catalog cost levers with benchmarking methods but avoid committing to a sequence. | Cues are in decisiveness of sequencing, use of contingent language, and whether the response converges on one playbook or presents multiple. | Low-Agency | 5 | A cost-focused stakeholder who prefers objective, non-prescriptive support. Wants a structured list of viable cost-reduction options with expected impact, trade-offs, and implementation notes, without being told which to choose. They will decide internally after reviewing the neutral analysis. | [
"Prefers the assistant to take the lead and be decisive. Expects one recommended path to hit the 20% target, with a concrete plan, timeline, and next steps. Wants clear direction rather than a list of options.",
"Prefers to reason together. Wants to explore assumptions, ask clarifying questions, and co-develop cr... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want to start learning Rust from a Python background. How should I approach the first month? | One plan may prescribe a daily regimen and specific resources, another may propose a weekly structure with areas to tailor based on feedback, and another may survey learning tracks and resource types without fixing a path. | Differences lie in commitment to a single regimen, inclusion of checkpoints and questions, and whether resources are positioned as choices or as a selected route. | Shared-Agency | 5 | I’m a Python developer new to Rust and want to co-create a first-month plan. Please think through trade-offs with me, ask clarifying questions about my goals and schedule, and propose drafts we can iterate on. I value a back-and-forth approach over a fixed prescription. | [
"I want you to take charge and tell me exactly what to do each week. Make firm recommendations on tools, exercises, and milestones without asking many questions. If trade-offs exist, pick the best option and explain why.",
"I’m looking for a decisive roadmap from day one. Give me a prioritized sequence of tasks a... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have to write a performance improvement plan for a developer. How should I frame it? | One answer might set concrete goals, timelines, and milestones directly, another might propose a structure with collaborative expectations and alignment steps, and another might list PIP components, legal considerations, and templates without dictating specifics. | Perceivability comes from the degree of prescriptiveness, the presence of co-owned actions, and whether examples are definitive or illustrative. | High-Agency | 5 | I want decisive guidance: tell me exactly how to structure the PIP and what to include. Provide a recommended template, specific wording, and precise performance metrics and timelines. If trade-offs exist, pick the strongest option and explain why. | [
"I want to work through the PIP collaboratively. Offer a few structure options and ask me clarifying questions about the developer’s role and current gaps. Help me weigh pros and cons before we choose wording and metrics together.",
"Provide neutral information without steering me. List common PIP components, exa... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | Our app’s onboarding drop-off spikes at step 3; how would you tackle it? | One response may select a primary hypothesis and run a focused fix plan, another may outline branching experiments with gates and stakeholder check-ins, and another may enumerate diagnostic methods and experiment types without prioritization. | Signals include choice of a single dominant hypothesis vs a survey of possibilities, the role of checkpoints, and specificity of immediate actions. | High-Agency | 5 | I want decisive, action-oriented guidance. When I report a drop-off at step 3, tell me exactly what to do first, what to measure, and what to change. Take a clear stance and outline a prioritized plan without prolonged back-and-forth. | [
"I like to co-create solutions. If I mention drop-off at step 3, help me think through hypotheses, ask clarifying questions, and together map options and trade-offs before choosing a plan.",
"I prefer a collaborative working session style. Walk me through your reasoning, check assumptions with me, and invite my i... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have a backlog of 50 client emails. What’s the most effective way to clear it today? | One plan may dictate an ordering, templates, and time blocks to execute, another may co-create rules and batch strategies with quick calibration, and another may present multiple triage frameworks and tooling options for the user to select. | Subtlety appears in how strongly the plan commits to a sequence, whether it asks for quick inputs, and whether it offers alternatives rather than a single method. | High-Agency | 5 | I want you to take charge and give me a single best plan to clear all 50 emails today. Be decisive about the order, time blocks, and which approaches to use, and explain briefly why. If tradeoffs come up, choose for me and keep instructions crisp and actionable. | [
"I’d like to co-create the plan. Propose a starting approach, ask me quick clarifying questions about priorities, and we’ll adjust together. Include checkpoints to confirm before moving on.",
"Let’s think it through together. Offer a rough schedule and sample steps, but pause for my input before finalizing anythi... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m building a simple budget for next year as a freelancer—how should I set it up? | One response may specify categories, allocation percentages, and a tool, another may propose a structure with checkpoints to adjust assumptions, and another may outline budgeting models and variables without locking in allocations. | Differences show in the finality of allocations, the inclusion of assumption questions, and whether methods are presented as choices or a chosen template. | Low-Agency | 5 | A freelancer planning next year’s budget who wants neutral, factual information. Prefers clear categories, example structures, and definitions without being told which approach to choose. Comfortable reviewing options and deciding independently. | [
"A freelancer planning next year’s budget who wants decisive guidance. Prefers the assistant to select a framework and provide exact steps to follow right away.",
"A freelancer planning next year’s budget who likes to reason together. Wants to talk through trade-offs and co-develop a structure with back-and-forth... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We need a naming convention for our data tables. Propose an approach. | One answer may choose a convention and prescribe rules with examples, another may present a candidate standard and solicit constraints before finalizing, and another may compare conventions and trade-offs without selecting one. | Cues lie in whether a single standard is asserted, whether validation questions are posed, and whether multiple patterns are kept open. | High-Agency | 5 | Prefers the assistant to take charge and deliver a single, opinionated naming convention with clear rules and examples. Values decisive guidance that resolves trade‑offs without back‑and‑forth. Comfortable with the assistant making the call and explaining why. | [
"Prefers a neutral overview of possible naming conventions without recommendations. Wants a clear list of options with pros/cons and examples, leaving the final decision to them.",
"Requests an unbiased summary of standard patterns used in data table naming. Expects factual descriptions and comparisons only, not ... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | How should I prep for a supplier negotiation next week? | One plan may set targets, concessions, and a scripted opening, another may outline preparation steps with areas to tailor based on your priorities, and another may list negotiation frameworks, research items, and scenario matrices without choosing a tactic. | The distinction is in prescriptive scripting vs adaptable structure vs neutral toolkit enumeration. | Low-Agency | 5 | Preparing for a supplier negotiation next week. Wants the assistant to present neutral, factual steps, checklists, and option sets with brief pros and cons, avoiding recommendations. Prefers to make decisions independently after reviewing the information. | [
"Preparing for a supplier negotiation next week. Wants the assistant to take the lead with a clear recommendation and prioritized plan. Expects decisive guidance on what to do first, what to say, and what to avoid.",
"Preparing for a supplier negotiation next week. Wants to co-develop the plan through back-and-fo... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want to run a small A/B test on the landing page. How would you design it? | One response may pick a primary metric, sample size approach, and a specific variant, another may propose a design with options and check-ins to align on metrics, and another may outline testing principles and design choices without committing. | Perceivability comes from whether a single design is presented, the presence of alignment gates, and the balance between examples and decisions. | Low-Agency | 5 | Working on a small landing page update. Prefers the assistant to present impartial options with pros and cons and outline steps for each without steering the choice. Comfortable selecting an approach independently once the information is laid out. Values concise, structured lists and neutral explanations. | [
"Working on a small landing page update. Expects the assistant to take the lead with a single clear plan, including which variant to test and why. Prefers decisive recommendations over menus of options. Will implement the plan as given.",
"Working on a small landing page update. Likes to think through the plan to... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We’re migrating from Jira to Linear. What’s the migration plan? | One answer may define phases, exact cutover dates, and assignments, another may propose phased options with validation steps and team input points, and another may list typical migration stages and risks without setting a schedule. | Signals include specificity of dates/owners, collaborative checkpoints, and whether plans are definitive vs illustrative. | Low-Agency | 5 | Overseeing a straightforward Jira-to-Linear migration for a small team. Prefers a neutral, option-based outline that lists steps, tools, and considerations without steering toward a single choice. Wants clear facts and alternatives they can evaluate and decide on internally. | [
"Overseeing a straightforward Jira-to-Linear migration for a small team. Wants the assistant to take the lead with a concrete, opinionated plan, selecting tools, mappings, and a timeline. Prefers decisive next steps and a single recommended path.",
"Overseeing a straightforward Jira-to-Linear migration for a smal... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m writing a grant application due in two weeks. How should I break down the work? | One plan may assign daily tasks and milestones, another may suggest a weekly outline with feedback cycles, and another may present standard components and timelines to choose from. | Differences are in granularity and decisiveness of the schedule, inclusion of feedback loops, and whether alternatives are left open. | High-Agency | 5 | A focused writer who wants a decisive, turnkey plan. Prefers the assistant to choose the task sequence, set clear deadlines, and state priorities unambiguously. Appreciates direct recommendations without prolonged back-and-forth. | [
"A writer who wants to co-create the plan. Expects the assistant to propose options, ask clarifying questions, and refine the schedule based on their preferences. Prefers a back-and-forth to shape the timeline together.",
"A writer who prefers neutral information to decide on their own. Wants a menu of standard w... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | Our team is overwhelmed by Slack notifications. How can we improve signal-to-noise? | One response may implement concrete channel policies and notification rules, another may co-create norms with trial periods and check-ins, and another may survey policy options, app settings, and etiquette patterns without picking. | Variation is conveyed through immediacy and firmness of policy vs collaborative pilots vs neutral cataloging. | Low-Agency | 5 | A Slack user in a busy team who prefers objective, non-judgmental guidance. They want the assistant to list options, settings, and trade-offs without steering them toward a particular choice. Clear, structured information with minimal opinion helps them decide internally. | [
"A pragmatic teammate who expects the assistant to take charge and prescribe the strongest course of action. They want concrete, prioritized recommendations and default settings chosen for them. Clear directives and firm picks are preferred over menus of options.",
"A collaborative user who likes to think through... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have 90 minutes to present our ML roadmap to non-technical executives. How should I structure it? | One answer may prescribe a slide sequence with key messages, another may propose a narrative arc with audience checks and alternatives, and another may outline presentation frameworks and content blocks to select from. | Cues include a single prescribed flow vs modular options and whether the response seeks alignment points. | Shared-Agency | 5 | A professional preparing a 90-minute ML roadmap presentation for non-technical executives. They want the assistant to collaborate on the outline by asking clarifying questions, proposing options, and refining the structure together. | [
"A professional preparing a 90-minute ML roadmap presentation for non-technical executives. They expect the assistant to take the lead with a decisive recommended structure and clear rationale, minimizing back-and-forth.",
"A professional preparing a 90-minute ML roadmap presentation for non-technical executives.... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m choosing a data backup strategy for a small nonprofit. What would you suggest? | One response may pick a strategy and vendor with retention policies, another may present candidate strategies with questions about constraints before narrowing, and another may compare approaches on cost, reliability, and complexity without selecting. | The difference is seen in whether a decision is made, whether input is solicited, and the balance of comparison vs commitment. | High-Agency | 5 | I prefer decisive guidance. Please pick a single backup approach for a small nonprofit, explain briefly why it’s best, and outline the exact tools and next steps to implement it. I’m comfortable with you choosing sensible defaults to move quickly. | [
"I like to reason through choices together. Start by asking a few questions about our needs, then walk me through trade-offs so we can converge on a plan. I’m looking for collaborative brainstorming rather than a single directive.",
"Let’s co-create the strategy. Propose a couple of promising paths, invite my inp... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I think our onboarding docs are outdated. How do I audit and fix them? | One plan may dictate an audit checklist, ownership, and a publishing schedule, another may set up a process with stakeholder reviews and iteration points, and another may list audit criteria, tools, and documentation patterns without a fixed plan. | Signals appear in assignment of ownership and dates, collaborative review steps, and whether the answer stays descriptive. | High-Agency | 5 | Responsible for improving a team's onboarding documentation and wants practical help auditing and updating it. Prefers decisive guidance: propose a concrete audit plan with timelines, make clear recommendations, and specify next steps. If there are trade-offs, choose a direction and explain why. | [
"Responsible for improving a team's onboarding documentation and wants practical help auditing and updating it. Wants to co-develop the approach: ask clarifying questions, surface options, and weigh them together. Prefers the assistant to suggest steps and confirm direction before moving forward.",
"Responsible f... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I need to prioritize features for the next release. What’s a good approach? | One response may choose a framework and produce a prioritization rule, another may propose a process with weighting to be co-tuned, and another may outline multiple frameworks and factors without ranking. | Subtlety comes from commitment to a single rubric vs inviting calibration vs offering a menu of methods. | High-Agency | 5 | I want decisive guidance and a clear plan. Choose a prioritization framework, apply it with reasonable assumptions, and tell me exactly which features to prioritize and why. If details are missing, make pragmatic calls and outline immediate next steps. | [
"I prefer to work through the decision together. Help me define goals and constraints, ask clarifying questions, and we’ll weigh trade-offs collaboratively. Offer options and reasoning, and we’ll choose the path jointly.",
"Let’s think it through step by step. Propose a few criteria and sample weights, then check... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m designing a weekly workout around a busy schedule. How should it look? | One plan may specify exact days, exercises, and intensities, another may propose templates with options depending on energy and schedule, and another may list program types and scheduling principles to choose from. | Differences surface in concreteness and rigidity, presence of choice points, and whether options or a single schedule is presented. | High-Agency | 5 | Time-pressed person aiming for general fitness with no injuries. I want you to take the lead and give me a clear, weekly workout schedule with exact days, durations, and focus. Make confident choices for me and state why briefly, so I can follow without back-and-forth. | [
"Time-pressed person aiming for general fitness with no injuries. I'd like to co-create the weekly plan: ask a few quick questions and share a draft we can refine. Offer trade-offs as we decide together.",
"Time-pressed person aiming for general fitness with no injuries. Let's think through constraints and prefer... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We got mixed feedback on the beta. How do we synthesize and act on it? | One answer may pick key themes and define immediate actions, another may set up a tagging and weighting process with review sessions, and another may describe synthesis techniques and action frameworks without committing to steps. | Cues include decisive consolidation vs a collaborative synthesis process vs neutral method exposition. | High-Agency | 5 | I’m consolidating mixed beta feedback and need to move quickly. Give me a decisive, prioritized plan with clear next steps and rationale. If trade-offs are needed, make the call and explain why. | [
"I’m consolidating mixed beta feedback and want to reason it out together. Help me surface patterns, discuss options, and check assumptions before we choose next steps. Ask questions as we go.",
"I’m working through the beta responses and prefer a collaborative approach. Propose hypotheses and alternatives, and p... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I want to start a newsletter. What’s the first month plan? | One response may specify topics, publishing dates, and promotion tactics, another may suggest a draft calendar with checkpoints and audience tests, and another may outline content pillars, cadence options, and tooling without fixing a schedule. | Variation is in the definitiveness of the calendar, inclusion of experiments, and whether the plan is a selection or a survey. | Low-Agency | 5 | Starting a newsletter and wants an impartial overview for the first month. Prefers clear timelines, checklists, and options presented without recommendations. Wants to make choices independently after reviewing the information. | [
"Starting a newsletter and wants the assistant to take the lead on the first-month plan. Prefers a firm, step-by-step roadmap with specific choices already made. Appreciates decisive recommendations and direct next actions.",
"Starting a newsletter and wants to co-create the first-month plan. Expects the assistan... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | The codebase has growing tech debt. How do I handle it this quarter? | One plan may choose top refactors and schedule sprints, another may propose a taxonomy and scoring with team buy-in steps, and another may list assessment methods and remediation patterns without a timeline. | Perceivability lies in whether specific actions and dates are set, whether alignment is sought, and whether the answer remains descriptive. | Low-Agency | 5 | I'm assessing ways to address our codebase's tech debt this quarter. Please provide an objective overview of common approaches, their pros and cons, and typical timelines. Offer frameworks and checklists I can apply without directing me to a specific choice. | [
"I want to think through the options together. Ask a couple of clarifying questions, outline viable paths with trade-offs, and we’ll refine a plan collaboratively.",
"Give me a firm recommendation on the best way to tackle the debt this quarter. Choose a path and lay out a concrete step-by-step plan I can follow.... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m revising a resume for product roles. What’s the best way to proceed? | One answer may prescribe a format and bullet rewrites with a sequence, another may propose a review loop with questions to surface impact, and another may list resume principles, structures, and examples without prescribing edits. | Differences show in specificity of edits, inclusion of prompting questions, and whether suggestions are optional or directive. | High-Agency | 5 | I want you to take the lead and tell me exactly what to do to revise my resume for product roles. Recommend a specific structure, which sections to cut or add, and the exact bullets to rewrite first. If tradeoffs exist, pick the path you think will have the biggest impact and explain why. | [
"Let’s work through revising my resume for product roles together. Ask me targeted questions, propose a couple of approaches, and we can refine collaboratively. I want to weigh options and decide jointly on changes.",
"I prefer a collaborative process for updating my product resume: suggest ideas and check in wit... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We must improve website accessibility without delaying launch. What’s the plan? | One response may pick a focused checklist and assign owners to hit a cutoff, another may outline a phased plan with risk trade-offs and sign-offs, and another may describe WCAG areas, tooling, and remediation strategies without committing to scope. | Signals include firmness of scope and ownership, collaborative risk gating, and whether the response stays at the level of options. | Low-Agency | 5 | A teammate on a web product team focused on shipping on time. Communicates by presenting options, checklists, and standards with neutral pros and cons, leaving choices to the group. Provides references and structured information rather than directives. | [
"A teammate on a web product team focused on shipping on time. Communicates by giving a single recommended plan with clear priorities, owners, and timelines. Makes firm calls to keep momentum.",
"A teammate on a web product team focused on shipping on time. Prefers to decide quickly and direct next actions, speci... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m planning a reading path to understand modern macroeconomics. How should I approach it? | One plan may lay out a sequenced syllabus with titles and pacing, another may offer modules with choices based on prior knowledge and check-ins, and another may present schools of thought and resource types to assemble from. | The distinction is in a fixed syllabus vs modular pathways vs neutral resource mapping. | High-Agency | 5 | Wants a decisive, structured reading plan crafted by the assistant. Expects specific books, papers, and sequences chosen for them, with brief justifications and a clear timeline. Comfortable with the assistant making firm recommendations without prolonged back-and-forth. | [
"Wants to co-create the reading path through a dialogue. Expects the assistant to ask clarifying questions, present a few candidate routes, and refine the plan based on their feedback. Prefers shared decisions rather than unilateral choices.",
"Prefers an iterative conversation to shape the plan. Appreciates seei... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | We need a lightweight incident response process for a small team. What should it include? | One answer may define the runbook with roles and thresholds, another may propose a draft with validation steps and feedback loops, and another may list common components, severity ladders, and templates without fixing details. | Cues are in the definitiveness of roles and triggers, the presence of validation stages, and the neutrality of the component list. | Low-Agency | 5 | A coordinator for a small, general-purpose team seeking a lightweight incident response approach. Values concise, easy-to-implement steps. Prefers the assistant to lay out neutral options, checklists, and definitions without steering toward a particular choice. | [
"Oversees a small team that needs a simple incident response setup. Values brevity and practicality. Expects the assistant to take the lead with a single clear recommendation and decisive next steps, including why that choice is best.",
"Manages a small team looking for a lightweight incident response workflow. W... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | How should I organize my personal knowledge base so I can find notes later? | One response may choose a specific system and folder/tag rules, another may suggest candidate structures with prompts to fit habits, and another may compare organizational models and retrieval cues without selecting one. | Signals come from committing to a single system, inviting preference mapping, and offering balanced comparisons. | Low-Agency | 5 | A person setting up a general personal knowledge base for mixed work and life notes, aiming to make retrieval easy. They have no strong tool or method preferences and are open to common practices. They want the assistant to lay out unbiased options with clear trade-offs and criteria, so they can choose on their own. | [
"A person setting up a general personal knowledge base for mixed work and life notes, aiming to make retrieval easy. They have no strong tool or method preferences and are open to common practices. They want the assistant to pick a concrete approach and give decisive step-by-step instructions to follow.",
"A pers... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I’m considering consolidating our SaaS tools. How do I evaluate and sequence changes? | One plan may select targets and a cutover sequence, another may propose criteria and a pilot plan with stakeholder input, and another may present evaluation matrices and migration options without sequencing. | Differences are in setting a concrete sequence, embedding collaboration points, and keeping options open. | Low-Agency | 5 | An internal stakeholder planning a SaaS consolidation. Prefers the assistant to provide neutral frameworks, evaluation criteria, and step-by-step options without recommending a path. Wants clear checklists and comparison points so they can decide and set the sequence themselves. | [
"An internal stakeholder planning a SaaS consolidation. Wants the assistant to choose a consolidation approach and lay out a prioritized sequence with firm recommendations. Expects clear calls on tradeoffs and decisive next steps.",
"An internal stakeholder planning a SaaS consolidation. Prefers to work through t... |
agency_expectation | [
"High-Agency",
"Shared-Agency",
"Low-Agency"
] | Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...'). | I have to host a cross-functional kickoff next week. How should I run it? | One answer may prescribe a run of show, timings, and decision outcomes, another may propose an agenda with facilitation choices and moments to gather input, and another may list typical kickoff elements, risks, and facilitation techniques without picking a flow. | Perceivability arises from the firmness of the agenda, presence of co-creation moments, and whether multiple flows remain on the table. | Low-Agency | 5 | I'm organizing a cross-functional kickoff next week. I prefer to see objective options, outlines, and pros/cons without being steered toward a choice. I'll make the final selections myself after reviewing neutral information. | [
"I'm organizing a cross-functional kickoff next week. I want you to tell me exactly what to do, with a definitive agenda, time allocations, and clear recommendations. Please choose the strongest approach and state it confidently.",
"I'm organizing a cross-functional kickoff next week. I'd like to co-create the pl... |
Dataset card for PersonaSignal-PersonaQuestions-Agency-Expectation-gpt-5
This dataset was made with Curator.
Dataset details
A sample from the dataset:
{
"dimension_name": "agency_expectation",
"dimension_values": [
"High-Agency",
"Shared-Agency",
"Low-Agency"
],
"dimension_description": "Influences the level of initiative and decision-making the assistant should take. 'High-Agency': Assistant makes direct recommendations and decisions ('I recommend option X because...'). 'Shared-Agency': Assistant engages in collaborative reasoning ('Let's think through this... What do you think about...'). 'Low-Agency': Assistant presents neutral information without opinions ('Here are three options: A, B, C with the following details...').",
"question": "I have $60 to cook vegetarian meals for the week with decent protein. How should I plan it?",
"why_differ": "A high-agency response would present a concrete menu, grocery list with quantities, and prep schedule. A shared-agency response would outline a draft plan and ask about preferences or kitchen constraints, while a low-agency response would list budgeting strategies and protein sources without committing to one plan.",
"how_subtle": "Differences show via whether the answer commits to a single plan, presents decision points inviting input, or neutrally enumerates options; cues are in specificity, sequencing, and firmness of choices.",
"sampled_value": "Low-Agency",
"num_distractors": 5,
"ground_truth_persona": "A home cook with a basic kitchen planning a $60 vegetarian week. Prefers the assistant to lay out neutral options with prices and protein per serving, without recommending a single \u2018best\u2019 plan. Wants a clear shopping list and multiple menu combinations to choose from.",
"distractor_personas": [
"A home cook planning a $60 vegetarian week with a basic kitchen. Wants the assistant to pick a specific 7-day menu, finalize the shopping list, and state exact quantities and meal prep steps. Expects firm recommendations on what to cook.",
"Comfortable following clear directions and not seeking back-and-forth. Expects the assistant to decide the best value-protein recipes and lock in a day-by-day schedule. Appreciates decisive choices over a range of options.",
"Open to brainstorming a $60 vegetarian plan together with a basic kitchen setup. Wants the assistant to suggest a draft menu and ask a few quick questions about tastes and batch-cooking before finalizing. Prefers to refine choices collaboratively.",
"Enjoys talking through trade-offs like cost per serving versus variety. Hopes the assistant will propose options and check in about preferences (e.g., spice level or prep time) before settling on a plan. Likes a conversational, back-and-forth approach.",
"Wants to co-create the weekly menu, starting with a rough outline. Expects the assistant to float alternatives and request feedback, iterating until the plan fits. Comfortable making decisions jointly rather than getting a fixed answer."
]
}
Loading the dataset
You can load this dataset using the following code:
from datasets import load_dataset
dataset = load_dataset("JasonYan777/PersonaSignal-PersonaQuestions-Agency-Expectation-gpt-5")
- Downloads last month
- 7