{"questions":[{"qnum":1,"section":1,"question_type":"profile","question":"Occupation: what is your primary professional/business field?","description":"Maps to Anthropic's expertise study occupation taxonomy (Management, Business/Financial, Computer/Math, Engineering, Science, etc.). This is the user's profession, not what they're doing in this session.","options":{"a":"Business and Finance","b":"Legal Services and Management","c":"Computer Software and Development","d":"Entertainment and Media","e":"Health and Medicine","f":"Life Sciences/Biotech","g":"Sales and Marketing","h":"Education","i":"Agriculture and Food","j":"Community and Social Services","k":"Other"},"maps_to":["occupation"]},{"qnum":2,"section":1,"question_type":"single_select","question":"Expertise: how would you rate yourself in your primary field?","description":"Maps to Anthropic's 5-level expertise taxonomy — verified task completion rates (Novice 14.5% → Beginner 20.9% → Intermediate 28.3% → Advanced 29.5% → Expert 32.9%).","options":{"a":"Novice: I'm new to this field and still learning the basics","b":"Beginner: I have some experience but still need guidance","c":"Intermediate: I have a good understanding and can handle most tasks","d":"Advanced: I have extensive experience and can solve complex problems","e":"Expert: I am highly skilled and can mentor others in this field"},"maps_to":["skill_level"]},{"qnum":3,"section":1,"question_type":"single_select","question":"Hours per day: roughly how many hours per day do you spend on AI-assisted tasks?","description":"Maps to task volume for cost estimation. Time is used as a proxy for session count on the backend.","options":{"a":"Less than 1","b":"1–2","c":"3–4","d":"5–6","e":"7–8","f":"9–12","g":"12+"},"maps_to":["estimated_monthly_tasks"]},{"qnum":4,"section":1,"question_type":"multiple_select","max_options":4,"question":"Primary AI tasks: select up to 4 that take up most of your time.","description":"Maps to prompt_overhead_ratio — estimated from work mode mix. Different work modes have different context overhead patterns.","options":{"a":"Building: Creating new products, services, tools or solutions","b":"Fixing: Patching bugs, troubleshooting issues, correcting errors","c":"Operating: Using AI to operate, monitor, or maintain systems","d":"Planning: Creating plans, strategies, or roadmaps","e":"Testing: Verifying behavior, content, and correctness","f":"Understanding: Reviewing and comprehending existing code, systems, or processes","g":"Analyzing: Exploring data, generating reports, and related tasks","h":"Communicating: Writing documentation, emails, marketing materials, or presentations"},"maps_to":["prompt_overhead_ratio"]},{"qnum":5,"section":1,"question_type":"single_select","question":"AI in your field: what share of your total AI use is in your primary occupation?","description":"Cross-references with Q2 (expertise level) to evaluate Myth 3: 'domain knowledge beats better models.' If someone is Expert + uses AI mostly in their domain, they should theoretically benefit most from domain knowledge.","options":{"a":"Almost none — I mostly use AI outside my primary professional/business occupation","b":"Some — a mix of tasks inside and outside my primary professional/business occupation","c":"About half — split between my primary professional/business occupation and other areas","d":"Most — I primarily use AI for tasks related to my primary professional/business occupation","e":"Almost all — nearly everything I do with AI is in my primary professional/business occupation"},"maps_to":["belief_myth_3_expert_users"]},{"qnum":6,"section":2,"question_type":"single_select","question":"Model selection: which types do you use for most tasks?","description":"Maps to premium_model_fraction — flagship vs. mid-tier vs. economy.","options":{"a":"The cheapest model available","b":"I use cheaper models for simple tasks, pricier for complex or important work","c":"I like to pick one medium-tier model and use it for most tasks, unless the task is very complex or important","d":"I generally use the best model available; I want the best results for every task"},"maps_to":["premium_model_fraction"]},{"qnum":7,"section":2,"question_type":"single_select","question":"Common tasks (summarizing, explaining code, simple edits): which model?","description":"Maps to model-task alignment — detects over-provisioning on simple tasks. Derived from Q6 vs Q7 to identify if users over-use flagship on routine work.","options":{"a":"The cheapest model available","b":"A mix of cheaper, mid-tier, or premium models depending on the level of detail and accuracy I need","c":"I usually use the same model for most tasks (I rarely switch models)","d":"The best model available; I want the best results for every task"},"depends_on":[6],"maps_to":["model_task_mismatch"]},{"qnum":8,"section":2,"question_type":"single_select","question":"Session start: how do you typically begin a new AI session?","description":"Maps to cache_hit_rate — compact-and-continue vs. clear-and-restart. Answers combined with Q8 to derive cache behavior.","options":{"a":"I start fresh every time","b":"I tend to resume previous sessions so the AI has the right context","c":"I'll launch a new session, but provide specific context to help the AI orient itself","d":"I'll explain what I need, provide all the relevant context, and then deliver my request","e":"I'm not really sure. I just ask the AI to do what I need"},"maps_to":["cache_hit_rate"]},{"qnum":9,"section":2,"question_type":"multiple_select","max_options":3,"question":"In-session habits: select up to 3 that describe how you handle AI-completed tasks.","description":"Maps to context_bloat_rate. Captures the user's engagement and prompt management patterns during a session. Different patterns produce different levels of context accumulation.","options":{"a":"I review its plan or approach before execution and correct gaps early","b":"I provide detailed instructions upfront, then let it run with minimal intervention","c":"I re-explain or re-send prior instructions as it works on a new subtask (to keep the model on track)","d":"I review its output as it comes and correct with targeted feedback","e":"I let it run when the goal is clear, but intervene when I see it drifting","f":"I step back entirely and only intervene if the output is completely wrong","g":"I write a handoff prompt so it can continue in a new session without re-establishing context","h":"I paste references or files into the session as needed","i":"I give it a task and let it search, read files, and figure out what context it needs"},"depends_on":[8],"maps_to":["context_bloat_rate"]},{"qnum":10,"section":2,"question_type":"matrix","question":"Retry rate: for each task type below, on average how many attempts does it take before the AI gets it right?","description":"Matrix: one selection per row (task type) across 4 retry columns. Aggregated with time-weighted average. Uses a-h option labels matching Q3 for consistency.","matrix_rows":{"a":"Building: Creating new code, content, or solutions","b":"Fixing: Patching bugs, troubleshooting issues","c":"Operating: Monitoring or maintaining systems","d":"Communicating: Writing docs, emails, marketing, or presentations"},"matrix_columns":{"1":"1–2 attempts","2":"3–5 attempts","3":"6–10 attempts","4":"Agent loops"},"depends_on":[3],"maps_to":["retry_multiplier"]},{"qnum":14,"section":3,"question_type":"belief","question":"\"If I spend time learning the domain, I'll get way better results than just using a better model.\"","description":"Tests Myth 3 — expert users. This belief is directly asked because expertise level (Q2) does not reliably predict it.","options":{"a":"Completely agree — domain knowledge is the biggest factor in getting good results","b":"Agree — understanding the problem is more important than model quality","c":"Somewhat agree — both matter, but model quality can compensate for some lack of domain knowledge","d":"Disagree — a good model can compensate for lack of domain knowledge"},"maps_to":["belief_myth_3_expert_users"]},{"qnum":15,"section":3,"question_type":"single_select","question":"Top priority: what would you most like to improve in your AI use?","description":"Sets the tone and focus for the LLM-generated executive summary and recommendations. Impacts which behavioral levers the LLM prioritizes.","options":{"a":"Reduce my AI/LLM costs","b":"Improve quality of AI outputs","c":"Speed up my workflow","d":"Reduce errors and AI task retries"},"maps_to":["priority_goal"]}],"total":12}