メニュー
Complete reference for EDSL survey construction: question types, Jinja2 templating, skip/navigation rules, and memory modes.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/expectedparrot/vernon --skill surveysコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Patterns for creating EDSL AgentLists from various sources: lists, CSV, Excel, DataFrame, and programmatic combinations.
Saving EDSL objects locally and publishing them to Coop (Expected Parrot's servers).
Templates for the standard EDSL study files: survey, scenarios, agents, models, and create_results.py.
SKILL.md を表示中
SOC 職業分類に基づく
| name | surveys |
| description | Complete reference for EDSL survey construction: question types, Jinja2 templating, skip/navigation rules, and memory modes. |
Consolidated reference for building surveys with EDSL: question types, Jinja2 templating, skip/navigation rules, memory modes, helper utilities, and visualization.
from edsl import Survey, QuestionFreeText, QuestionMultipleChoice
q1 = QuestionFreeText(
question_name="name",
question_text="What is your name?"
)
q2 = QuestionMultipleChoice(
question_name="color",
question_text="What is your favorite color?",
question_options=["Red", "Blue", "Green", "Yellow"]
)
q3 = QuestionFreeText(
question_name="why_color",
question_text="Why do you like that color?"
)
survey = Survey([q1, q2, q3])
Survey is immutable - each operation returns a new Survey instance:
survey = Survey()
survey = survey.add_question(QuestionFreeText(
question_name="q1",
question_text="First question?"
))
survey = survey.add_question(QuestionFreeText(
question_name="q2",
question_text="Second question?"
))
# Add at specific index
survey = survey.add_question(new_question, index=1)
Instructions are displayed to respondents between questions:
from edsl import Survey, Instruction
instruction = Instruction(
text="Please answer the following questions honestly.",
name="intro"
)
survey = survey.add_instruction(instruction)
survey = survey.add_instruction(
Instruction(text="Now for some demographic questions...")
)
IMPORTANT:
Instructiontext is NOT checked for scenario field references. If your scenarios have fields that must be used, reference them in a question'squestion_text, not in an Instruction. EDSL'sJobsCompatibilityErroronly looks at question text.
All questions require question_name (a valid Python identifier) and question_text.
To see the question types available:
from edsl import Question
Question.available()
Open-ended text responses without constraints.
from edsl import QuestionFreeText
q = QuestionFreeText(
question_name="feedback",
question_text="What do you think about our service?"
)
Single selection from a predefined list of options.
from edsl import QuestionMultipleChoice
q = QuestionMultipleChoice(
question_name="color",
question_text="What is your favorite color?",
question_options=["Red", "Blue", "Green", "Yellow"]
)
Multiple selections from a predefined list (checkbox-style).
from edsl import QuestionCheckBox
q = QuestionCheckBox(
question_name="features",
question_text="Which features do you use? (Select all that apply)",
question_options=["Feature A", "Feature B", "Feature C", "Feature D"],
min_selections=1,
max_selections=3
)
Numeric responses with optional min/max bounds.
from edsl import QuestionNumerical
q = QuestionNumerical(
question_name="age",
question_text="How old are you?",
min_value=0,
max_value=120
)
Simple binary yes/no question (derived from MultipleChoice).
from edsl import QuestionYesNo
q = QuestionYesNo(
question_name="consent",
question_text="Do you agree to participate in this survey?"
)
# Options are automatically ["Yes", "No"]
Linear scale with customizable range and endpoint labels.
from edsl import QuestionLinearScale
q = QuestionLinearScale(
question_name="satisfaction",
question_text="How satisfied are you with our service?",
question_options=[1, 2, 3, 4, 5],
option_labels={1: "Very Unsatisfied", 5: "Very Satisfied"}
)
Standard 5-point Likert scale (agree/disagree).
from edsl import QuestionLikertFive
q = QuestionLikertFive(
question_name="statement_agree",
question_text="I find the product easy to use."
)
# Options: Strongly disagree, Disagree, Neutral, Agree, Strongly agree
Response as a list of items.
from edsl import QuestionList
q = QuestionList(
question_name="top_movies",
question_text="List your top 3 favorite movies.",
max_list_items=3
)
Ranking/ordering items by preference.
from edsl import QuestionRank
q = QuestionRank(
question_name="priority",
question_text="Rank these features by importance (1 = most important):",
question_options=["Speed", "Security", "Price", "Support"],
num_selections=4
)
Grid-based responses with rows (items) and columns (options).
from edsl import QuestionMatrix
q = QuestionMatrix(
question_name="product_ratings",
question_text="Rate each product on the following attributes:",
question_items=["Product A", "Product B", "Product C"],
question_options=["Poor", "Fair", "Good", "Excellent"],
)
Allocating a fixed budget across multiple options.
from edsl import QuestionBudget
q = QuestionBudget(
question_name="time_allocation",
question_text="How would you allocate 100 hours across these activities?",
question_options=["Work", "Exercise", "Leisure", "Sleep"],
budget_sum=100
)
Response as key-value pairs (structured data).
from edsl import QuestionDict
q = QuestionDict(
question_name="contact_info",
question_text="Provide your contact information:",
answer_keys=["name", "email", "phone"]
)
Extracting specific information from text.
from edsl import QuestionExtract
q = QuestionExtract(
question_name="entities",
question_text="Extract all company names from the following text: {{ text }}",
answer_template={"companies": "List of company names"}
)
BM25-powered search through large option sets.
from edsl import QuestionDropdown
q = QuestionDropdown(
question_name="country",
question_text="Select your country:",
question_options=["Afghanistan", "Albania", ..., "Zimbabwe"]
)
Multiple choice with an "Other" option for custom responses.
from edsl import QuestionMultipleChoiceWithOther
q = QuestionMultipleChoiceWithOther(
question_name="source",
question_text="How did you hear about us?",
question_options=["Google", "Friend", "Advertisement"],
other_option_label="Other (please specify)"
)
Checkbox with an "Other" option for custom responses.
from edsl import QuestionCheckboxWithOther
q = QuestionCheckboxWithOther(
question_name="interests",
question_text="What are your interests?",
question_options=["Sports", "Music", "Reading"],
other_option_label="Other"
)
Select top K items from a list.
from edsl import QuestionTopK
q = QuestionTopK(
question_name="favorites",
question_text="Select your top 3 favorite items:",
question_options=["A", "B", "C", "D", "E"],
k=3
)
Python function-based question (not sent to LLM - computed locally).
from edsl import QuestionFunctional
def compute_sum(scenario, agent):
numbers = scenario.get("numbers", [])
return sum(numbers)
q = QuestionFunctional(
question_name="total",
question_text="Calculate the sum",
func=compute_sum
)
Use custom Pydantic models as response schemas.
from edsl import QuestionPydantic
from pydantic import BaseModel
class PersonInfo(BaseModel):
name: str
age: int
occupation: str
q = QuestionPydantic(
question_name="person",
question_text="Describe a person:",
pydantic_model=PersonInfo
)
Responses with markdown formatting.
from edsl import QuestionMarkdown
q = QuestionMarkdown(
question_name="formatted_response",
question_text="Write a formatted response with headers and lists."
)
All questions support these common parameters:
| Parameter | Description |
|---|---|
question_name | Unique identifier (valid Python identifier) |
question_text | The question text (supports Jinja2 templating) |
answering_instructions | Optional custom instructions for the LLM |
question_presentation | Optional custom presentation template |
| Type | Use Case | Key Parameter |
|---|---|---|
QuestionFreeText | Open-ended responses | - |
QuestionMultipleChoice | Single selection | question_options |
QuestionCheckBox | Multiple selections | question_options |
QuestionNumerical | Numbers | min_value, max_value |
QuestionYesNo | Binary yes/no | - |
QuestionLinearScale | Numeric scale | question_options, option_labels |
QuestionLikertFive | 5-point agree/disagree | - |
QuestionList | List of items | max_list_items |
QuestionRank | Ordering | question_options, num_selections |
QuestionMatrix | Grid/table | question_items, question_options |
QuestionBudget | Budget allocation | question_options, budget_sum |
QuestionDict | Key-value pairs | answer_keys |
QuestionExtract | Extract from text | answer_template |
QuestionDropdown | Large option sets | question_options |
EDSL uses Jinja2 templating to create dynamic questions. Template variables are enclosed in {{ }} and are rendered at runtime.
from edsl import QuestionFreeText, Scenario
q = QuestionFreeText(
question_name="opinion",
question_text="What do you think about {{ scenario.fruit }}?"
)
scenarios = [
Scenario({"fruit": "apples"}),
Scenario({"fruit": "oranges"}),
]
results = q.by(scenarios).run()
from edsl import QuestionFreeText, Agent
q = QuestionFreeText(
question_name="perspective",
question_text="As a {{ agent.occupation }}, what do you think about remote work?"
)
agent = Agent(traits={"occupation": "software engineer", "age": 35})
results = q.by(agent).run()
Reference previous answers within a survey:
q1 = QuestionFreeText(
question_name="name",
question_text="What is your name?"
)
q2 = QuestionFreeText(
question_name="greeting",
question_text="Hello {{ name.answer }}! How are you today?"
)
survey = Survey([q1, q2])
q = QuestionMultipleChoice(
question_name="preference",
question_text="Which {{ scenario.category }} do you prefer?",
question_options=[
"{{ scenario.option_a }}",
"{{ scenario.option_b }}",
"{{ scenario.option_c }}"
]
)
| Variable | Access Pattern | Example |
|---|---|---|
| Scenario value | {{ scenario.key }} | {{ scenario.fruit }} |
| Agent trait | {{ agent.trait }} | {{ agent.occupation }} |
| Prior answer | {{ question_name.answer }} | {{ q1.answer }} |
Rules define conditional navigation through surveys using Jinja2 template expressions.
survey = survey.add_skip_rule(
"pet_name",
"{{ has_pet.answer }} == 'No'"
)
survey = (survey
.add_rule("preference", "{{ preference.answer }} == 'A'", "section_a")
.add_rule("preference", "{{ preference.answer }} == 'B'", "section_b"))
survey = survey.add_stop_rule(
"eligibility",
"{{ eligibility.answer }} == 'No'"
)
from edsl.surveys.navigation_markers import EndOfSurvey
survey = survey.add_rule("screening", "{{ screening.answer }} == 'disqualified'", EndOfSurvey)
"{{ question_name.answer }} == 'value'"
"{{ age.answer }} > 18"
"{{ q1.answer }} == 'yes' and {{ q2.answer }} != 'no'"
| Task | Method |
|---|---|
| Add skip rule | survey.add_skip_rule(question, expression) |
| Add navigation rule | survey.add_rule(question, expression, next_question) |
| Add stop rule | survey.add_stop_rule(question, expression) |
| Set priority | survey.add_rule(q, expr, next_q, priority=1) |
| View rules | survey.show_rules() |
| Jump to end | survey.add_rule(q, expr, EndOfSurvey) |
Memory controls which previous question-answer pairs an agent sees when answering each question.
| Mode | Method | Use Case |
|---|---|---|
| No Memory | (default) | Independent questions |
| Full Memory | survey.set_full_memory_mode() | Conversational surveys |
| Lagged Memory | survey.set_lagged_memory(lags=N) | Recent context only |
| Targeted Memory | survey.add_targeted_memory(focal, prior) | Precise control |
| Memory Collection | survey.add_memory_collection(focal, [priors]) | Multiple targeted |
from edsl.surveys.survey_helpers.matrix_combiner import combine_multiple_choice_to_matrix
new_survey = combine_multiple_choice_to_matrix(
survey=survey,
question_names=["trust_freelancer", "trust_ai", "trust_agency"],
matrix_question_name="trust_matrix"
)
Survey methods can be chained since each returns a new Survey:
survey = (Survey([q1, q2, q3])
.add_rule("q1", "{{ q1.answer }} == 'skip'", "q3")
.add_skip_rule("q2", "{{ q1.answer }} == 'skip'")
.add_question(q4))
| Task | Method |
|---|---|
| Create survey | Survey([q1, q2, q3]) |
| Add question | survey.add_question(q, index=None) |
| Add instruction | survey.add_instruction(inst) |
| Get question | survey.get("question_name") |
| List questions | survey.questions |
| Question names | survey.question_names |
| Number of questions | len(survey) |