> ## Documentation Index > Fetch the complete documentation index at: https://mintlify.com/Metaculus/metaculus/llms.txt > Use this file to discover all available pages before exploring further. # Question Types > Detailed explanation of all question types on Metaculus Metaculus supports five distinct question types, each designed for different kinds of predictions. Understanding these types will help you make better forecasts and interpret results correctly. ## Binary Questions Binary questions ask for the probability of a yes/no outcome. ### Structure * **Question format**: "Will \[event] happen by \[date]?" * **Input**: A single probability value between 0% and 100% * **Output**: The probability that the event occurs ### Examples * "Will SpaceX successfully land humans on Mars before 2030?" * "Will global temperatures rise by more than 1.5°C by 2025?" * "Will the S\&P 500 close above 5000 by end of 2024?" ### How to Forecast Your forecast represents your belief about the probability of the event occurring: * **50%**: You think it's equally likely to happen or not happen * **90%**: You're quite confident it will happen * **10%**: You think it's unlikely to happen * **100%**: You're certain it will happen (rarely appropriate) * **0%**: You're certain it won't happen (rarely appropriate) Avoid extreme probabilities (0% or 100%) unless you're absolutely certain. Even very likely events deserve probabilities like 95-99%, not 100%. ### Resolution Binary questions resolve to either: * **Yes** (1.0): The event occurred according to the resolution criteria * **No** (0.0): The event did not occur * **Ambiguous**: The resolution criteria cannot be determined * **Annulled**: The question was invalidated ### Scoring Binary questions are typically scored using logarithmic scoring rules that reward: * Accuracy (being close to the actual outcome) * Calibration (90% predictions should happen about 90% of the time) * Resolution (avoiding 50% predictions when you have information) ## Numeric Questions Numeric questions ask for predictions about continuous numerical values. ### Structure * **Question format**: "What will \[value] be on \[date]?" * **Input**: A probability distribution over a range of values * **Range**: Defined by `range_min` and `range_max` * **Scale**: Linear or logarithmic (for values spanning orders of magnitude) * **Bounds**: Can be open or closed ### Examples * "What will be the global average temperature anomaly in 2025?" * "How many electric vehicles will be sold in the US in 2024?" * "What will Bitcoin's price be on December 31, 2024?" ### Key Concepts **Scale Types:** Values are evenly spaced. Best for: * Temperature changes * Percentage points * Small ranges Example: 0°C to 5°C in 0.1°C increments Values are spaced exponentially. Best for: * Stock prices * Population figures * Values spanning multiple orders of magnitude Example: $1 to $1,000,000 where each doubling is equally likely **Bounds:** * **Closed bounds**: The resolution must fall within the defined range * **Open lower bound**: Resolution can be below `range_min` * **Open upper bound**: Resolution can be above `range_max` ### How to Forecast You define your prediction by specifying percentiles: * **25th percentile**: You think there's a 25% chance the value will be below this * **50th percentile (median)**: Your central estimate * **75th percentile**: You think there's a 75% chance the value will be below this Metaculus generates a smooth probability distribution (CDF) from your percentiles. The system uses 200 discrete points (by default) to represent your continuous distribution internally. ### Resolution Numeric questions resolve to the actual observed value according to the resolution criteria. If the value falls outside the defined range and bounds are open, it resolves to the out-of-bounds value. ## Date Questions Date questions ask when an event will occur. ### Structure * **Question format**: "When will \[event] happen?" * **Input**: A probability distribution over a date range * **Range**: Defined by `range_min` and `range_max` (as Unix timestamps) * **Output**: A CDF representing probability of occurrence over time ### Examples * "When will the first human set foot on Mars?" * "When will the US unemployment rate drop below 4%?" * "When will the next major earthquake (>7.0 magnitude) occur in California?" ### How to Forecast Similar to numeric questions, you specify percentiles: * **25th percentile date**: 25% chance the event occurs before this date * **50th percentile date**: Your median estimate * **75th percentile date**: 75% chance the event occurs before this date For date questions, consider both the earliest possible date (due to constraints like technology, politics, or physics) and the likelihood of delays. ### Resolution Resolves to the specific date when the event occurred according to resolution criteria. Time is typically recorded to the day or specific timestamp depending on the question. ## Multiple Choice Questions Multiple choice questions ask you to distribute probability across several discrete options. ### Structure * **Question format**: "Which outcome will occur?" * **Input**: Probability for each option (must sum to 100%) * **Options**: 2 or more discrete choices defined in the `options` field * **Options history**: Changes to options over time are tracked ### Examples * "Who will win the 2024 Presidential election?" (Options: Candidate A, Candidate B, Candidate C, Other) * "What will be the primary cause of the next major recession?" (Options: Inflation, Financial crisis, Geopolitical event, Other) * "Which company will reach a \$5 trillion market cap first?" (Options: Apple, Microsoft, Amazon, Google, Other, None by 2030) ### How to Forecast Assign a probability to each option: 1. Review all options carefully 2. Assign probabilities based on your assessment 3. Ensure total equals 100% 4. Submit your forecast If the options change after you've made a forecast (new options added or removed), your forecast is automatically adjusted to account for the changes. ### Option Ordering Questions can have different ordering modes: * **DEFAULT**: Creation order in forecast maker, CP order in views * **CP\_DESC**: All views sort by descending community prediction ### Resolution Resolves to the index of the correct option. The resolution is stored as a single selected option. ### Special Cases * **"None of the above" or "Other"**: Often included as catch-all options * **Dynamic options**: Some questions may add options over time * **Options history**: Changes are tracked with timestamps ## Discrete Questions Discrete questions are hybrid questions with discrete probability bins over a continuous-like range. ### Structure * **Question format**: "What will \[value] be?" with discrete buckets * **Input**: Probability distribution across discrete bins * **Range**: Defined by `range_min` and `range_max` * **Bins**: Number of discrete outcomes (`inbound_outcome_count`, default: 200) * **Bounds**: Can be open or closed ### Examples * "How many seats will a political party win?" (Bins: 0-10, 11-20, 21-30, etc.) * "What will the final vote percentage be?" (Bins: 0-5%, 5-10%, 10-15%, etc.) ### How to Forecast Similar to numeric questions but with discrete bins: 1. View the discrete bins 2. Assign probability to each bin (or use percentile mode) 3. Ensure probabilities sum to 100% 4. Submit ### Differences from Numeric Questions | Feature | Numeric | Discrete | | - | - | - | | **Distribution** | Smooth continuous CDF | Discrete probability mass | | **Zero point** | Can have logarithmic zero | No zero point | | **Bins** | Continuous | Discrete buckets | | **Precision** | High (200+ points) | Medium (based on bin count) | ## Question Metadata All questions share common metadata: ### Time Fields * **Open time**: When forecasting begins * **Scheduled close time**: When forecasting is scheduled to end * **Scheduled resolve time**: Expected resolution date * **Actual close time**: When forecasting actually ended * **Actual resolve time**: When the outcome became known * **CP reveal time**: When the community prediction is shown (can be delayed) ### Scoring Fields * **Default score type**: Usually "peer" or "spot\_peer" * **Question weight**: Importance multiplier (default: 1.0) * **Include bots in aggregates**: Whether bot forecasts are included ### Resolution Criteria The specific, objective criteria used to determine the outcome. This should be clear, verifiable, and unambiguous. Additional details, edge cases, and clarifications about the resolution criteria. Background information and context to help forecasters understand the question. ## Question Status Questions progress through several statuses: Question created but not yet open for forecasting (`open_time` in the future). Currently accepting forecasts (between `open_time` and `scheduled_close_time`). No longer accepting forecasts but not yet resolved (after `scheduled_close_time`). Outcome determined and resolution set. ## Choosing the Right Question Type When creating or forecasting on questions: * Yes/no outcomes * Event occurrence * Simple true/false statements * Continuous measurements * Values with high precision * Scientific/economic data * Timing predictions * "When will X happen?" * Temporal milestones * Several distinct outcomes * Categorical predictions * "Which X will Y?" ## Next Steps Learn how to submit forecasts for each type Understand how community predictions work See how your forecasts are scored Integrate question data into your applications