> ## 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.

# Track Record

> How to view and interpret your forecasting performance on Metaculus

Your track record shows how accurate your predictions have been over time. Understanding these metrics helps you improve your forecasting skills and demonstrates your expertise to the community.

## Accessing Your Track Record

You can view track records for yourself and other users:

<Steps>
  <Step title="Navigate to Profile">
    Click on any username or go to your own profile at `/accounts/profile/[id]`
  </Step>

  <Step title="Select Track Record Tab">
    Click the "Track Record" tab to view forecasting performance metrics
  </Step>

  <Step title="Explore the Data">
    Review charts, statistics, and historical performance
  </Step>
</Steps>

## Key Performance Metrics

Your track record includes several important statistics:

### Average Peer Score

**What it is**: Your mean peer score across all scored questions

**How to interpret**:

* **Positive scores**: You're beating the community prediction
* **Negative scores**: The community prediction is beating you
* **Score near 0**: You're performing similarly to the CP
* **Higher is better**: More positive scores indicate better performance

<Tip>
  An average peer score above 0 means you're consistently adding value beyond the community aggregate - a sign of strong forecasting skill.
</Tip>

### Total Predictions

**What it is**: The total number of forecasts you've submitted

**Why it matters**:

* More predictions = more data points for evaluating your skill
* Quantity alone doesn't indicate quality
* Shows your engagement level

<Note>
  Each update to a forecast counts as a new prediction. Active forecasters who update regularly will have higher counts.
</Note>

### Questions Predicted

**What it is**: The number of unique questions you've forecasted on

**Why it matters**:

* Shows breadth of forecasting activity
* Higher diversity generally improves calibration
* Indicates your coverage across topics

### Questions Scored

**What it is**: The number of resolved questions where you received a score

**Why it matters**:

* Only scored questions contribute to your track record statistics
* This is the most important number for evaluating performance
* More scored questions = more reliable performance estimates

<Warning>
  A forecast only receives a score if:

  1. The question resolves (not annulled or ambiguous)
  2. Your forecast was active during the scoring period
  3. You forecasted within the valid time window (between open\_time and close\_time)
</Warning>

## Track Record Visualizations

The track record page includes several charts to help you understand your performance:

### Score Scatter Plot

**What it shows**: Individual scores for each question plotted over time

**How to read it**:

* **X-axis**: Time when the question was resolved
* **Y-axis**: Your score on that question
* **Points above 0**: Questions where you beat the CP
* **Points below 0**: Questions where the CP beat you
* **Patterns**: Look for trends over time

<Accordion title="What to look for">
  - **Improving trend**: Points moving upward over time indicates you're getting better
  - **Consistent performance**: Points clustered around a specific value
  - **Outliers**: Unusually high or low scores worth investigating
  - **Volatility**: How spread out your scores are
</Accordion>

### Calibration Curve

**What it shows**: How well your predicted probabilities match actual outcomes

**How to read it**:

* **X-axis**: Your predicted probability
* **Y-axis**: Actual frequency of occurrence
* **Diagonal line**: Perfect calibration
* **Your curve**: Your actual calibration

<Tabs>
  <Tab title="Well-calibrated">
    Your curve closely follows the diagonal line:

    * When you say 70%, it happens about 70% of the time
    * When you say 30%, it happens about 30% of the time
    * This is the goal!
  </Tab>

  <Tab title="Overconfident">
    Your curve is below the diagonal:

    * When you say 90%, it only happens 70% of the time
    * You're too confident in your predictions
    * Consider expressing more uncertainty
  </Tab>

  <Tab title="Underconfident">
    Your curve is above the diagonal:

    * When you say 60%, it actually happens 80% of the time
    * You're not confident enough
    * You can be more decisive
  </Tab>
</Tabs>

<Note>
  Calibration curves are most reliable when you have many resolved binary predictions. With fewer predictions, random variation can make the curve appear miscalibrated.
</Note>

### Score Histogram

**What it shows**: The distribution of your scores across all questions

**How to read it**:

* **X-axis**: Score bins
* **Y-axis**: Number of questions in each bin
* **Shape**: Shows the pattern of your performance

**Interpretations**:

* **Right-skewed**: More high scores than low (good!)
* **Left-skewed**: More low scores than high (room for improvement)
* **Normal distribution**: Mix of good and bad predictions
* **Bimodal**: Two distinct types of performance (might indicate domain expertise)

## Score Types

Metaculus uses several scoring methods. The main ones you'll see:

### Peer Score

**How it works**: Compares your forecast to the community prediction (CP) at each point in time

**Calculation**:

* Both you and the CP are scored against the actual outcome using a proper scoring rule (usually Brier or log score)
* Your peer score = Your score - CP score
* Positive means you beat the CP; negative means the CP beat you

**Why it's used**:

* Measures your added value over the crowd
* Accounts for question difficulty
* Fair comparison across different questions

<Tip>
  Peer score is the default and most common score type on Metaculus. Focus on this for evaluating your performance.
</Tip>

### Baseline Score

**How it works**: Compares your forecast to a baseline forecast (often 50% for binary, or a naïve prediction)

**Why it's used**:

* Shows absolute performance, not relative
* Useful for understanding raw accuracy
* Less common than peer score

### Spot Scores

**How it works**: Evaluates your forecast at a specific point in time (the `spot_scoring_time` or `cp_reveal_time`)

**Types**:

* **Spot Peer**: Your score vs. CP at the spot time
* **Spot Baseline**: Your score vs. baseline at the spot time

**Why it's used**:

* Reduces incentive to update constantly
* Evaluates skill at a specific meaningful moment
* Common in tournaments and challenges

<Note>
  If `spot_scoring_time` isn't set, it defaults to `cp_reveal_time`, then `actual_close_time`, then `scheduled_close_time`.
</Note>

### Coverage

Along with scores, you'll see coverage metrics:

**What it is**: The fraction of the scoring period during which you had an active forecast

**Values**:

* **1.0 (100%)**: You had a forecast active for the entire period
* **0.5 (50%)**: You forecasted for half the period
* **0.0 (0%)**: No forecast during the scoring period

**Why it matters**:

* Some scoring methods weight by coverage
* Higher coverage = more credit for your forecasts
* Encourages early and sustained participation

## Additional Statistics

Beyond forecasting performance, track records show:

### Authoring Stats

<CardGroup cols={2}>
  <Card title="Questions Authored" icon="pen">
    Number of questions you've created that were approved
  </Card>

  <Card title="Forecasts on Authored Questions" icon="users">
    How many predictions your questions have received
  </Card>

  <Card title="Notebooks Authored" icon="book">
    Articles and analyses you've published
  </Card>

  <Card title="Comments Authored" icon="comments">
    Total comments you've written
  </Card>
</CardGroup>

## Improving Your Track Record

Here's how to improve your forecasting performance:

<AccordionGroup>
  <Accordion title="1. Improve Calibration">
    **If overconfident**:

    * Express more uncertainty in your forecasts
    * Use wider probability distributions
    * Avoid extreme probabilities (0%, 100%)
    * Consider alternative scenarios

    **If underconfident**:

    * Trust your analysis more
    * Be more decisive when you have good information
    * Don't automatically regress to 50%
  </Accordion>

  <Accordion title="2. Update Forecasts">
    * Set calendar reminders to review questions
    * Update when significant new information emerges
    * Don't be afraid to change your mind
    * More updates generally improve scores (especially with recency weighting)
  </Accordion>

  <Accordion title="3. Focus on Your Strengths">
    * Identify topics where you consistently score well
    * Develop domain expertise in specific areas
    * Don't spread yourself too thin
    * Quality over quantity
  </Accordion>

  <Accordion title="4. Learn from Others">
    * Read comments from top forecasters
    * Check the community prediction and ask why it differs from yours
    * Study track records of skilled forecasters
    * Ask for feedback on your reasoning
  </Accordion>

  <Accordion title="5. Use Systematic Methods">
    * Develop a consistent forecasting process
    * Use reference classes and base rates
    * Break down complex questions
    * Document your reasoning
  </Accordion>

  <Accordion title="6. Avoid Common Pitfalls">
    * Don't anchor on initial impressions
    * Beware of groupthink
    * Watch for motivated reasoning
    * Don't overweight vivid scenarios
    * Remember regression to the mean
  </Accordion>
</AccordionGroup>

## Track Record Privacy

You can control the visibility of your track record:

1. Go to **Account Settings**
2. Navigate to **Privacy Settings**
3. Choose who can see your:
   * Full track record
   * Individual predictions
   * Statistics

<Warning>
  Public track records help build credibility and trust in the forecasting community. Consider keeping yours public unless you have specific privacy concerns.
</Warning>

## Tournament and Project Leaderboards

Beyond personal track records, you can appear on leaderboards:

### Types of Leaderboards

* **Tournament Leaderboards**: Specific to tournament projects
* **Global Leaderboards**: Site-wide performance over time periods
* **Project Leaderboards**: Performance within specific projects

### Leaderboard Metrics

<Tabs>
  <Tab title="Score">
    Total score across all questions in the leaderboard's scope
  </Tab>

  <Tab title="Rank">
    Your position relative to other forecasters
  </Tab>

  <Tab title="Coverage">
    Fraction of questions you participated in
  </Tab>

  <Tab title="Contribution Count">
    Number of questions where you had an active forecast
  </Tab>

  <Tab title="Medal">
    Gold/Silver/Bronze designation for top performers
  </Tab>

  <Tab title="Prize">
    Monetary prize amount (if applicable)
  </Tab>
</Tabs>

### Medal Exclusions

Some users are excluded from medals and prizes:

* **Staff**: Metaculus employees
* **Project owners**: Organizers of the tournament
* **Disqualified**: Users who violated rules
* **Other**: Custom exclusions

<Note>
  Excluded users may still appear on leaderboards with `show_anyway=True`, but they don't receive ranks or prizes.
</Note>

## Historical Context

Your track record evolves over time:

* **Early predictions**: May have higher variance and less reliability
* **Learning curve**: Most forecasters improve significantly in their first 6-12 months
* **Plateau**: Performance often stabilizes after gaining experience
* **Specialization**: Many successful forecasters focus on specific domains

<Tip>
  Don't be discouraged by early poor performance. Forecasting is a skill that improves with practice, feedback, and learning.
</Tip>

## Comparing Track Records

When comparing forecasters:

<AccordionGroup>
  <Accordion title="Look at sample size">
    More scored questions = more reliable estimate of skill
    Compare forecasters with similar numbers of predictions
  </Accordion>

  <Accordion title="Consider recency">
    Recent performance may be more indicative than all-time
    People improve (or decline) over time
  </Accordion>

  <Accordion title="Check specialization">
    Some forecasters excel in specific domains
    Overall scores may not reflect domain expertise
  </Accordion>

  <Accordion title="Account for difficulty">
    Peer scores already account for question difficulty
    But topic selection still matters
  </Accordion>
</AccordionGroup>

## API Access

You can retrieve track record data programmatically:

```python theme={null}
# Get user profile with track record
profile = ServerProfileApi.getProfileById(user_id)

# Access statistics
average_score = profile.average_score
forecasts_count = profile.forecasts_count
questions_predicted = profile.questions_predicted_count
score_count = profile.score_count

# Charts
scatter_plot = profile.score_scatter_plot
calibration = profile.calibration_curve
histogram = profile.score_histogram
```

See the API reference for complete documentation.

## Next Steps

<CardGroup cols={2}>
  <Card title="Making Predictions" icon="chart-line" href="/guides/making-predictions">
    Learn how to submit better forecasts
  </Card>

  <Card title="Question Types" icon="list-check" href="/guides/question-types">
    Understand what you're predicting
  </Card>

  <Card title="Aggregation Methods" icon="users" href="/guides/aggregation-methods">
    Learn how the CP is calculated
  </Card>

  <Card title="Scoring API" icon="code" href="/api/scores">
    Access score data programmatically
  </Card>
</CardGroup>