Balance Analytics

Advanced · 10 min read

Balance Analytics is Chitmunk's data-driven dashboard for analyzing the content and balance of your card game. It reads directly from your project's CSV data and surfaces patterns, outliers, and potential balance issues that are hard to spot by reading cards one at a time.

Accessing the Analytics Dashboard

Open Balance Analytics from the Game Home dashboard sidebar. Navigate to the Balance Lab view under the Design Catalog section. The dashboard analyzes your current project's CSV data automatically.

The dashboard has five main sections, accessible via the tab bar at the top:

1. Overview
2. Keyword Frequency
3. Duplicates & Near-Duplicates
4. Balance Heatmaps
5. Insights

Overview

The Overview tab gives you a quick summary of your dataset:

• Total cards: Number of rows in your CSV (across all card types in the project).
• Columns detected: All CSV columns with their data types (text, numeric, boolean).
• Numeric column summaries: For each numeric column, shows min, max, mean, median, and standard deviation. An unusually high standard deviation can indicate a wide power spread.
• Empty cell rate: Percentage of blank cells per column. High empty rates in key columns may indicate incomplete data entry.
• Card type breakdown: If your project has multiple card types, a pie chart shows how many cards belong to each type.

Keyword Frequency Analysis

The Keyword Frequency tab tokenizes all text content in your CSV (card names, descriptions, effect text) and counts how often each term appears.

What It Shows

• Word cloud: Most frequent terms displayed at proportional sizes. Dominant mechanics and themes appear largest.
• Frequency table: Sortable table with each unique word, its count, and the percentage of cards containing it.
• n-gram analysis: Common 2- and 3-word phrases (e.g., "draw two cards", "end of turn", "deal 1 damage"). These reveal repeated mechanic language.

Using Keyword Data

Keyword frequency tells you which mechanics dominate your design. If "draw" appears on 60% of cards, your game may over-rely on card draw as a resource. If a keyword like "discard" only appears twice in a 60-card deck, it may not have enough support to function as a viable strategy.

Use the frequency table to identify your game's primary keyword vocabulary — the 5–10 mechanics that define the core gameplay loop — and ensure they appear at intentional frequencies.

Duplicate and Near-Duplicate Detection

This tab identifies cards that are identical or nearly identical in their text content.

• Exact duplicates: Cards with identical values in all text columns. These are almost always data entry errors (copied rows that weren't updated).
• Near-duplicates: Cards with very similar text (above a configurable similarity threshold, default 85%). These may be intentional variants (e.g., "Deal 1 damage" vs "Deal 2 damage") or accidental near-copies.

Similarity Threshold

Adjust the similarity slider (50%–100%) to control sensitivity. At 95%, only cards with nearly identical text are flagged. At 70%, broader similarities are surfaced — useful for identifying cards that are mechanically similar even if worded differently.

Reviewing Flagged Cards

Each flagged pair shows both cards side by side with differences highlighted in red. Click either card to jump to that row in the Data Mode spreadsheet editor to fix it.

Balance Heatmaps

Heatmaps visualize how numeric values are distributed across your card set. Select any two numeric columns from your CSV to generate a scatter plot or heatmap showing how they correlate.

Common Balance Charts

• Cost vs. Power: Plot card cost (mana, action points) on the X axis and some power metric (damage, life gain, card draw) on the Y axis. A balanced game has points clustering near a diagonal line — high-cost cards should have proportionally higher power.
• Category distribution: If you have a "Type" or "Faction" column, plot power metrics broken down by category to see if any faction has a systematic advantage.
• Value histogram: A distribution chart for a single column showing how many cards have each value. A healthy distribution is usually a bell curve; a bimodal distribution (lots of very low and very high values, nothing in between) often indicates balance issues.

Color Coding

Points on the scatter plot are color-coded by a third categorical column (e.g., card type, rarity, or faction) if you select one from the "Color By" dropdown. This makes it easy to see if one category systematically over- or under-performs.

Interpreting Insights

The Insights tab runs a set of automated checks and surfaces findings in plain language:

• Outlier cards: Cards with values more than 2 standard deviations from the mean in any numeric column. These may be intentionally powerful (rares, boss cards) or accidental errors.
• Underrepresented keywords: Keywords that appear fewer than 3 times in the entire set — too few to form a consistent strategy around.
• Category imbalance: If your cards have a Type/Faction column, flags any category with fewer than 15% of total cards as potentially underrepresented.
• Numeric column anomalies: Columns where the maximum value is more than 5x the mean (sharp outliers), or where many cells are empty (possible missing data).
• Power curve assessment: If a cost and power column are detected, grades the correlation as Strong, Moderate, or Weak. Weak correlation suggests cost and power are not well-calibrated.

Drill-Down Navigation

Throughout the analytics dashboard, clicking on a card name, keyword, or data point opens a drill-down view:

• Click a card name anywhere in the dashboard to jump to that card in the Data Mode spreadsheet editor, with the card's row highlighted.
• Click a keyword in the frequency table to see all cards containing that keyword, displayed as a filtered card grid.
• Click a heatmap point to see the card at that data point's position, with all its CSV values.
• Click an insight to expand it and see the list of affected cards with links to each.

Drill-down navigation makes it easy to move from a high-level pattern to the specific cards involved, investigate, and jump directly to editing — without losing your place in the dashboard.

Tips for Using Analytics Effectively

• Run analytics at milestones: Analytics is most valuable after completing a full draft (50+ cards), not during early prototyping. Run it at version checkpoints to track how the design evolves.
• Define your intended power curve first: Before interpreting heatmaps, write down your intended cost-to-power ratio. Is a cost-2 card supposed to deal 3 damage? Having that baseline makes the scatter plot meaningful.
• Use near-duplicate detection before playtesting: Catching accidental duplicates before a playtest saves time. Run this check every time you add a batch of new cards.
• Track keyword frequency across versions: Export the keyword frequency table as CSV (copy-paste from the table) and compare it to earlier versions. Seeing how your vocabulary shifts over iterations reveals whether you are refining or drifting from your original design intent.
• Analytics works on all card types: The dashboard analyzes all card types in your project together. Use the "Filter by Card Type" dropdown to analyze each deck type separately if needed.