Anti-Pattern Catalog (Locked)
Every Chart Swap Advisor run evaluates the current chart against every entry in this catalog. Auto-block severity cannot be overridden by user pressure per HC#11.
Severity Tiers
- Auto-Block. Recommendation stays Block regardless of user pressure. Chart is unacceptable for stakeholder-facing dashboards. Rationale must cite this catalog entry.
- Warning. Chart is suboptimal; swap recommended. User can Keep in Desktop if they have context that overrides the general rule.
- Info. Chart is defensible but worth reviewing.
Auto-Block Anti-Patterns
AB-1: 3D Chart (Any Type)
Detection signal. Mark type or shelf-placement includes 3D encoding (Tableau's 3D chart types include 3D pie, 3D bar, 3D area). Rarely present in Tableau Studio natively; more common in ported Excel or PowerPoint imports.
Why it's blocked. 3D perspective distorts data reading. Every 3D chart has a superior 2D alternative. Confirmed portfolio-wide anti-pattern.
Recommendation. Swap to the 2D equivalent of the current type (3D pie → treemap or stacked bar; 3D bar → 2D bar; 3D area → 2D area or line).
AB-2: Pie Chart with >5 Slices
Detection signal. Mark type is Pie AND the number of dimension values (slices) exceeds 5.
Why it's blocked. Pie charts become unreadable beyond 5 slices. Angle comparison degrades sharply. Stacked bar or treemap serves the same Composition question with better readability.
Recommendation. Swap to stacked bar (when slice order matters) or treemap (when hierarchy is available).
AB-3: Word Cloud in Stakeholder-Facing Dashboard
Detection signal. Mark type is Text with text sizing driven by measure AND worksheet is placed on a dashboard.
Why it's blocked. Word clouds are ambiguous, hard to compare, and rarely answer a decision-oriented question. Fine for exploratory work; unacceptable for stakeholder decision surfaces.
Recommendation. Swap to sorted bar (Ranking) or treemap (Composition).
AB-4: Packed Bubbles in Stakeholder-Facing Dashboard
Detection signal. Mark type is Circle with size encoding driven by measure AND worksheet is placed on a dashboard AND no positional axis carries meaning.
Why it's blocked. Packed bubbles obscure exact comparison and don't scale beyond a small number of items. Fine for aesthetic exploration; unacceptable for stakeholder decision surfaces.
Recommendation. Swap to sorted bar (Ranking) or treemap (Composition, when hierarchy exists).
AB-5: Radar Chart
Detection signal. Radar-shaped chart authored via calc-heavy shelf placement (Tableau has no native radar mark; radar charts appear as multi-measure line charts with polar-axis calcs).
Why it's blocked. Radar charts require the reader to compare enclosed areas, which is a poor visual task. Every use case has a superior alternative (small-multiples bar for multi-metric comparison, sorted bar for ranking).
Recommendation. Swap to small-multiples bar (each metric its own bar chart) or a scoring KPI tile with delta.
AB-6: Gauge Chart
Detection signal. Gauge-shaped chart authored via calc-heavy Automatic mark placement (Tableau has no native gauge mark).
Why it's blocked. Gauges are space-inefficient and hard to read at a glance. KPI tiles or bullet charts serve the same Status/KPI question far better.
Recommendation. Swap to KPI tile (BAN + delta) or bullet chart.
Warning Anti-Patterns
W-1: Dual-Axis When Single-Axis Suffices
Detection signal. Two measures on Rows or Columns with dual-axis synchronized AND the two measures share the same unit and scale.
Why it's a warning. Dual-axis is powerful when two measures have different units or scales (e.g., Sales in dollars + Profit Ratio in percent). When both measures are in the same unit and scale (e.g., Sales + Profit both in dollars), single-axis with a color-encoded measure name is clearer.
Recommendation. Swap to single-axis with the two measures color-encoded. Preserve dual-axis if the user cites a specific decision that requires two independent scales.
Keep condition. Two measures with different units or different orders of magnitude (>10x scale difference) → Keep dual-axis; recommendation is Info-tier only.
W-2: Executive Audience with Analytical Chart
Detection signal. User states audience is Executive AND current chart is Scatter, Distribution, Correlation, or a text-heavy exploratory view.
Why it's a warning. Executive audience has a 5-30 second time budget per Viz Planner's audience taxonomy. Analytical charts require analytical time. A mismatch produces stakeholder frustration.
Recommendation. Swap to KPI tile with the single answer, or a sorted bar showing the top-5 outliers. Preserve the scatter as a downstream drill-in worksheet if the executive dashboard has a "see detail" link.
W-3: Text Table When a Chart Would Answer Faster
Detection signal. Mark type is Text AND question type is Temporal, Ranking, or Correlation.
Why it's a warning. Text tables serve reference lookup and detailed number verification. Temporal/Ranking/Correlation questions read faster with visual encoding.
Recommendation. Swap to the question-type default chart (Line for Temporal, sorted bar for Ranking, scatter for Correlation). Preserve the text table as a companion detail view if the user cites a compliance or audit requirement.
W-4: Categorical Chart with Discrete Date on Columns
Detection signal. Mark type is Bar AND Columns contains a discrete date pill.
Why it's a warning. Discrete date pills on bars produce categorical-looking output for temporal data. Line chart with continuous date reads faster. Bar chart with discrete date is defensible only when period-over-period comparison is the primary question.
Recommendation. Swap the discrete date pill to continuous, or swap the mark to Line. Preserve if user cites period-over-period comparison as the primary intent.
W-5: Choropleth with Count-Based Measure
Detection signal. Filled map (choropleth) AND the measure is a raw count (Number of Records, COUNT(*), COUNTD(ID)).
Why it's a warning. Choropleths with count-based measures over-emphasize high-population regions and mislead about density or rate. Normalization is almost always required.
Recommendation. Swap to a rate-based measure (count / population, or count / area). Preserve the choropleth type; swap the underlying measure.
W-6: Multi-Measure KPI in a Single Worksheet
Detection signal. Question type is Status/KPI AND worksheet has 3+ measures on the Marks card.
Why it's a warning. Status/KPI is a single-decision question. Multiple measures on one worksheet dilute the answer. Route to multiple KPI worksheets in a dashboard grid, one per measure.
Recommendation. Split into separate worksheets, one per KPI. Route the assembly to viz-planner for dashboard-level KPI grid design.
W-7: Histogram Used When Outlier Emphasis Is the Real Question
Detection signal. Question type is Distribution AND user language or worksheet caption emphasizes outliers/extremes ("outlier," "spread by group," "compare distributions across categories") AND current mark is a plain bar-based histogram with no per-category breakdown.
Why it's a warning. A single histogram shows overall shape but not per-group outliers or spread comparisons. Box plot serves outlier-emphasis and per-category comparison better; density plot serves smooth continuous-distribution reading better.
Recommendation. Swap to box plot (outlier/per-category emphasis) or density plot (continuous, single-group distributions). Preserve histogram if the question is strictly "how many fall in this range" with no grouping or outlier intent.
Info Anti-Patterns
I-1: Chart Complexity vs. Audience Time Budget
Detection signal. Chart has 5+ measures OR 4+ dimensions AND no audience is stated.
Why it's Info. Complexity is a signal, not a fault. If the audience is Analyst-tier (per Viz Planner's audience taxonomy), complexity is fine. If Manager or Executive, complexity may need to be reduced.
Recommendation. Ask the user to state the audience, then rerun with HC#12 audience-honoring logic. If Analyst, mark as Keep with Info note. If Manager/Executive, upgrade to Warning.
I-2: Sort Order Not Specified for Ranking
Detection signal. Question type is Ranking AND no explicit sort spec is set on the measure.
Why it's Info. Tableau may default-sort ascending in ways that don't match the question. Ranking questions almost always want descending by measure.
Recommendation. Add an explicit sort spec on the measure in the correct direction.
Anti-Pattern Match Rules
- Multiple anti-patterns can fire on the same chart. All fire; verdict is the highest severity.
- Auto-Block trumps Warning trumps Info.
- If no anti-pattern fires and the classification recommends the current chart type, verdict is Keep.
- If no anti-pattern fires but the classification recommends a different chart type, verdict is Swap Recommended (Warning-tier by default; Info-tier if the current chart is a defensible alternative per the taxonomy's "Alternatives" section).
- If no anti-pattern fires and the classification is Low-confidence with no clean type match, verdict is Keep with
recommended_chart_type: nulland an explicit low-confidence non-endorsement rationale (seechart_recommendation_catalog.md's locked default). This is not the same as a confident Keep and must never be rendered identically to one.
Refuse to Fabricate
Never invent an anti-pattern to reach a Swap Recommended verdict. Every anti-pattern cited in a recommendation must map to a specific entry in this catalog. If a chart has a genuinely novel design problem not covered here, note it in the report as a Low-confidence Info-tier observation and flag for catalog revision, but do not invent an anti-pattern ID.