Evidence-Led Tables and Visuals

Visuals are optional. Add one only when it answers an analytical question more clearly than prose or a small table. Do not generate a fixed count, create decorative cover imagery by default, or call another skill or external image service.

Required data contract

Before creating a chart, define:

• claim IDs the visual supports;
• input source IDs and calculation ID;
• geography, population, and coverage;
• observed/estimated/forecast status;
• period and retrieval cutoff;
• currency, base year, and price basis;
• stock/flow/count/share/rate/price/index;
• units and denominator;
• taxonomy and version;
• revision status;
• suppression, missing values, and material limitations.

Build visuals from the validated local data, never from an image prompt that contains unsupported numbers.

Choose the smallest useful display

Avoid pie/donut charts for many categories, dual axes, 3-D effects, area scaling without explanation, radar charts for precise comparison, and unlabeled quadrant scores.

Historical and forecast data

• Separate observed and forecast periods with a clear boundary.
• Use different line styles as well as color.
• Label scenarios as conditional; do not label the outer paths as a confidence interval.
• Show revisions or vintage when they materially change the historical path.
• Do not splice incompatible series without a break marker and explanation.
• If a historical source ends before the forecast base year, identify the bridge estimate.

TAM/SAM/SOM

Prefer a table or nested bar chart to concentric circles because area can imply precision and proportionality poorly.

Show:

• top-down and bottom-up TAM separately;
• SAM filters and retained share;
• SOM capture share and horizon;
• downside/base/upside values;
• reconciliation gap;
• denominator and coverage keys;
• sensitivity to principal assumptions.

The caption must say that outputs are conditional scenarios.

Competitive evidence

Feature matrix

Use statuses yes, partial, no, unknown, and not-applicable. Each cell must map to source IDs and a common product scope, geography, and as-of date. Do not treat unknown as no.

Market shares

Use bars rather than a crowded pie. State:

• relevant product/geographic scope;
• share metric and denominator;
• period;
• unknown/residual share;
• whether values are reported, estimated, or calculated;
• source IDs for numerator and total.

If scope uncertainty is material, show alternate definitions instead of one authoritative-looking chart.

Positioning maps

Use only with measurable, defined axes. Publish the scoring rule, evidence per point, treatment of unknown values, and sensitivity to axis choice. Do not place firms based on impression alone.

Survey evidence

Display weighted estimates only with:

• unweighted sample base;
• target population and frame;
• field dates and mode;
• weighting statement;
• appropriate uncertainty or an explicit reason it is unavailable;
• wording or instrument link;
• non-probability label where applicable.

Do not imply that overlapping intervals prove equivalence or that nonoverlap is the only test of a meaningful difference.

Regulatory timelines

Distinguish:

• announced or consulted;
• enacted/adopted;
• effective;
• stayed, repealed, or superseded.

Use the official publication and effective dates, not a news article date.

Accessibility

• Do not rely on color alone.
• Use direct labels where possible.
• Ensure meaningful reading order and alt text.
• Keep text legible at final output size.
• Use patterns or line styles for scenarios.
• Avoid red/green-only encodings.
• Provide the underlying table or data file.

Caption template

[What the display shows]. Geography: [scope]. Period: [period].
Measure: [unit, denominator, stock/flow]. Currency/base: [if applicable].
Observed through [date]; conditional scenarios thereafter.
Sources: [S-...]. Calculation: [CALC-...]. Retrieval cutoff: [date].
Limitations: [material caveat].

Visual audit

• Every plotted value is reproducible from a local table.
• All source and calculation IDs exist.
• Units, denominators, dates, and price bases are visible.
• Axis starts, scales, truncation, and sorting do not mislead.
• Missing/suppressed values are not rendered as zero.
• Forecast and historical data are visibly distinct.
• Rounding is consistent with evidence precision.
• The visual works in grayscale and with screen readers.
• The report does not require any figure to be considered complete.