Communicate with evidence

Lead with the decision or finding. Show the mechanism, then the smallest chart that makes the evidence clear. Put detailed settings, records, and caveats in agent notes. A reader should understand the question before opening a technical appendix.

Choose the picture that answers the question

The examples below are deliberately synthetic formatting fixtures. They show how to communicate a future study. Their values are not estimates from the arena and must never be cited as strategic results.

Curves: show what changes

Use curves for calibration, build timing, cost sweeps, or changes through a session. If measurements are noisy, publish intervals in the accompanying table or add a reviewed interval-capable design component. The current generic line plot does not draw uncertainty bands. Do not imply that a smooth line supplies uncertainty information.

Scatter plots: show the price of an improvement

A win-versus-concession plot can reveal whether social gains require costly departures from the fixed reference. Compare equivalent compute budgets or label their differences. A point needs a retained cohort behind it before it can be marked measured.

Heatmaps: show where an approach fails

A four-player cell must identify all three opponents. Link a measured heatmap to a report carrying planned, completed, incomplete, and invalid counts, uncertainty, and the exact lineup for each cell. Blank cells mean unmeasured, never a zero win rate. This compact component displays recorded cell values; the report owns their statistical interpretation.

Write an approach page

Use the authoring template. Every idea gets a matching registered SVG near the top, one concrete decision, a mechanism, a falsifiable study, an appropriate visual outcome to collect, and an explicit evidence status. Keep an approach as a proposal until a retained result supports an update.

For a new animation, design a specific explanatory sequence. A tradeoff frontier should look different from a belief update or an expansion race. Use a gentle loop that stays readable throughout. Respect reduced-motion preferences and omit animation pause buttons. Captions should describe the mechanism in one short sentence.

Asset and component contract

Rich MDX accepts literal string attributes only. No JavaScript, imports, ad hoc HTML, styles, or remote embeds. Visual implementations live in the design Storybook, exported through @settlers/design/research; web loads and validates the data. The same registered figures are available in agent notebook rendering.

<StrategyFigure approach="reputation" />
<StrategyAtlas asset="strategy-atlas" />
<PliabilityLab />
<ResearchPlot asset="example-tradeoff" />

Store small JSON exhibits in content/assets/. Plot fields include schema: 1, kind: "plot", variant, title, caption, evidence, source, labelled x and y domains, and up to five named series of points. Variants are line, scatter, and heatmap. A heatmap uses integer cell coordinates, category tick labels, and a color domain with a unit label. The checked-in examples are complete schemas to copy and replace.

Use illustrative for synthetic teaching data, derived for an exact calculation with its formula, and measured only for retained observations. Include a run or artifact reference and denominator in a measured source/caption and link the evidence report nearby. Schema validation checks the data shape; it cannot establish that a claimed source is true. Evidence review remains required.