Generate a chart that can answer the user's question; combine multiple charts only when complementary perspectives are needed. By default, show only the title, chart, units and real Tooltip, retaining necessary static legends and color scales; other interactions come only from explicit requirements.
Choose meaningful charts
First clarify what the observations are, what is to be compared, field meanings/units and data ranges. Only check missingness, valid samples within groups, constants or point density that would affect the choice; do not automatically generate audit reports. A numeric ID does not become a measurement metric just because it is a number, and uniqueness alone does not prove that a field is an ID.
According to question and data conditions choose the chart meaning. When the meaning is insufficient, clarify only the questions that would change the chart meaning. Respect the user's choice when a chart type has already been specified; if the data does not satisfy the contract, explain the reason, and do not silently change the chart, guess bandwidth or fabricate results.
Locate index the entry of the selected type, read its guide, typeFile and necessary import dependencies, and the example's minimal input. The index has one chart per row and can be located by type; do not read the complete joint declaration or all guides. Only for special variants, read the options and fixed links for the corresponding type in variantFile. For common numeric, interval or color conditions, see common constraints. TS is a field reference, and final validation is still performed by the common validate/compile.
Generation and delivery
Reuse the generator; do not rewrite HTML or drawing calculations.Input and interaction protocol maintains table binding, direct Spec, budget and on-demand controls.examples/ consists of six executable demos; the basic examples have no controls, interactive.json is for reference only for explicit interaction requirements, and demo data must not be treated as user data.
python3 /实际技能目录/scripts/generate_visualization.py \
--request /实际请求目录/request.json \
--output /新的输出目录
Python 3.11+, CSV/TSV/JSON only require the standard library; XLSX optionally uses openpyxl and requires an explicit sheet. When the default interpreter is too old, specify an existing available version. Output HTML, a replayable request and data with provenance; the output directory must not exist. HTML embeds the data used, with fixed VChart/statplots 2.1.8-alpha.0 CDN; opening requires network and external script permissions, and does not require npm, TypeScript, MCP or a local server.
The title should state the actual variables, comparison objects and necessary ranges, and axes/Tooltips should retain units. Deliver the HTML link and the data ranges, omitted missingness or small-sample limitations that affect interpretation; do not treat correlation as causation and do not add analysis panels. Do not expand aggregate frequencies, do not sample/truncate by default, and do not guess intervals or dimensionality-reduction coordinates; use the analysis workflow for original model computation and full inference reports.
Acceptance and continued chart modification
Actually open the HTML to check the chart elements, Tooltips and the requested interactions.window.__DSV_STATUS__.ready only indicates the end of the batch; success requires each chart's status=rendered; input/omission counts and detailed errors are kept in the status object. A file or canvas alone cannot prove usability; when the host/CDN is restricted, explain the incomplete acceptance.
For conversational chart modification, read visualization.request.json from disk, retain all used rows, IDs, units, fixed conditions and on-demand options, modify the request and generate a new directory. When adding fields that were not embedded, restore them from the known original data by real ID; the request is not a full-column backup. Subsequent browser selections are not automatically written back to disk; when they cannot be read, confirm the specific conditions; provide saveView only when the user requests saving the current view.
Maintainers modify the upstream controller source and generate copies via the sync script; do not manually edit assets/statplot-controller.js. references/statplots is a generated snapshot of release assets from the same version; types are copied byte-for-byte, and the guide is extracted by chapter. See the manifest for sources and conversion rules. Users do not need to run the sync.