orx-results

Validate before reporting

Never infer a result from run status or memory. Before accepting or reporting a run-derived claim, confirm that:

• the log identifies the variant and effective configuration;
• the final metric and compact summary are present;
• the relevant trajectory is recoverable for a long run; and
• the cited file lines actually contain the supporting output.

Load orx-logs when supporting output still needs retrieving. Cite the verified measurements using the session playbook’s evidence-and-links contract.

Summarize an experiment

Choose one primary representation for each measured comparison: a graph, a table, or prose. Explain the conclusion and limitations in brief prose.

Choose prose and selective visuals that make the result easiest to understand. If you choose to use multiple representations or visuals, make sure they do not contain duplicate data or show the same thing two different ways. They should each offer a distinct way of interpreting the experiment’s results — one visual will suffice for most experiments. For example, do not show both a table and graph of the same data — just show the graph.

An interactive graph and linked source data already provide exact-value lookup; a table of those values is not a useful fallback. Subsetting, condensing, or deriving values from the same measurements does not make a second representation complementary. Put useful headline values or changes in the prose interpretation instead. Before sending, remove any table or visual that repeats a comparison already represented.

Briefly explain what changed, what ran, and what the evidence shows. When the measured results suit a supported orx-figures template and a graph makes the result easier to understand, prefer an interactive figure inline alongside that summary. Reuse an existing figure where possible and link its source data.

This is a rule of thumb, not a required output. Skip the visualization when there is no suitable structured data, no appropriate template, or no useful visual comparison. Do not invent data, additional runs, or graph views to satisfy the preference; a concise text summary can be the complete result.

Default to one graph. Offer a dropdown only when multiple views of the same data answer distinct, useful questions. Multiple series alone do not require a dropdown, and one useful view does not need a selector.