Share results

A standalone report helps a collaborator read the question, results and interpretation. A repository publication also preserves declared replication material, supporting inputs and authoring sources for review.

Prepare the local report

Start with the workspace tutorial, then check:

uv run brainlesslab workspace check WORKSPACE --json
uv run brainlesslab report check WORKSPACE/reports/REPORT_ID --json
uv run brainlesslab report build WORKSPACE/reports/REPORT_ID

Expect valid checks. Inspect build/index.html in a browser. The question, conditions, units, uncertainty, exclusions and limitations should be understandable without access to the conversation or workspace. Share the HTML directly when the article meets the collaborator’s need. The core and declared supplements supply replication material separately.

Choose supporting material

Retain the exact plans, captured source, resolved execution settings, seed definitions, actual environment and external inputs needed by declared results. Include analysis procedures and summaries. Compact trial tables are usually useful; large trajectories and recordings need a stated purpose.

Explain what is retained, omitted and regenerable. State the practical cost or limit of regeneration and which original analyses become unavailable. Required dependencies cannot be discarded to meet the 25 MiB total revision cap. The format reference describes retention and capabilities. A checksum inventory does not verify a rerun.

Export an immutable revision

uv run brainlesslab report export WORKSPACE/reports/REPORT_ID --revision 1 --output research/reports/REPORT_ID/v1

The new directory contains the article, authoring, core, selected supplements and release manifest. Export refuses to overwrite an existing revision. If the total exceeds the cap, inspect the size breakdown and revise optional retention or report scope.

Keep mutable local records and builds out of Git. Submit the complete exported revision in a contribution PR. Explain its question, selection, evidence state, retention limits and local checks. Keep failed or excluded attempts visible when they affect the claim.

Publish the article

The exported report is the reader-facing article. The article list automatically includes the latest revision, using its title, authors and description. Write that description as a short account of the question or finding. Replications use the same format, with links to the original study and an explanation of what changed.

Keep proposed investigations in the article’s further-work discussion. Add related article links when another contribution is published. For the current plasticity articles, the site adds these links through RelatedResearch.astro, preserving the original exported text.

Optional overview

A separate overview can help synthesise several articles when there is enough material to warrant one. The existing Initial weights and plasticity summary retains its address for readers who already link to it.

Create site/src/content/docs/experiments/SLUG.mdx. Its experiment_article frontmatter records authors, an ISO date, evidence_state and a versioned report route such as /reports/REPORT_ID/v1/. Quote date strings. Use updated for substantive changes after publication, keeping the original edition date. The shared page header renders these details.

Lead with findings, then explain the question, controls, task outcomes, uncertainty and interpretation. Define mechanisms and score units before using them to explain results. Keep the next proposed experiment distinct from completed work.

Use ReportFigure.astro to include a named raster view from an exported revision. It reuses the validated figure and caption, reserves image dimensions and links to the full-size image. Keep the Research sidebar focused on Articles and Contribute. The project skill’s references/experiment-articles.md contains a complete metadata and component example.

For an editorial correction, update the docs article. When results, methods or report presentation change, export a new report revision, then review the article’s claims, figures and revision links together. Earlier exported revisions stay intact.

Check the publication change

Include complete exported packages for every report revision linked by the article. Review the staged file list and outgoing commits so the publication scope is clear. Keep unrelated changes separate. Run the site tests, typecheck and build from a clean checkout of the intended commit; this catches reliance on local untracked packages. Check the rendered article, figures and downloads at desktop and narrow widths. A successful build verifies packaging and rendering; content review checks the interpretation against the retained analysis.

Review and website hosting

Reports use the existing Astro/Starlight website and Cloudflare Pages deployment. They require no account, database or upload service. The report-only validation job treats submitted source and analysis code as data. It checks versions, inventories, hashes, references, declared capabilities, size and article consistency.

Human maintainer review and merge accept a contribution. A maintainer can choose a rerun case by case and link its comparison receipt. Acceptance does not establish scientific confirmation or verified reproduction. For a completed rerun, the numerical comparison command records differences and declared tolerances. Retain full original run data when this independent check matters. Link its separate receipt; report export does not set verified flags merely from an authored claim. Use the report’s reproductions field to include a supported receipt, with both full original and replay records. Export recomputes comparisons before deriving the declared rerun capability.

Published reports appear under /reports/. Each revision has a stable route at /reports/REPORT_ID/vREVISION/; /reports/REPORT_ID/ selects the latest revision. The article links to declared downloads. Search indexes questions, findings and interpretation rather than large provenance tables.

Use a new revision for corrections. Accepted bundles remain unchanged, and older revision links stay valid. Link reproductions as separate checks rather than counting them as independent experimental results.