MCRP / Research prototype

Put the candidate on publication surfaces without changing its meaning

Historical preparation plan. See the later launch decision for the authorized agent-first deployment and adopted packet licenses.

Prepared integration plan for the existing outward-facing Jekyll research site. No site file, public repository, deployment setting or external account was changed by this packet. Paths below are proposed placements, not existing public URLs.

Use the existing site for the introductory post and a link to the versioned research object. Serve the complete reading/code export as a plain static artifact or downloadable ZIP. Keep its source and result manifest alongside it. A separate static export repository, if approved, should contain only the curated export, not the private research repository’s Git history.

The candidate builder includes .nojekyll for a standalone static export root. Do not copy that file to the existing Jekyll site’s root: doing so would change how that whole site builds. The marker is a hosting hint, not an integrity proof; the chosen deployment route still needs an actual retrieval check.

A full export contains Markdown sources, including front matter. Copying all of those sources into a Jekyll tree and assuming they will be served unchanged is unsafe: the site generator can interpret source files or create competing rendered paths. Either serve the curated export without that transformation, or publish its ZIP and PDF as static downloads while intentionally integrating the blog. The download is the canonical self-contained reproduction object.

Prepare the site post

The editorial source is blog-introduction.md. Its front matter includes the current site’s layout, categories, publication_lane, AI disclosure and related-post conventions, with published: false. It is ready for a draft integration, not an authorization to turn that flag on.

A proposed filename is _drafts/mcrp-what-can-we-rely-on.md. On a later approved publication, choose the actual date and _posts/YYYY-MM-DD-...md filename. The site’s layout supplies the visible title, so remove the source’s duplicate H1 when adapting it. Preserve the contribution disclosure and distinction between internal agent review and external human review.

Resolve all relative links against the selected public export, including these entry points:

Link in editorial source Artifact entry
Human exercise onboarding/human-path.html
Implementer quickstart onboarding/agent-path.html
Domain cases domains/README.html
Collective critique reviews/COLLECTIVE.html
Positioning comparison publication/positioning-and-adoption.html

Do not invent a working destination or DOI before one exists. A draft can retain clearly labeled unresolved destinations; the publication candidate cannot. If the initial surface provides only a ZIP, make the download the entry and explain that index.html is its local starting page. Do not publish relative links that happen to work only in the authoring checkout.

Exact release sequence

  1. Select the responsible attribution, rights disposition, destination and feedback owner for a specific source and export inventory. The recommendation remains public release of a research seed, with the declared synthetic evidence limits.
  2. Build and inspect the post in the site’s own toolchain. Check navigation, intended links, contribution disclosure, mobile layout and dark mode. The standalone packet’s browser checks do not substitute for this integration check.
  3. Record the final site-post bytes, exported ZIP digest and public path mapping. Site layout changes can alter the rendered post; bind the approval to the rendered candidate as well as the editorial source.
  4. Publish only within the approved scope. Retrieve the public ZIP and manifest, compare them with the approved local bytes, and inspect the actual public post. A successful build or deployment command is not evidence of successful delivery.
  5. Record a dated observation. Later source changes require a new version and a new decision for the affected public artifact. Keep old versions identifiable; never silently replace a DOI or download label with different evidence.

These steps make the proposed launch concrete. The current task prepares the candidate and its evidence; it does not claim that an external page was deployed or observed. A next publication instruction can name the exact destination and byline rather than reopen the research design.