Reading the learning path on the web

The project publishes its public Markdown documentation as a small static website. Static means the server returns ready-made pages: no account, browser script or AI service is needed to read them. The learning path is the home page, with links to the glossary, Monday demonstration, remaining work and Ask OKF.

Open the learning website or its direct learning-path address. The first public release was verified on 20 September 2026 against source commit 3f72ebc30128c2a3171951050a566d3ed8db7c16: the manifest and all 108 output files returned HTTP 200 and matched their expected byte counts and hashes. The retained observation is a dated publication check, not a guarantee about future availability or later versions. A separate browser check opened the learning path and glossary without console errors or page-level horizontal overflow.

Where the words come from

The Markdown files in this repository remain the authored source. The build renders Git-tracked files in docs/, evaluation/ and a small explicit set of root documents. It excludes private correspondence, untracked research and the large corpus bodies. The separately admitted Evidence workbench catalogue also publishes 40 selected evidence packages and their checked delivery parts. It does not copy arbitrary evaluation folders. A source link on each page opens that exact Git version. Links between published guides stay on the website; evidence files open their commit-bound repository location. The page footer links a manifest listing the source and output hashes. A hash is a fingerprint used to detect changed bytes.

The renderer escapes source HTML and runs no page scripts. The learning path's source-roles Mermaid diagram has a checked static SVG for this website, with its full explanation and links in the table below it. GitHub can render the authored Mermaid diagram. Other Mermaid blocks remain readable diagram source on the website. Browser keyboard and narrow-screen checks are reported separately from publication success.

Rebuild and publish

uv sync --locked --project tools/learning-site
uv run --locked --project tools/learning-site python -m unittest discover -s tools/learning-site -p test_learning_site.py
uv run --locked python scripts/source_roles_diagram.py --check
uv run --locked --project tools/learning-site python scripts/build_learning_site.py --commit "$(git rev-parse HEAD)"

The output directory must be new: the build refuses to overwrite a candidate. The renderer has its own locked environment under tools/learning-site/; this keeps the frozen evidence producers’ dependency identity unchanged. It is the same Markdown renderer family used by OKF Explorer. Do not edit generated HTML. Change the Markdown, build and test, then use a normal reviewed pull request. After protected main passes the full bundle validation, GitHub Actions builds its exact commit and uploads that single artefact to Pages. The Pages job does not regenerate or change evidence bundles.

Verify the retained first-publication receipt without contacting the website:

uv run --locked python scripts/check_learning_site_observation.py

Publication of a guide does not establish complete answerability, specialist acceptance or a source document's current legal applicability. Read the notice and remaining work.

Verify a workbench publication

The current exporter uses a version 3 manifest when the 40-question workbench is present. It admits only the declared public packages and parts, within a 64 MiB total website limit. The older version 2 verifier remains available for its historical publications.

node --test scripts/test_learning_site_v3.mjs
node scripts/verify_learning_site_v3.mjs --repo REPOSITORY --site BUILT_SITE --commit FULL_COMMIT --output FRESH_OBSERVATION_DIRECTORY

The version 3 check compares the live files with an exact local build and its immutable Git inputs. It reads at most four public resources concurrently, bounds downloads and records failures without silently retrying. This checks publication identity; the separately pinned all-question reconstruction test checks the delivery contract, and browser journeys check actual use. None establishes a correct benefits answer.