Vuer documentation preservation — 2026-09-07
Source preservation is complete for all 149 upstream tags, including prereleases and historical aliases. Exact source commits are on archive/<tag> branches in https://github.com/vuer-ai/vuer-docs. Independently migrated Dockit branches are on docs/<tag>, each initially created with the original source commit as its parent, then advanced through tooling-only updates, with a docs-site/migration-source.json provenance record. The branch manifests include the source and migration commit IDs; remote refs were checked against every entry at capture time. A separately attached final branch manifest records subsequent tooling updates; the original snapshot remains immutable.
The ReadTheDocs rendered-build capture is partial: 270 files across 73 versions were recovered before ordinary HTTP requests began receiving Cloudflare challenge responses. The crawler stopped without solving or bypassing those challenges. All 117 versions reported as built by the public API have manifests; these do not imply complete snapshots. The existing ReadTheDocs project remains the authoritative original rendered archive. ReadTheDocs HTML ZIP downloads were not enabled and returned HTTP 404.
rtd-versions.json contains all 168 public version records (119 active, 117 built). rtd-builds.json contains all 681 publicly accessible build-attempt records. The authenticated UI reported 711 attempts; the additional records were not available in the public API capture. These are provenance records, not 681 separately recoverable HTML builds. 534 public attempts succeeded, spanning 243 distinct commit IDs. Deleted or overwritten previous build output is not claimed recovered.
rendered-archive/objects/<sha256> preserves downloaded response bytes without rewriting. Each version manifest records original URL, path, content type, byte count, SHA-256, Last-Modified, ETag, and failures. External-host resources are recorded as references and are not mirrored. Crawl coverage includes HTML src/href/srcset/poster/data-src, CSS url/@import, and Sphinx search-index docnames. Checkpointing permits resuming after an authorized access change. The crawler honors Retry-After, uses a global throttle, and stops globally on a Cloudflare challenge.
Restore saved files with checksum verification:
python3 restore_rtd.py rendered-archive restored-siteThe resulting tree preserves /en/<version>/... URL paths. Do not treat the restored partial tree as a complete replacement for ReadTheDocs.
generate_doc_branches.py records the deterministic overlay procedure and refuses to overwrite an existing differing branch. The runtime source is the historical commit, not current source relabeled as an old release. Earliest releases without documentation render their own README and their own Python module API (tassa before vuer). Production builds were checked for the first tag, RST-only documentation, and current documentation; the autodoc repository has a separate generation/link validation report for all 149 tags.
Completed Dockit version migration
All 149 historical source tags were rebuilt with DreamLake Dockit and published as separate branch deployments. verified-version-deployments.json records every immutable URL, deployment ID, exact documentation commit and upstream source commit. Each deployed site also exposes /build-provenance.json. All 117 original published Read the Docs versions are represented; the other 19 Read the Docs records outside the tag set were inactive, never-built development branches.
The new production site is https://docs.vuer.ai. Its version menu uses these verified immutable builds. snapshot-* deployment branches are preservation-only: never replace their latest successful deploys. Corrections must use a new snapshot branch name. Git-based deployment of editable docs/* source branches remains separately configured.
snapshot-publisher-scripts.tar.gz contains the reproducible build, guarded upload and catalog reconciliation scripts. snapshot-http-audit.json records HTTP checks of every snapshot homepage, its local scripts and stylesheets, and its source provenance. MIGRATION-SHA256SUMS covers these additional migration artifacts. These migrated builds are separate from the explicitly partial original rendered capture above.
