MARS website

The Metropolitan Aerospace Rocket Society website uses Zensical 0.0.68.

The production configuration in zensical.toml retains https://marstmu.com/. The private preview uses a separate ignored configuration and output directory; it never replaces the public site's files or DNS.

Install and preview

Python 3.11 or newer is required; the lockfile and this preview were tested with Python 3.12.

python3.12 -m venv .venv
.venv/bin/python -m pip install --require-hashes -r requirements.txt
.venv/bin/python scripts/preview.py --build
.venv/bin/python scripts/preview.py --serve

Backend: http://127.0.0.1:8887/.

Private review URL: https://workstation.tailee9084.ts.net:8887/. Tailscale is required. The HTTPS route proxies only this localhost backend; Funnel is not used. Preview HTML and responses carry noindex, nofollow, noarchive; robots.txt blocks crawling. The private sitemap is retained as a route inventory and uses tailnet URLs only; the public CNAME is removed from generated preview output. These precautions do not substitute for the tailnet access boundary. Text responses negotiate gzip while retaining no-store; images, videos, private attachments and all byte-range responses preserve their original bytes. This is private-server transport optimization, not proof of public-host performance.

The requested MARS Winter 2026 trailer is available as a 1080p H.264/AAC MP4 with native playback controls. Its highest-quality downloaded master (H.264/Opus in MKV, 1920×1080) remains untouched outside published docs in .local-media/. The preview-only /downloads/mars-video-full-quality route serves that one explicit file as an attachment; it does not expose the media directory. Both playback and the master download support HTTP byte ranges for seeking/resuming. The master is excluded from Git and the generated/public site. Any later public master hosting remains a separate publication decision.

To update the review copy after editing, rerun scripts/preview.py --build and refresh the browser. The static server intentionally does not rebuild large historical image collections continuously.

Migration and verification

Zensical's native blog and exclude implementations replace the active MkDocs plugins. Search is deliberately disabled, including its button and keyboard overlay. The modern theme uses a compact custom header, an image-led homepage, system-controlled light/dark appearance, and a clean social footer. Blog date/slug URLs, authors, archives, assets, and the original /sponsors/package.pdf download are preserved.

Markdown extensions have one consolidated configuration. The old duplicate YAML key is fixed, and Zensical uses its own documented emoji/icon helpers. Copyright is updated through 2026. The new dependency lock replaces separate floating package lists in local setup, CI, and the devcontainer.

# Build public-format output locally; this does not upload or publish anything.
.venv/bin/zensical build --strict
.venv/bin/python scripts/prepare_site.py site
.venv/bin/python scripts/check_site.py site

Image delivery and private originals

Both the preview build and CI run scripts/prepare_site.py after Zensical. Display-only WebP variants use content-addressed names, actual pixel widths, responsive srcset, EXIF orientation and embedded-profile color handling. Photographs are compressed; diagrams and transparent branding use lossless encoding. SVGs, favicons and animated GIFs remain unchanged. Existing manual previews resolve back to their original source before encoding, preserving the homepage's separate portrait/landscape crops. Original image files and gallery download links are retained verbatim. The video poster is encoded from a lossless decoded frame at 70 seconds, not recompressed from the old JPEG. This requires ffmpeg (installed in CI).

Public guide images and author avatars are pinned in docs/assets/external/ with durable source URLs and SHA-256 checksums. Routine builds require those verified copies and do not fetch remote images. To intentionally pin a newly added public image, first build locally and run scripts/prepare_site.py site --fetch-external, then review its manifest. This command only inspects images rendered in that build; it does not fetch excluded team pages. The generated delivery inventory is site/assets/display/manifest.json. Pages without .arithmatex do not download the MathJax engine. Math pages retain the pinned 3.2.2 renderer, loaded on demand after checking for real formulas.

.venv/bin/python -m unittest discover -s scripts -p 'test_*.py'

Optional browser/lab checks require Node 22.19+ (24 in CI), Chrome and the pinned CI tools:

npm ci --prefix ci
node scripts/image-performance-smoke.cjs
node scripts/mathjax-smoke.cjs
node scripts/check_lighthouse.mjs --preview --full --report-only

The image check verifies delivery bytes, decoded dimensions, responsive descriptors, transparency, lazy/eager loading and original download links across mobile DPR 2/3, desktop and both color schemes. Lighthouse reports are local simulated lab measurements, not field Core Web Vitals. MACH's reference targets are performance 75, accessibility/best-practices 95 and SEO 90; the private preview intentionally excludes its noindex SEO score from gating, never removes its privacy protection. Reports live under ignored .preview/ directories. CI blocks on image/media and math regressions; Lighthouse is explicitly diagnostic (--report-only): public-host performance is not yet validated against the reference 75 target, and CI's uncompressed stdlib server is not the private preview's gzip transport. Its reports are retained separately from the Pages artifact.

The separate original-image archive command and its verified manifest remain private under .local-media/; archive creation is never part of CI or the public build. The exact preview-only /downloads/mars-images-full-resolution attachment route supports range requests without exposing its directory. “Full resolution” means the best source files available in this checkout and fetched source URLs, not a claim of unavailable camera masters. Historical source exports and known minimized variants are identified by the archive manifest.

# Explicit private archive operation; never invoked by preview/CI builds.
.venv/bin/python scripts/archive_images.py
.venv/bin/python scripts/archive_images.py --verify-only
# Existing archives are protected; reviewed regeneration requires --replace.

mkdocs.yml is retained as the comparison/rollback configuration, with its duplicate extension key repaired. Optional legacy-builder dependencies live in requirements-legacy.txt; they are not installed for the Zensical preview.

requirements.in declares the intended runtime. requirements.txt locks its transitive dependencies and hashes. For a reviewed dependency update, use a separate maintenance environment with pip-tools==7.6.2 and regenerate:

pip-compile --generate-hashes --output-file requirements.txt requirements.in

Reviewed publication

The workflow builds and validates on pull requests and main-branch pushes. Publishing requires an explicitly requested manual workflow run with publish=true; the default is false. Zensical has no gh-deploy command, so the site uses GitHub Pages artifacts instead of force-pushing gh-pages.

After a maintainer approves publication, dispatch the existing workflow:

gh workflow run ci.yml --ref main -f publish=true

Before an approved public migration, verify representative historical URLs, blog images/comments, gallery, navigation, disabled search shortcuts, mobile layouts, both themes, and downloadable documents; configure GitHub Pages to use Actions; confirm the existing custom domain and DNS are retained; and review the content and updated sponsorship document. The local preview never changes GitHub settings or DNS.

The complete checker also gates internal pages/fragments (including absolute canonical-domain links), CSS/font dependencies and preserved downloads. CI runs all Python suites plus header/menu/history/parallax/founder and facade regression checks. scripts/browser-smoke.cjs --public tests public-format output without requiring the private server's noindex headers; the default still verifies them. Founder provenance and superseded/source portrait drafts are excluded from Git and generated outputs. Their untouched sources and lineage remain in the private image archive; selected portraits and original gallery downloads stay available.

References: Zensical migration, native blog, GitHub Pages publishing.