brfid.github.io

Source for brfid.github.io, a Hugo-based personal site, blog, and resume deployed to GitHub Pages.

The live site is built and deployed via a Hugo + vintage computing pipeline (VAX/PDP-11 via SIMH). The vintage pipeline generates rendered artifacts that Hugo includes in the final build.

Hugo

Public landing content lives in site.yaml; Hugo templates and assets live under hugo/. The site builds with Hugo extended ≥ 0.156.0 (ARM64 binary available from Hugo releases).

Front-end assets are vendored, not fetched at build or view time: the Newsreader and IBM Plex Mono fonts ship as woff2 files in hugo/static/fonts/ (self-hosted via @font-face, no CDN) and custom styles live in hugo/assets/css/extended/. Hugo bundles both — there is no separate asset build step.

Every HTML surface carries a site-wide noindex, nofollow robots directive, including blog posts, the resume, and the generated vintage build log, and Hugo does not publish a sitemap. robots.txt leaves HTML crawlable so search engines can observe noindex; it blocks the machine-readable pipeline status, public resume PDF, and RSS feeds. The deploy workflow fails if any published HTML omits the directive or a sitemap reappears.

Local preview

make preview

Open http://localhost:1313/. The preview command syncs site.yaml and resume.yaml, builds the public resume HTML and phone-free site/resume.pdf, and runs Hugo from that output directory. It is production-equivalent by default, and neither output reads the private phone overlay. Files under hugo/ live-reload. Restart make preview after changing resume.yaml, or after layout and CSS changes when the PDF itself also needs to be regenerated.

To generate the application PDF with a phone number, copy resume.private.example.yaml to the gitignored resume.private.yaml, edit the value, and run make resume-pdf-application. That explicit target writes local/bradley-fidler-resume.pdf outside the served and deployed site/ tree. Run make preview again afterward because the application build refreshes site/. make preview-public remains as a compatibility alias for the same phone-free preview. Set PREVIEW_PORT to use a different port, for example make preview PREVIEW_PORT=1314.

Build to site/

hugo --source hugo --destination ../site

Note: --destination is relative to the source directory, so ../site writes to site/ at the repo root.

Publish: pushing to main runs the vintage pipeline, builds the public resume HTML and phone-free PDF, verifies their navigation and privacy boundaries, and deploys the complete site to GitHub Pages. To skip a deploy, include [nopublish] anywhere in the commit message. workflow_dispatch is available for manual re-runs without a new commit.

Blog posts

Published posts are Hugo page bundles under hugo/content/posts/<slug>/. Create a draft bundle with make new-post POST_SLUG=<slug>, put the post in its generated index.md, and place any referenced images beside it. The scaffold is a draft by default; make preview-drafts includes it locally, while production ignores it until its front matter sets draft: false.

Working essay sources stay in the private career repository. Only approved publish copy and its public assets cross into this public repository. The Blog index is /posts/; RSS is available at the historical /index.xml route and the section-specific /posts/index.xml route.

Vintage pipeline (publish path)

The vintage pipeline's sole content artifact is the landing-page bio. The resume does not pass through the emulators; Hugo renders its HTML and Playwright prints the public PDF separately during the same production job:

• build/vintage/brad.bio.txt is the landing bio: bradman.c composes it as troff on the VAX, and the PDP-11 runs nroff to fill and justify it.
• hugo/data/bio.yaml is generated from brad.bio.txt and consumed by the landing template; it is not published as a file.
• hugo/static/build.log.html is the published machine-boundary build log with VAX and PDP-11 console sections, rendered by resume_generator/build_log.py.
• hugo/static/pipeline-status.json is the machine-readable result, build identity, commit, completion time, and per-stage artifact counts.

site.yaml owns the public identity (name, headline) and icon links; the landing bio text lives in resume.yaml's basics.summary, shared with the resume Summary and PDF. The pipeline flattens the identity from site.yaml and that summary from resume.yaml into build/vintage/bio.vintage.yaml; the link list goes directly to Hugo. Only basics.summary enters this pipeline; the rest of the resume document (work, skills, etc.) does not.

The pipeline uses pexpect to drive SIMH emulators via stdin/stdout (no telnet ports, no sleep-based timing):

• Stage B: VAX (4.3BSD) compiles and runs bradman.c to generate brad.bio.roff (troff) from bio.vintage.yaml, then uuencodes it to brad.bio.uu (scripts/vax_pexpect.py)
• Stage A: PDP-11 (2.11BSD) uudecodes the spool and runs nroff (no macro package) to fill and justify brad.bio.roff → brad.bio.txt (scripts/pdp11_pexpect.py)
• Stage A+B: VAX generates, host couriers the spool, PDP-11 renders

Both stages are implemented and validated on main. See git log for validation history.

See docs/integration/INDEX.md for design details.

Setup

Use Python 3.11 or newer.

python3 -m venv .venv
.venv/bin/python -m pip install -e '.[dev]'
.venv/bin/python -m playwright install chromium

Run make help to list common commands for tests, the Hugo build, and resume generation.

Quality checks

make check

make check is the shared local and CI quality gate: Ruff lint and format checks, mypy, all non-Docker/non-slow tests, Pylint, and Vulture. Use make test when only the fast test suite is needed.

Publish path

deploy.yml runs the whole publish path on the ubuntu-latest runner itself: the vintage pipeline, Hugo build, phone-free resume PDF generation, privacy checks, and Pages deploy. There is no execution host to start or stop and no cloud account involved.

bash scripts/edcloud-vintage-runner.sh "build-$(date -u +%Y%m%d-%H%M%S)" > /tmp/stdout.txt

That single line is the whole pipeline invocation. The runner does exec 3>&1 before redirecting its own stdout to a log file, so redirecting it to a file puts the base64 artifact markers on the parent's stdout while the verbose pipeline log stays in $LOG_DIR, which is how CI extracts artifacts with awk.

Single orchestration entrypoint: scripts/edcloud-vintage-runner.sh (the edcloud in the name is vestigial; it has no cloud coupling). Prebuilt GHCR images provide the emulators and disk images, while the runner bind-mounts the current checkout's pexpect scripts and shared session helper over the copies baked into those images. This keeps cached infrastructure from silently pinning older orchestration code. Set KEEP_IMAGES=1 to preserve Docker images between runs (avoids rebuild on retry).

Historically (until 2026-07-25): CI authenticated to AWS, started an EC2 instance ("edcloud"), ran the same script there over SSM, and extracted the artifacts from the SSM output. That account was decommissioned; the brfid/edcloud repo is retained but no longer provisions anything for this site: brfid/edcloud.

Testing the pipeline without publishing

.github/workflows/vintage-validate.yml (workflow_dispatch only) runs the pipeline on a hosted runner, publishes nothing, verifies that the rendered bio round-trips the public fields in site.yaml (name and headline verbatim) and resume.yaml (basics.summary word-for-word since nroff reflows it), and optionally compares brad.bio.txt against a known-good hash. SIMH emulates the guest CPU deterministically, so an unchanged render is evidence that a host or toolchain change was inert.

# Record the current hash, comparing against nothing
gh workflow run vintage-validate.yml --ref main

# Compare against a recorded baseline
gh workflow run vintage-validate.yml --ref main \
  -f expected_sha256=<recorded-hash>

The bio carries no date (bradman stamps the build date only in a troff comment, which nroff strips), so its raw SHA256 is stable across runs and directly comparable to a baseline. The baseline is tied to the current site.yaml; update it when the landing copy changes.

Artifacts and the full pipeline log are uploaded on success and failure alike.

Vintage pipeline prerequisites

• docker, git, python3 on the host. Nothing else — no AWS, no secrets.
• Pre-built images pulled from ghcr.io/brfid/{vax,pdp11}-pexpect:latest, with a local docker build fallback. Both are amd64-only, so they run natively on ubuntu-latest but require emulation (or a multi-arch rebuild) on an arm64 Mac.

Source of truth

Doc	Role
STATUS.md	Operational state for site, pipeline, resume surfaces; procedure and active ops priorities
~/src/career/STATUS.md	Strategic arc (sequencing, posture, when/why phases run) that gates this site's reactivation
docs/integration/INDEX.md	Vintage pipeline design, stages, artifacts, constraints

Cold start order

1. This file
2. STATUS.md
3. hugo/ (Hugo site root — theme, content, config)
4. ~/src/career/STATUS.md (only if touching site reactivation, resume surfaces, or any phase-gated work)

Then apply AGENTS.md constraints.