Skip to content

site: enable GitHub Pages deploy and serve the live curated corpus - #71

Open
andrewsu wants to merge 1 commit into
mainfrom
worktree-gh-pages-setup
Open

site: enable GitHub Pages deploy and serve the live curated corpus#71
andrewsu wants to merge 1 commit into
mainfrom
worktree-gh-pages-setup

Conversation

@andrewsu

@andrewsu andrewsu commented Sep 3, 2026

Copy link
Copy Markdown
Member

Pages is now enabled on the repo with build_type=workflow (site URL: https://sulab.github.io/DrugMechDB-agentic/). This wires up the deploy and switches the site off the bundled sample data.

What changed

.github/workflows/deploy-site.yml — dormant → live. The workflow already existed with two deliberate safeties; both are lifted:

  • added the push trigger on main for web-revamp-mockup/** and the workflow file itself
  • widened the deploy guard to github.event_name == 'push' || (workflow_dispatch && inputs.deploy == 'true'), so a merge publishes automatically while a manual dispatch stays a build-only dry run unless someone passes deploy=true

The stale DORMANT / maintainer-checklist header is replaced with what the workflow actually does now. Added a sanity check that data/index.json is non-empty — the site runs in live mode, so without it the browse page would deploy green and render nothing.

Site → live data. dmdb-data.js flips SOURCE to "live". record.html reads ?id= so browse results deep-link to a specific record. browse.html gains a Curation run facet and shows per-record sourced/flagged edge counts.

scripts/build_site_data.py + web-revamp-mockup/data/ — the generator and its output: index.json + 135 records across the 4 agentic curation runs (59 distinct path ids).

Run Records
Opus (model eval) 40
Sonnet (model eval) 35
Phase-3 eval 31
Pilot run 29

Why commit the generated data

The deploy stays dependency-free (no Python, no build step) and the published site is exactly the reviewed diff — a data refresh goes through normal review like any other change. Re-run python scripts/build_site_data.py after a curation and commit the result. Total payload is ~1.4 MB.

Verification

  • workflow YAML parses; trigger and guard expressions confirmed
  • a dry-run dispatch on main succeeded (13s) — artifact packaging works against the real Pages config
  • the generator reproduces the committed data byte for byte (diff -rq clean)
  • all 135 index rows resolve to record files, no orphans in either direction
  • every page and JSON endpoint returns 200 over local HTTP
  • no root-absolute asset paths, so the site works under the /DrugMechDB-agentic/ subpath

Note for the reviewer

The site changes here (browse.html, dmdb-data.js, record.html, build_site_data.py, data/) were sitting uncommitted in the working tree; they're carried in as-is, unmodified. Deploying without them would have published the sample-data mockup instead. The mock banners now read LIVE · … but still say the design itself is a mockup.

On merge this publishes automatically to https://sulab.github.io/DrugMechDB-agentic/.

🤖 Generated with Claude Code

Pages is now configured with build_type=workflow, so wire up the deploy and
switch the site off the bundled sample data.

deploy-site.yml: lift the two dormancy safeties — add the `push` trigger on
main for web-revamp-mockup/** (plus the workflow itself), and widen the deploy
guard so a merge publishes automatically while a manual dispatch stays a
build-only dry run unless deploy=true. Stale DORMANT/maintainer-checklist
header replaced with what the workflow now does. Added a sanity check for a
non-empty data/index.json, since the site runs in live mode and would render
an empty browse page without it.

dmdb-data.js: flip SOURCE to "live". record.html reads ?id= so browse results
deep-link to a specific record; browse.html gains a "Curation run" facet and
shows per-record sourced/flagged edge counts.

scripts/build_site_data.py + web-revamp-mockup/data/: the generator and its
output (index.json + 135 records across the 4 agentic curation runs, 59
distinct path ids). Data is committed rather than rebuilt in CI so the deploy
stays dependency-free and the published site is exactly the reviewed diff;
re-run the script after a curation and commit the result.

Verified: workflow YAML parses; a dry-run dispatch on main packaged the
artifact successfully; the generator reproduces the committed data byte for
byte; all 135 index rows resolve to record files with no orphans; every page
and JSON endpoint returns 200 over local HTTP; no root-absolute asset paths,
so the site works under the /DrugMechDB-agentic/ subpath.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant