Roadmap

Four milestones, in dependency order. Each has an umbrella tracking issue in this repo where progress is checked off; implementation issues are filed in the repo where the work happens (where to file issues) and linked from the umbrella’s checklist. This page is the narrative; the issues are the ledger. When they disagree, fix whichever is wrong.

flowchart LR
  M0["M0 · Operational readiness<br/>hygiene, CI, health checks"] --> M1["M1 · Production cutover<br/>flipped 2026-09-28"]
  M1 --> M2["M2 · Full disconnect<br/>retire master, staging,<br/>build system, stats"]
  M1 --> M3["M3 · Search + redesigns<br/>prove the PR process"]

M0 — Operational readiness (#1)

The floor the rest stands on: every repo public, consistently named, licensed, contributable, tested, and watched. Mostly done 2026-08-21.

Acceptance criteria

• Four public repos under the bioc-* convention, each with LICENSE, CONTRIBUTING, CODE_OF_CONDUCT (the original four: bioc-website, bioc-edge, bioc-registry, bioc-infrastructure; bioc-build and bioc-manifest, added since, have all three as of 2026-09-30)
• Histories verified free of credentials and undisclosed findings
• CI with badges on every repo
• Daily black-box health checks with status badges — serving (health) and data plane (freshness)
• CD: the workers deploy from CI on main, not from laptops
• This roadmap, with umbrella issues

Dependencies: none.

M1 — Production cutover (#2)

Done: bioconductor.org has been served by the new edge since 2026-09-28 (nameservers to Cloudflare at 17:30 UTC, the site at about 20:09 UTC; runbook and log). What remains is closing out the flip: the week of monitoring ends 2026-10-05, and Route53 stays the rollback target, with a record freeze, until 2026-10-12. How the system works now is on How bioconductor.org works; who runs it and the rollback steps are on Operations.

Several pre-flip gates were overtaken by the flip itself. They are restated below as what the running system has to show, with what replaced each.

Acceptance criteria

• DNS flipped (2026-09-28)
• BiocManager::install() acceptance passes for release + devel against bioconductor.org (3.23 and 3.24, after the flip)
• Behaviour under real load measured, in place of a synthetic load test: 530,455 requests in two hours on 2026-09-29, 63% served from the edge cache, no Worker exceptions (Logpush, 16:00-18:00 UTC)
• Rollback written up: Operations, roll back DNS and the DNS runbook. Not rehearsed; the lever stays in place until 2026-10-12
• Parity with master, in place of the pre-flip crawl diff: a probe of 13,271 paths from production logs (2026-09-29) found the differences in bioc-edge #43-#48, all open; parity.yml compares production with master daily
• A week of monitoring with no unexplained regressions (ends 2026-10-05). Regressions found and fixed so far are in improvements
• Logging continuity: gap alerting is running (bioc-logpush-check, CI & scheduled jobs); both log eras queryable as one is not
• Upstream security findings resolved with the Bioconductor core team
• Core team sign-off on the cutover as it stands (no sign-off is recorded here)

Checkboxes here mirror #2.

Tasks

• Cutover runbook page (DNS, verification, rollback): DNS runbook
• Flip
• Fix bioc-edge #43-#48, then close M1
• Decide www. / secure. and satellite hostnames (DNS runbook, open questions)
• Walk the security findings with the core team
• After 2026-10-12: end the record freeze and retire the Route53 zone, or decide to keep it

Dependencies: M0 — the daily checks are what watches the flip.

M2 — Full disconnect from legacy infrastructure (#3)

The new stack serves production but still depends on master: packages arrive by the hourly sync from master’s docroot. (The /packages/stats/ tree no longer does: it is generated by bioc-traffic and served from R2.) M2 ends the dependency: master, staging, the legacy build system, and the legacy stats pipeline all retire, and roughly $5,000/month of AWS with them. Parity with master (bioc-edge #43-#48) has to be clean before master goes.

Acceptance criteria

• Prose content changes by PR in bioc-website. Pages are being ported one at a time; until the core team decides who owns the content, Bioconductor/bioconductor.org stays the source (Astro site build)
• Hourly rsync from the legacy docroot retired — new content arrives only via CI builds and registry propagation
• Annotation, experiment, and workflow packages have a build home (r-universe covers software only — the open workstream)
• Download stats derive from the new edge logs, historical eras joined
• Search off legacy Solr: production search is Pagefind over the Astro build (checked 2026-09-30); the design ADR stays in M3
• master + staging snapshotted, powered off, DNS retired
• The legacy AWS bill ends

Tasks

• Port prose pages into bioc-website one at a time; decide content ownership with the core team
• Sync-source retirement plan (what still changes upstream post-cutover, and who publishes it instead)
• Software repositories served from bioc-registry instead of the BBS mirror: plan, #35
• Build-home decision for annotation/experiment/workflow, with the core team and the r-universe track
• Stats surface + continuity decision; then drop the /packages/stats/ passthrough to master: master’s design kept, pages generated from the R2 tree (bioc-traffic#28); passthrough dropped 2026-10-16 (bioc-edge#68)
• Final snapshots per the decommission plan; decommission
• Record the end state in the migration record

Dependencies: M1, done except for the parity fixes, which M2 needs too.

M3 — Search and page redesigns, through the public PR process (#4)

Two goals in one: a real search story, and visible redesign work used deliberately to prove the contribution loop — PR → preview URL → review → merge → route flip.

Acceptance criteria

• Search design recorded as an ADR (scope, index strategy, fate of the Solr routes)
• Search live on the production Astro-built routes
• Two or more page redesigns shipped end-to-end through the public PR flow
• Contributor docs good enough that an outside contributor gets a preview URL unaided

Tasks

• Revisit the July 2026 findability and UX/IA audits (audit/review/); pick candidates
• Search ADR — Pagefind already indexes the Astro build; measure it at full-site scale first. Input: the unified search proposal
• Prototype search over the full ported surface
• Ship redesign PRs #1 and #2; fold the friction into CONTRIBUTING
• Invite at least one outside contributor through the flow

Dependencies: none left; M1 put the Astro build in production.