ADR 018: AV migration track starts now, in parallel — supersedes ADR 009's strict ordering (not its risk analysis)
Status: Accepted — 2026-09-04, ratified by Yuji Shinozaki, Than Grove and Xiaoming Wang. Date: 2026-09-04 Deciders: Yuji Shinozaki (Lead Architect), Than Grove, Xiaoming Wang — all three present when this was decided.
Header correction, 2026-09-04 (metadata only, decision text untouched): this ADR was first recorded as "accepted by Yuji alone, to be confirmed with Than and Xiaoming." That was wrong — Than and Xiaoming were in the session and the decision was ratified by all three at the time. Corrected in the same way ADR 004's Status line was updated when ADR 013 superseded it; nothing below this header changed. Supersedes: ADR 009 — the ordering only (Phase 3's "AV strictly after Texts/Sources close" sequencing). ADR 009's risk analysis, its Phase 0/1 content, and the "AV is hardest" conclusion are unchanged and restated below, not revised. Relates to: Sprint 2 (closing out today), AV/Sources/Texts migration complexity comparison (2026-09-01 — the evidence this ADR relies on and does not revise), Spike 7 (Kaltura, still Pending), Spike 11 (AV transcript replication, still Pending), roadmap.md (the still-open "AV transcript format" Phase 0 item this ADR reactivates). Forward references added 2026-09-04, same day, once scheduling detail resolved into concrete sprints: Sprint 3 (AV core — this ADR's track, made concrete) and Sprint 4 (AV transcripts, split out as a dependent, independently-schedulable sprint — see Sprint 3/4 for the rationale; this ADR's own Decision text still describes AV as one track, which the sprint split refines without reversing).
ADRs are immutable; this ADR does not edit ADR 009's content. ADR 009's Status line is updated to point here, matching the precedent set by ADR 013 superseding ADR 004. ADR 009's Context, Phase 0/1 content, and consequences remain accurate history and active reasoning — only the Phase 3 ordering claim ("AV last, after other risks are retired") is superseded.
Context
ADR 009 sequenced migration as: Images pilot (Phase 1) → fork to parallel site tracks, Texts (Than) and Sources (Xiaoming) (Phase 2) → AV last (Phase 3), "intentionally last, after other risks are retired." That ordering was not a guess — the 2026-09-01 complexity comparison, built from all three sites' real content-model audits, scored AV hardest (3.8/5) vs. Sources (3.2) and Texts (2.0), and explicitly concluded it validates ADR 009's sequencing rather than revising it.
That risk assessment has not changed. What has changed is team availability:
- Than is closing out Sprint 2 today and will be away for the next two weeks. Than is the team's most up-to-speed person on both Phase 2 tracks (Texts and Sources) — those tracks stall in his absence regardless of any AV decision.
- Yuji (Lead Architect) is better suited to AV of the two remaining active drivers.
- Sprint 2 (the last piece of Phase 1's Images work, per the roadmap's phasing) closes today, so the team is at a natural fork point already.
Two Phase 0 items from the original roadmap were never actually closed out, and remain
open regardless of this ADR: Spike 7 (Kaltura module landscape / upload-ingest
path — still ○ Pending, completely unstarted) and the AV transcript format triage
(plain text vs. structured/time-coded — the roadmap's one still-open question, gating
whether AV transcript work can reuse Spike 4b's now-proven structured-Tibetan
rich-text pattern, closed 2026-08-07, or needs its own proof from scratch).
Decision
AV becomes an active parallel track starting now, alongside whatever pace Texts/ Sources continue at without Than, rather than waiting for Phase 2 to fully close before Phase 3 begins. This is a resourcing decision, not a revision of the risk call: AV is still the hardest site by the team's own evidence, and this ADR does not dispute that.
Concretely, AV work begins with the two unclosed Phase 0 items, in the same spirit as how Images' own pilot began with cheap de-risking probes before real content-type migration — not with implementation:
- Spike 7 (Kaltura integration) — module landscape survey, playback-path prototype, and in particular the upload/ingest path and the Kaltura partner/credential re-provisioning question, since that is the one AV risk that is an external operational dependency rather than a Drupal-side data-shape problem.
- AV transcript format triage — resolve plain-text vs. structured/time-coded before scoping Spike 11, so it's known up front whether Spike 4b's proof transfers or a new one is needed.
Real AV migration implementation (content-type migrations, the OG access-realm work,
the MISSING_TYPE corrupted-bundle triage) follows only once those land, mirroring
Phase 1's own "probe, then build" structure.
This does not, by itself, decide what happens to Texts/Sources while Than is out — whether they pause, or someone else covers a reduced slice — that is a separate call for the team, not resolved here.
Why this doesn't contradict the 2026-09-01 comparison
That comparison scored AV hardest and said so explicitly to validate ADR 009, not
license moving it earlier. Nothing in this ADR disagrees — AV remains the compounding,
higher-risk site (nested field_collection entities, an external Kaltura dependency,
two extra OG access realms, the most unresolved unknowns). The reason to start it now
anyway is that the alternative — leaving AV untouched until Texts/Sources fully close —
assumed both of those tracks would have a driver making steady progress throughout, and
that assumption no longer holds for the next two weeks. Two Phase 0 probes (not
content-type migration) is exactly the calibrated-risk shape ADR 009 itself already
established as the right way to start a hard, unknown-heavy track.
Consequences
- AV's Sprint 2 content-model audit (C1) already ran ahead of the original per-owner sequencing, as a pulled-forward group effort (2026-08-28 team decision, recorded in the Sprint 2 doc). This ADR continues that same trajectory into real spike work; it is not the first time AV has moved earlier than ADR 009's letter.
- The single-shared-pattern rationale from ADR 009's Phase 1 (build the pattern once, mob-style, before forking) is unaffected — AV still reuses the Migrate API consolidation pattern, KMaps productionization, and proxy-auth integration Phase 1 already proved. Starting AV's probes earlier in calendar time doesn't mean reinventing that foundation.
- Kaltura re-provisioning is a live, unretired risk (Spike 7's central question) — if it turns out non-trivial, that could stall the newly-parallel AV track on its own, independent of anything Texts/Sources-related. This was always a known risk under ADR 009; it is simply reachable sooner now.
- The roadmap's open "AV transcript format" question must actually be answered as part of this track's first work, not deferred again — it directly gates Spike 11's scope.
- If Than's absence resolves faster than two weeks, or Texts/Sources need a driver sooner than expected, revisit — this ADR reflects the circumstances of 2026-09-04, not a permanent reprioritization.
Verification
Not yet started — this ADR authorizes Spike 7 and the AV transcript format triage to begin; neither has run as of this writing.