Each one already carries a recommended option (marked rec). Confirm, redirect, or skip — nothing here is locked until you say so.
Who’s deciding?
1 · R2 orphan-sweeper + lifecycle + customMetadata?
Closes the design’s named “sweep later” gap — a daily Cron job reconciles R2 against D1. No new product surface.
2 · App-level artifact versioning?
R2 has no native versioning — this adds versioned keys + a history read tool, reaped by lifecycle. Opt-in, no hot-path change.
3 · Vectorize semantic search beside FTS5?
Hybrid keyword + meaning. Adds a 4th write target + Workers AI dependency. The plan is written and ready.
4 · Vectorize tenant-isolation posture (if #3 isn’t “drop”)
Vectorize is a separate store — the tenant_id invariant doesn’t auto-extend. This is the answer that keeps tenants from leaking.
5 · Image attachments via Images-on-R2?
Transform straight off the existing R2 bucket — no migration, Free tier (5k/mo). The lightest-touch path.
6 · Cloudflare Stream video hosting?
Hosting recordings vs linking them. URL-ingest needs a direct MP4 — Loom/Zoom share links are HTML pages.
7 · What ships in the first commit?
Sets the branch and the first PR scope.
Any constraints, timing, or sequencing changes?
Budget windows, a deadline, a tenant rollout, or a reason to reorder — anything that should shape the plan.
Anything else?