admin panel long running imrovements, billing fixes, ui cleanup

1. Multiple registry domains + per-user domain preference The biggest feature. The appview can serve several registry domains (e.g. buoy.cr, atcr.io), and users can now pick which one shows up in their pull/push commands. - Lexicon/record: adds registryDomain (and documents ociClient) to the sailor profile (lexicons/.../profile.json, pkg/atproto/lexicon.go). - DB: new registry_domain column on users (schema.sql + migration 0027), with GetUserByDID/Handle reads, UpdateUserRegistryDomain writer, and Jetstream caching it on profile updates (writes unconditionally so clearing propagates). - UI/handlers: new UpdateRegistryDomainHandler + /api/profile/registry-domain route, a <select> in the user settings panel (only shown when >1 domain configured), and resolveRegistryURL() which falls back to the primary domain if the user's pref is stale/removed. Tests added for all of it. 2. default_hold_did removed → first managed_holds entry is the default Consolidates two overlapping config fields into one. ServerConfig.DefaultHoldDID is gone; PrimaryHoldDID() now returns managed_holds[0]. managed_holds is now REQUIRED. Updated in config, validation, server wiring, test harness, example YAML, and the deploy template. 3. Admin long-running operations → generic background-job framework New pkg/hold/admin/jobs.go introduces a reusable startJob/jobRegistry pattern (a detached context.Background() job + a /admin/api/jobs/{key}/status polling endpoint). This replaces the bespoke scan-backfill goroutine state machine, and now also wraps crew tier remap and crew import all three previously looped synchronously on the request context and got 504'd/cancelled mid-run by the reverse proxy. Forms switched from POST-redirect to htmx fragments (job_progress.html, job_result.html, crew_import_results.html) the old crew_import_results.html page and scan_backfill_progress.html partial were deleted. This is also captured as a new rule in CLAUDE.md. 4. Cascade-delete manifest on last-tag deletion DeleteTagHandler now, after removing the last tag pointing to a digest, cascade-deletes the manifest itself (PDS + DB + hold blob purge) — but only if it's not a child of a manifest list (multi-arch parent). New GetTagDigest and ShouldCascadeDeleteManifest queries back it, plus cascade_delete_test.go. Also switches tag rkey computation to the atproto.RepositoryTagToRKey helper. 5. Billing simplification Drops the OwnerBadge config option (hold-owner supporter badge). The user-profile template no longer special-cases an "owner" badge value (only "Captain"). Example tiers renamed to the nautical scheme (deckhand/bosun/quartermaster). 6. Build/deploy: go generate always runs via Make make generate is now a phony target that always runs go generate ./... (regenerating cbor_gen, icon sprites, etc.), and build-trixie depends on it. The deploy tooling (provision.go/update.go) drops its own runGenerate calls since the Makefile handles it. 7. New cmd/firehose-tap tool (untracked) A standalone CLI that subscribes to a com.atproto.sync.subscribeRepos endpoint and pretty-prints events, with emphasis on Sync 1.1 compliance fields (per-op prev CIDs, commit prevData) and a --validate CI mode. Fits with the recent "more sync1.1 compliant" commit.
2026-08-31 23:36:57 +00:00 · 2026-06-05 20:57:25 -05:00
parent efabb677e4
commit ab4a4ebf9d
43 changed files with 1768 additions and 387 deletions
@@ -198,6 +198,7 @@ See `config-appview.example.yaml` and `config-hold.example.yaml` for all options
 - **Hold stats are ATProto records in CAR store** — `io.atcr.hold.stats` records are stored via `repomgr.PutRecord()`, not in SQLite. Lost if CAR store is lost without backup.
 - **PLC auto-update on boot** — When using did:plc, `LoadOrCreateDID()` calls `EnsurePLCCurrent()` every startup. If local signing key or URL doesn't match plc.directory, it auto-updates (requires rotation key on disk).
 - **Hold CAR store is the source of truth** — Captain, crew, layer, stats, scan records, Bluesky posts, profiles are all ATProto records in the CAR store. SQLite holds only the records index and events.
+- **Admin panel long-running loops MUST use background jobs** — Any admin handler that loops over many items doing per-item PDS writes/network calls (bulk crew tier remap, crew import, scan backfill) must NOT run synchronously on the request context. The reverse proxy 504s such a request around the 10s mark and cancels `r.Context()`, aborting the loop mid-flight (`getb tx, context canceled`). Use the `startJob` helper in `pkg/hold/admin/jobs.go`: do fast validation synchronously (return a 200 fragment on failure, never a 302 — the forms are htmx-driven), then kick off the loop via `ui.startJob(key, title, resultTemplate, timeout, fn)` and render `partials/job_progress.html`. The fragment polls `/admin/api/jobs/{key}/status` until done. `fn` runs under its own detached `context.Background()` timeout, so it survives the request ending. The GC subsystem (`gc.startBackground`) is a peer implementation in the `gc` package (which must not import `admin`).

 ## Common Tasks

@@ -23,8 +23,13 @@ GENERATED_ASSETS = \
 	pkg/appview/public/js/lucide.min.js \
 	pkg/appview/licenses/spdx-licenses.json

-generate: $(GENERATED_ASSETS) ## Run go generate to download vendor assets
+generate: ## Run go generate ./... (always — regenerates cbor_gen, icon sprites, vendor assets)
+	@echo "→ Running go generate ./..."
+	go generate ./...

+# File rule: lazily download missing vendor assets for fast incremental local builds.
+# Production builds depend on the phony `generate` target instead so generated code
+# (cbor_gen.go, icon sprites, etc.) is always up to date.
 $(GENERATED_ASSETS):
 	@echo "→ Generating vendor assets and code..."
 	go generate ./...
@@ -66,7 +71,7 @@ build-oauth-helper: ## Build OAuth helper only
 # 2.43, which otherwise stamps sqrtf@GLIBC_2.43 onto cgo-linked output).
 TRIXIE_BUILDER_IMAGE ?= golang:1-trixie

-build-trixie: $(GENERATED_ASSETS) ## Build all production binaries (appview, hold, credential-helper, scanner, labeler) for linux/amd64 in a Debian 13 (glibc 2.41) container
+build-trixie: generate ## Build all production binaries (appview, hold, credential-helper, scanner, labeler) for linux/amd64 in a Debian 13 (glibc 2.41) container
 	@echo "→ Building in $(TRIXIE_BUILDER_IMAGE) for glibc 2.41 compatibility..."
 	@mkdir -p bin
 	docker run --rm \
@@ -0,0 +1,395 @@
+// firehose-tap subscribes to an ATProto com.atproto.sync.subscribeRepos endpoint
+// and pretty-prints every event it sees, with extra emphasis on Sync 1.1 fields
+// (per-op prev CIDs, commit-level prevData) so you can verify a hold or relay
+// is emitting compliant events.
+//
+// Typical use:
+//
+//	# Watch new events as you trigger pushes/deletes on production
+//	go run ./cmd/firehose-tap wss://hold01.atcr.io
+//
+//	# Replay from the beginning of the persisted buffer
+//	go run ./cmd/firehose-tap --cursor 0 wss://hold01.atcr.io
+//
+//	# CI mode: subscribe for 30s, exit non-zero if any update/delete op is
+//	# missing prev (Sync 1.1 inductive-firehose requirement).
+//	go run ./cmd/firehose-tap --validate --duration 30s wss://hold01.atcr.io
+//
+// The endpoint may be a single hold or a relay. The wss:// scheme is required.
+package main
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"net/http"
+	"net/url"
+	"os"
+	"os/signal"
+	"strings"
+	"sync/atomic"
+	"syscall"
+	"time"
+
+	comatproto "github.com/bluesky-social/indigo/api/atproto"
+	"github.com/bluesky-social/indigo/events"
+	"github.com/bluesky-social/indigo/events/schedulers/sequential"
+	"github.com/gorilla/websocket"
+)
+
+// ANSI color codes (disabled via --no-color or NO_COLOR env)
+var (
+	cRed    = "\033[31m"
+	cGreen  = "\033[32m"
+	cYellow = "\033[33m"
+	cCyan   = "\033[36m"
+	cBold   = "\033[1m"
+	cDim    = "\033[2m"
+	cReset  = "\033[0m"
+)
+
+func disableColors() {
+	cRed, cGreen, cYellow, cCyan, cBold, cDim, cReset = "", "", "", "", "", "", ""
+}
+
+type config struct {
+	endpoint  string
+	cursor    int64
+	cursorSet bool
+	didFilter string
+	validate  bool
+	duration  time.Duration
+	maxEvents int64
+	verbose   bool
+	noColor   bool
+}
+
+// violationCounts tracks Sync 1.1 compliance issues seen so far.
+type violationCounts struct {
+	missingPrev     atomic.Int64 // update/delete op without prev
+	createWithPrev  atomic.Int64 // create op with non-nil prev (spec forbids)
+	deleteWithCid   atomic.Int64 // delete op with non-nil cid (spec forbids)
+	missingPrevData atomic.Int64 // commit with since but no prevData
+}
+
+func (v *violationCounts) total() int64 {
+	return v.missingPrev.Load() + v.createWithPrev.Load() + v.deleteWithCid.Load() + v.missingPrevData.Load()
+}
+
+func main() {
+	cfg := parseFlags()
+	if cfg.noColor || os.Getenv("NO_COLOR") != "" {
+		disableColors()
+	}
+
+	ctx, cancel := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
+	defer cancel()
+	if cfg.duration > 0 {
+		var timeoutCancel context.CancelFunc
+		ctx, timeoutCancel = context.WithTimeout(ctx, cfg.duration)
+		defer timeoutCancel()
+	}
+
+	if err := run(ctx, cfg); err != nil && !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+		fmt.Fprintf(os.Stderr, "%sfirehose-tap: %v%s\n", cRed, err, cReset)
+		os.Exit(2)
+	}
+}
+
+func parseFlags() *config {
+	cfg := &config{}
+	flag.Int64Var(&cfg.cursor, "cursor", -1, "subscribeRepos cursor (-1 = new events only, 0 = replay from start)")
+	flag.StringVar(&cfg.didFilter, "did", "", "only show events for this DID (case-sensitive match on commit.repo)")
+	flag.BoolVar(&cfg.validate, "validate", false, "exit non-zero if any Sync 1.1 violation is observed")
+	flag.DurationVar(&cfg.duration, "duration", 0, "exit cleanly after this duration (e.g. 30s)")
+	flag.Int64Var(&cfg.maxEvents, "max-events", 0, "exit cleanly after seeing this many events (0 = unlimited)")
+	flag.BoolVar(&cfg.verbose, "verbose", false, "print full ops for create/no-prev-needed ops too")
+	flag.BoolVar(&cfg.noColor, "no-color", false, "disable ANSI colors")
+
+	// Detect whether --cursor was explicitly set so we can omit it from the URL
+	// when the user didn't pass one (default -1 means "live, no backfill").
+	flag.Usage = func() {
+		fmt.Fprintf(os.Stderr, "Usage: %s [flags] <subscribeRepos-endpoint>\n\n", os.Args[0])
+		fmt.Fprintln(os.Stderr, "Endpoint examples:")
+		fmt.Fprintln(os.Stderr, "  wss://hold01.atcr.io")
+		fmt.Fprintln(os.Stderr, "  wss://relay1.us-east.bsky.network")
+		fmt.Fprintln(os.Stderr, "  ws://localhost:8080")
+		fmt.Fprintln(os.Stderr)
+		fmt.Fprintln(os.Stderr, "Flags:")
+		flag.PrintDefaults()
+	}
+	flag.Parse()
+
+	flag.Visit(func(f *flag.Flag) {
+		if f.Name == "cursor" {
+			cfg.cursorSet = true
+		}
+	})
+
+	args := flag.Args()
+	if len(args) != 1 {
+		flag.Usage()
+		os.Exit(2)
+	}
+	cfg.endpoint = args[0]
+	return cfg
+}
+
+func run(ctx context.Context, cfg *config) error {
+	wsURL, err := buildWSURL(cfg.endpoint, cfg.cursor, cfg.cursorSet)
+	if err != nil {
+		return err
+	}
+
+	fmt.Fprintf(os.Stderr, "%sfirehose-tap%s connecting to %s%s%s\n", cBold, cReset, cCyan, wsURL, cReset)
+	if cfg.validate {
+		fmt.Fprintf(os.Stderr, "%svalidate mode: process will exit non-zero on Sync 1.1 violations%s\n", cDim, cReset)
+	}
+
+	header := http.Header{}
+	header.Set("User-Agent", "atcr-firehose-tap/1.0")
+
+	dialer := websocket.DefaultDialer
+	dialer.HandshakeTimeout = 30 * time.Second
+	conn, resp, err := dialer.DialContext(ctx, wsURL, header)
+	if err != nil {
+		if resp != nil {
+			return fmt.Errorf("dial %s: %w (HTTP %d)", wsURL, err, resp.StatusCode)
+		}
+		return fmt.Errorf("dial %s: %w", wsURL, err)
+	}
+	defer conn.Close()
+
+	v := &violationCounts{}
+	var eventsSeen atomic.Int64
+
+	callbacks := &events.RepoStreamCallbacks{
+		RepoCommit: func(evt *comatproto.SyncSubscribeRepos_Commit) error {
+			n := eventsSeen.Add(1)
+			if cfg.didFilter == "" || cfg.didFilter == evt.Repo {
+				printCommit(evt, v, cfg)
+			}
+			if cfg.maxEvents > 0 && n >= cfg.maxEvents {
+				return errMaxEvents
+			}
+			return nil
+		},
+		RepoIdentity: func(evt *comatproto.SyncSubscribeRepos_Identity) error {
+			eventsSeen.Add(1)
+			if cfg.didFilter == "" || cfg.didFilter == evt.Did {
+				printIdentity(evt)
+			}
+			return nil
+		},
+		RepoAccount: func(evt *comatproto.SyncSubscribeRepos_Account) error {
+			eventsSeen.Add(1)
+			if cfg.didFilter == "" || cfg.didFilter == evt.Did {
+				printAccount(evt)
+			}
+			return nil
+		},
+		RepoSync: func(evt *comatproto.SyncSubscribeRepos_Sync) error {
+			eventsSeen.Add(1)
+			if cfg.didFilter == "" || cfg.didFilter == evt.Did {
+				printSync(evt)
+			}
+			return nil
+		},
+		RepoInfo: func(evt *comatproto.SyncSubscribeRepos_Info) error {
+			fmt.Printf("%s#info%s name=%s message=%s\n", cYellow, cReset, evt.Name, derefStr(evt.Message))
+			return nil
+		},
+		Error: func(evt *events.ErrorFrame) error {
+			fmt.Printf("%s#error%s %s: %s\n", cRed, cReset, evt.Error, evt.Message)
+			return nil
+		},
+	}
+
+	sched := sequential.NewScheduler("firehose-tap", callbacks.EventHandler)
+	streamErr := events.HandleRepoStream(ctx, conn, sched, nil)
+	if errors.Is(streamErr, errMaxEvents) {
+		streamErr = nil
+	}
+
+	printSummary(eventsSeen.Load(), v, cfg)
+
+	if cfg.validate && v.total() > 0 {
+		os.Exit(1)
+	}
+	return streamErr
+}
+
+// errMaxEvents is returned from the commit handler to short-circuit
+// HandleRepoStream once --max-events has been reached.
+var errMaxEvents = errors.New("max events reached")
+
+func buildWSURL(endpoint string, cursor int64, cursorSet bool) (string, error) {
+	if !strings.Contains(endpoint, "://") {
+		endpoint = "wss://" + endpoint
+	}
+	u, err := url.Parse(endpoint)
+	if err != nil {
+		return "", fmt.Errorf("parse endpoint: %w", err)
+	}
+	switch u.Scheme {
+	case "wss", "ws":
+		// already a websocket URL
+	case "https":
+		u.Scheme = "wss"
+	case "http":
+		u.Scheme = "ws"
+	default:
+		return "", fmt.Errorf("unsupported scheme %q (use ws:// or wss://)", u.Scheme)
+	}
+
+	if !strings.Contains(u.Path, "subscribeRepos") {
+		u.Path = strings.TrimRight(u.Path, "/") + "/xrpc/com.atproto.sync.subscribeRepos"
+	}
+
+	if cursorSet {
+		q := u.Query()
+		q.Set("cursor", fmt.Sprintf("%d", cursor))
+		u.RawQuery = q.Encode()
+	}
+	return u.String(), nil
+}
+
+func printCommit(evt *comatproto.SyncSubscribeRepos_Commit, v *violationCounts, cfg *config) {
+	prevData := "(none)"
+	if evt.PrevData != nil {
+		prevData = shortCID(evt.PrevData.String())
+	}
+	since := "(none)"
+	if evt.Since != nil {
+		since = *evt.Since
+	}
+
+	// Spec: commits beyond the very first one should carry prevData.
+	if evt.Since != nil && evt.PrevData == nil {
+		v.missingPrevData.Add(1)
+	}
+
+	fmt.Printf("%s#commit%s seq=%d repo=%s rev=%s since=%s prevData=%s ops=%d\n",
+		cGreen, cReset, evt.Seq, evt.Repo, evt.Rev, since, prevData, len(evt.Ops))
+
+	for _, op := range evt.Ops {
+		printOp(op, v, cfg)
+	}
+}
+
+func printOp(op *comatproto.SyncSubscribeRepos_RepoOp, v *violationCounts, cfg *config) {
+	action := op.Action
+
+	cidStr := "(nil)"
+	if op.Cid != nil {
+		cidStr = shortCID(op.Cid.String())
+	}
+	prevStr := "(nil)"
+	if op.Prev != nil {
+		prevStr = shortCID(op.Prev.String())
+	}
+
+	violated := false
+	var noteParts []string
+
+	switch action {
+	case "update", "delete":
+		if op.Prev == nil {
+			v.missingPrev.Add(1)
+			violated = true
+			noteParts = append(noteParts, "MISSING prev (Sync 1.1)")
+		}
+	case "create":
+		if op.Prev != nil {
+			v.createWithPrev.Add(1)
+			violated = true
+			noteParts = append(noteParts, "create MUST NOT have prev")
+		}
+	}
+	if action == "delete" && op.Cid != nil {
+		v.deleteWithCid.Add(1)
+		violated = true
+		noteParts = append(noteParts, "delete MUST have nil cid")
+	}
+
+	// In non-verbose mode, only print ops that updated/deleted (the interesting
+	// Sync 1.1 cases) or that violated a rule.
+	if !cfg.verbose && action == "create" && !violated {
+		return
+	}
+
+	actionColor := cCyan
+	if violated {
+		actionColor = cRed
+	}
+	note := ""
+	if len(noteParts) > 0 {
+		note = fmt.Sprintf(" %s[%s]%s", cRed, strings.Join(noteParts, "; "), cReset)
+	}
+	fmt.Printf("    %s%-6s%s %s cid=%s prev=%s%s\n",
+		actionColor, action, cReset, op.Path, cidStr, prevStr, note)
+}
+
+func printIdentity(evt *comatproto.SyncSubscribeRepos_Identity) {
+	handle := "(nil)"
+	if evt.Handle != nil {
+		handle = *evt.Handle
+	}
+	fmt.Printf("%s#identity%s seq=%d did=%s handle=%s\n",
+		cYellow, cReset, evt.Seq, evt.Did, handle)
+}
+
+func printAccount(evt *comatproto.SyncSubscribeRepos_Account) {
+	status := "(nil)"
+	if evt.Status != nil {
+		status = *evt.Status
+	}
+	fmt.Printf("%s#account%s seq=%d did=%s active=%v status=%s\n",
+		cYellow, cReset, evt.Seq, evt.Did, evt.Active, status)
+}
+
+func printSync(evt *comatproto.SyncSubscribeRepos_Sync) {
+	fmt.Printf("%s#sync%s seq=%d did=%s rev=%s\n",
+		cYellow, cReset, evt.Seq, evt.Did, evt.Rev)
+}
+
+func printSummary(seen int64, v *violationCounts, cfg *config) {
+	fmt.Fprintln(os.Stderr)
+	fmt.Fprintf(os.Stderr, "%s--- summary ---%s\n", cBold, cReset)
+	fmt.Fprintf(os.Stderr, "events seen: %d\n", seen)
+	if v.total() == 0 {
+		fmt.Fprintf(os.Stderr, "%sno Sync 1.1 violations%s\n", cGreen, cReset)
+		return
+	}
+	fmt.Fprintf(os.Stderr, "%sviolations:%s\n", cRed, cReset)
+	if n := v.missingPrev.Load(); n > 0 {
+		fmt.Fprintf(os.Stderr, "  %d update/delete op(s) missing prev\n", n)
+	}
+	if n := v.createWithPrev.Load(); n > 0 {
+		fmt.Fprintf(os.Stderr, "  %d create op(s) with non-nil prev\n", n)
+	}
+	if n := v.deleteWithCid.Load(); n > 0 {
+		fmt.Fprintf(os.Stderr, "  %d delete op(s) with non-nil cid\n", n)
+	}
+	if n := v.missingPrevData.Load(); n > 0 {
+		fmt.Fprintf(os.Stderr, "  %d commit(s) with since but missing prevData\n", n)
+	}
+	if !cfg.validate {
+		fmt.Fprintf(os.Stderr, "%s(--validate not set, exiting 0 anyway)%s\n", cDim, cReset)
+	}
+}
+
+func shortCID(s string) string {
+	if len(s) <= 16 {
+		return s
+	}
+	return s[:8] + "…" + s[len(s)-6:]
+}
+
+func derefStr(s *string) string {
+	if s == nil {
+		return ""
+	}
+	return *s
+}
@@ -25,8 +25,6 @@ server:
    addr: :5000
    # Public-facing URL for OAuth callbacks and JWT realm. Auto-detected if empty.
    base_url: ""
-    # DID of the hold service for blob storage, e.g. "did:web:hold01.atcr.io" (REQUIRED).
-    default_hold_did: ""
    # Allows HTTP (not HTTPS) for DID resolution and uses transition:generic OAuth scope.
    test_mode: false
    # Display name shown on OAuth authorization screens.
@@ -34,10 +32,10 @@ server:
    # Short name used in page titles and browser tabs.
    client_short_name: ATCR
    # Separate domains for OCI registry API (e.g. ["buoy.cr"]). First is primary. Browser visits redirect to BaseURL.
-    registry_domains: []
-    # DIDs of holds this appview manages billing for. Tier updates are pushed to these holds.
+    registry_domains: [127.0.0.1:5000, atcr.io]
+    # DIDs of holds this appview manages billing for (REQUIRED). The first entry is the default blob-storage hold. Tier updates are pushed to these holds.
    managed_holds:
-      - did:web:172.28.0.3%3A8080 
+      - did:web:172.28.0.3%3A8080
 # Web UI settings.
 ui:
    # SQLite/libSQL database for OAuth sessions, stars, pull counts, and device approvals.
@@ -164,5 +162,3 @@ billing:
          ai_advisor: true
          # Show supporter badge on user profiles for subscribers at this tier.
          supporter_badge: true
-    # Show supporter badge on hold owner profiles.
-    owner_badge: true
@@ -10,7 +10,6 @@ log_shipper:
 server:
    addr: :5000
    base_url: "https://seamark.dev"
-    default_hold_did: "{{.HoldDid}}"
    client_name: Seamark
    test_mode: false
    client_short_name: Seamark
@@ -18,6 +17,8 @@ server:
      - "buoy.cr"
      - "bouy.cr"
      - "seamark.cr"
+    managed_holds:
+      - "{{.HoldDid}}"
 ui:
    database_path: "{{.BasePath}}/ui.db"
    theme: seamark
@@ -367,10 +367,6 @@ func cmdProvision(token, zone, plan, sshKeyPath, s3Secret string, withScanner, w
 	if appviewCreated || holdCreated {
 		rootDir := projectRoot()

-		if err := runGenerate(rootDir); err != nil {
-			return fmt.Errorf("go generate: %w", err)
-		}
-
 		if err := runMakeBuildTrixie(rootDir); err != nil {
 			return fmt.Errorf("build: %w", err)
 		}
@@ -422,9 +418,6 @@ func cmdProvision(token, zone, plan, sshKeyPath, s3Secret string, withScanner, w
 	// not freshly created (the appviewCreated branch above already handled it).
 	if state.LabelerEnabled && !appviewCreated {
 		rootDir := projectRoot()
-		if err := runGenerate(rootDir); err != nil {
-			return fmt.Errorf("go generate: %w", err)
-		}
 		labelerLocal := filepath.Join(rootDir, "bin", "atcr-labeler")
 		if err := runMakeBuildTrixie(rootDir); err != nil {
 			return fmt.Errorf("build labeler: %w", err)
@@ -122,13 +122,10 @@ func cmdUpdate(target string, withScanner, withLabeler bool) error {
 		return fmt.Errorf("unknown target: %s (use: all, appview, hold)", target)
 	}

-	// Run go generate before building
-	if err := runGenerate(rootDir); err != nil {
-		return fmt.Errorf("go generate: %w", err)
-	}
-
 	// Build all binaries via `make build-trixie` so output links against
 	// glibc 2.41 (the deploy target's glibc) regardless of the host's glibc.
+	// build-trixie depends on the `generate` make target, which runs
+	// `go generate ./...` on the host before invoking the trixie container.
 	if err := runMakeBuildTrixie(rootDir); err != nil {
 		return fmt.Errorf("build: %w", err)
 	}
@@ -347,17 +344,6 @@ func configValsFromState(state *InfraState) *ConfigValues {
 	}
 }

-// runGenerate runs go generate ./... in the given directory using host OS/arch
-// (no cross-compilation env vars — generate tools must run on the build machine).
-func runGenerate(dir string) error {
-	fmt.Println("Running go generate ./...")
-	cmd := exec.Command("go", "generate", "./...")
-	cmd.Dir = dir
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	return cmd.Run()
-}
-
 // runMakeBuildTrixie shells out to `make build-trixie`, which builds all
 // production binaries (appview, hold, credential-helper, labeler, scanner)
 // inside a Debian 13 container so they link against glibc 2.41. Centralizing
@@ -17,8 +17,9 @@ services:
    environment:
      # ATCR_SERVER_CLIENT_NAME: "Seamark"
      # ATCR_SERVER_CLIENT_SHORT_NAME: "Seamark"
+      # First entry is the default blob-storage hold. Comma-separate for multiple:
+      # ATCR_SERVER_MANAGED_HOLDS: "did:web:a,did:web:b" (Viper splits on commas).
      ATCR_SERVER_MANAGED_HOLDS: did:web:172.28.0.3%3A8080
-      ATCR_SERVER_DEFAULT_HOLD_DID: did:web:172.28.0.3%3A8080
      # Labeler URL (HTTP for dev — ParseLabelerURL accepts it directly so we don't
      # have to round-trip through did:web → https:// resolution).
      ATCR_LABELER_DID: did:web:172.28.0.4%3A5002
@@ -412,7 +412,7 @@ func buildAppViewConfig(addr, baseURL, holdDID, dbPath string) *appview.Config {
 	cfg.LogLevel = "warn"
 	cfg.Server.Addr = addr
 	cfg.Server.BaseURL = baseURL
-	cfg.Server.DefaultHoldDID = holdDID
+	cfg.Server.ManagedHolds = []string{holdDID}
 	cfg.Server.TestMode = true
 	// Registry domain is a bare hostname (no port). DomainRoutingMiddleware
 	// strips ports before matching, so "127.0.0.1" is what /v2/ requests
@@ -19,6 +19,17 @@
            "type": "boolean",
            "description": "Automatically delete manifest records that become untagged after a tag overwrite. Layers are cleaned up by hold garbage collection."
          },
+          "ociClient": {
+            "type": "string",
+            "maxLength": 32,
+            "knownValues": ["docker", "podman", "buildah", "nerdctl", "crane", "none"],
+            "description": "Preferred client for pull commands (docker, podman, buildah, nerdctl, crane). 'none' shows the image reference only. Defaults to docker if empty."
+          },
+          "registryDomain": {
+            "type": "string",
+            "maxLength": 255,
+            "description": "Preferred registry domain for UI display. Must be one of the appview's configured registry domains. Empty means the primary (first configured) domain."
+          },
          "createdAt": {
            "type": "string",
            "format": "datetime",
@@ -47,9 +47,6 @@ type ServerConfig struct {
 	// Public-facing URL for OAuth callbacks and JWT realm.
 	BaseURL string `yaml:"base_url" comment:"Public-facing URL for OAuth callbacks and JWT realm. Auto-detected if empty."`

-	// DID of the default hold service for blob storage.
-	DefaultHoldDID string `yaml:"default_hold_did" comment:"DID of the hold service for blob storage, e.g. \"did:web:hold01.atcr.io\" (REQUIRED)."`
-
 	// Allows HTTP (not HTTPS) for DID resolution.
 	TestMode bool `yaml:"test_mode" comment:"Allows HTTP (not HTTPS) for DID resolution and uses transition:generic OAuth scope."`

@@ -62,8 +59,18 @@ type ServerConfig struct {
 	// Separate domains for OCI registry API. First entry is the primary (used for JWT service name and UI display).
 	RegistryDomains []string `yaml:"registry_domains" comment:"Separate domains for OCI registry API (e.g. [\"buoy.cr\"]). First is primary. Browser visits redirect to BaseURL."`

-	// DIDs of holds this appview manages billing for.
-	ManagedHolds []string `yaml:"managed_holds" comment:"DIDs of holds this appview manages billing for. Tier updates are pushed to these holds."`
+	// DIDs of holds this appview manages billing for. The first entry is also
+	// the default blob-storage hold (used when a user has no hold selected).
+	ManagedHolds []string `yaml:"managed_holds" comment:"DIDs of holds this appview manages billing for (REQUIRED). The first entry is the default blob-storage hold. Tier updates are pushed to these holds."`
+}
+
+// PrimaryHoldDID returns the appview's default blob-storage hold, which is the
+// first managed hold. Empty if none configured.
+func (s ServerConfig) PrimaryHoldDID() string {
+	if len(s.ManagedHolds) > 0 {
+		return s.ManagedHolds[0]
+	}
+	return ""
 }

 // UIConfig defines web UI settings
@@ -163,7 +170,6 @@ func setDefaults(v *viper.Viper) {
 	// Server defaults
 	v.SetDefault("server.addr", ":5000")
 	v.SetDefault("server.base_url", "")
-	v.SetDefault("server.default_hold_did", "")
 	v.SetDefault("server.test_mode", false)
 	v.SetDefault("server.client_name", "AT Container Registry")
 	v.SetDefault("server.client_short_name", "ATCR")
@@ -235,7 +241,6 @@ func ExampleYAML() ([]byte, error) {
 	cfg.Billing.Currency = "usd"
 	cfg.Billing.SuccessURL = "{base_url}/settings/billing"
 	cfg.Billing.CancelURL = "{base_url}/settings/billing"
-	cfg.Billing.OwnerBadge = true
 	cfg.Billing.Tiers = []billing.BillingTierConfig{
 		{Name: "deckhand", Description: "Get started with basic storage", MaxWebhooks: 1},
 		{Name: "bosun", Description: "More storage with scan-on-push", StripePriceMonthly: "price_xxx", StripePriceYearly: "price_yyy", MaxWebhooks: 5, WebhookAllTriggers: true, SupporterBadge: true},
@@ -281,8 +286,8 @@ func LoadConfig(yamlPath string) (*Config, error) {
 	}

 	// Validation
-	if cfg.Server.DefaultHoldDID == "" {
-		return nil, fmt.Errorf("server.default_hold_did is required (env: ATCR_SERVER_DEFAULT_HOLD_DID)")
+	if len(cfg.Server.ManagedHolds) == 0 {
+		return nil, fmt.Errorf("server.managed_holds is required (at least one hold DID; env: ATCR_SERVER_MANAGED_HOLDS)")
 	}
 	if cfg.Labeler.DID != "" && !strings.HasPrefix(cfg.Labeler.DID, "did:") {
 		return nil, fmt.Errorf("labeler.did must be a DID (did:plc:... or did:web:...), got %q", cfg.Labeler.DID)
@@ -350,7 +355,7 @@ func buildDistributionConfig(cfg *Config, v *viper.Viper) (*configuration.Config
 	distConfig.Storage = buildStorageConfig()

 	// Middleware (ATProto resolver)
-	distConfig.Middleware = buildMiddlewareConfig(cfg.Server.DefaultHoldDID, cfg.Server.BaseURL, cfg.Server.TestMode)
+	distConfig.Middleware = buildMiddlewareConfig(cfg.Server.PrimaryHoldDID(), cfg.Server.BaseURL, cfg.Server.TestMode)

 	// Auth (use values from cfg.Auth)
 	// Realm always points to BaseURL where auth endpoints live
@@ -176,9 +176,9 @@ func TestLoadConfig(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			if tt.setHoldDID {
-				t.Setenv("ATCR_SERVER_DEFAULT_HOLD_DID", tt.envHoldDID)
+				t.Setenv("ATCR_SERVER_MANAGED_HOLDS", tt.envHoldDID)
 			} else {
-				os.Unsetenv("ATCR_SERVER_DEFAULT_HOLD_DID")
+				os.Unsetenv("ATCR_SERVER_MANAGED_HOLDS")
 			}

 			// Clear other env vars to use defaults
@@ -207,8 +207,8 @@ func TestLoadConfig(t *testing.T) {
 				t.Errorf("HTTP addr = %v, want :5000", got.Server.Addr)
 			}

-			if got.Server.DefaultHoldDID != tt.envHoldDID {
-				t.Errorf("default hold DID = %v, want %v", got.Server.DefaultHoldDID, tt.envHoldDID)
+			if got.Server.PrimaryHoldDID() != tt.envHoldDID {
+				t.Errorf("primary hold DID = %v, want %v", got.Server.PrimaryHoldDID(), tt.envHoldDID)
 			}

 			if got.UI.DatabasePath != "/var/lib/atcr/ui.db" {
@@ -0,0 +1,199 @@
+package db
+
+import (
+	"database/sql"
+	"errors"
+	"testing"
+	"time"
+)
+
+// seedCascadeFixture inserts a user and a single manifest. Returns the
+// manifest's row id so callers can attach references (for the multi-arch case).
+func seedCascadeFixture(t *testing.T, db *sql.DB, didStr, repo, digest string) int64 {
+	t.Helper()
+
+	user := &User{
+		DID:         didStr,
+		Handle:      "tester.example.com",
+		PDSEndpoint: "https://test.pds.example.com",
+		LastSeen:    time.Now(),
+	}
+	if err := UpsertUser(db, user); err != nil {
+		t.Fatalf("UpsertUser: %v", err)
+	}
+
+	id, err := InsertManifest(db, &Manifest{
+		DID:           didStr,
+		Repository:    repo,
+		Digest:        digest,
+		HoldEndpoint:  "did:web:hold.example.com",
+		SchemaVersion: 2,
+		MediaType:     "application/vnd.oci.image.manifest.v1+json",
+		CreatedAt:     time.Now(),
+	})
+	if err != nil {
+		t.Fatalf("InsertManifest: %v", err)
+	}
+	return id
+}
+
+func TestGetTagDigest_ReturnsDigestForKnownTag(t *testing.T) {
+	db, err := InitDB(":memory:", LibsqlConfig{})
+	if err != nil {
+		t.Fatalf("InitDB: %v", err)
+	}
+	defer db.Close()
+
+	const did = "did:plc:tagdigest"
+	const repo = "myapp"
+	const digest = "sha256:aaa"
+
+	seedCascadeFixture(t, db, did, repo, digest)
+
+	if err := UpsertTag(db, &Tag{
+		DID:        did,
+		Repository: repo,
+		Tag:        "latest",
+		Digest:     digest,
+		CreatedAt:  time.Now(),
+	}); err != nil {
+		t.Fatalf("UpsertTag: %v", err)
+	}
+
+	got, err := GetTagDigest(db, did, repo, "latest")
+	if err != nil {
+		t.Fatalf("GetTagDigest: %v", err)
+	}
+	if got != digest {
+		t.Errorf("digest mismatch: got %q want %q", got, digest)
+	}
+}
+
+func TestGetTagDigest_UnknownTagReturnsErrNoRows(t *testing.T) {
+	db, err := InitDB(":memory:", LibsqlConfig{})
+	if err != nil {
+		t.Fatalf("InitDB: %v", err)
+	}
+	defer db.Close()
+
+	seedCascadeFixture(t, db, "did:plc:tagdigest2", "myapp", "sha256:bbb")
+
+	_, err = GetTagDigest(db, "did:plc:tagdigest2", "myapp", "does-not-exist")
+	if !errors.Is(err, sql.ErrNoRows) {
+		t.Errorf("expected sql.ErrNoRows for unknown tag, got %v", err)
+	}
+}
+
+// TestShouldCascadeDeleteManifest_LastTagAndNoParent: the common case —
+// digest has no remaining tags and is not referenced by any manifest list.
+// Cascade should fire.
+func TestShouldCascadeDeleteManifest_LastTagAndNoParent(t *testing.T) {
+	db, err := InitDB(":memory:", LibsqlConfig{})
+	if err != nil {
+		t.Fatalf("InitDB: %v", err)
+	}
+	defer db.Close()
+
+	const did = "did:plc:cascade1"
+	const repo = "myapp"
+	const digest = "sha256:lonely"
+
+	seedCascadeFixture(t, db, did, repo, digest)
+
+	// No tags pointing to this digest, no manifest_references entries.
+	ok, err := ShouldCascadeDeleteManifest(db, did, repo, digest)
+	if err != nil {
+		t.Fatalf("ShouldCascadeDeleteManifest: %v", err)
+	}
+	if !ok {
+		t.Error("expected cascade=true when manifest is untagged and unreferenced")
+	}
+}
+
+// TestShouldCascadeDeleteManifest_RemainingTagBlocks: another tag still
+// points to this digest → keep the manifest alive.
+func TestShouldCascadeDeleteManifest_RemainingTagBlocks(t *testing.T) {
+	db, err := InitDB(":memory:", LibsqlConfig{})
+	if err != nil {
+		t.Fatalf("InitDB: %v", err)
+	}
+	defer db.Close()
+
+	const did = "did:plc:cascade2"
+	const repo = "myapp"
+	const digest = "sha256:shared"
+
+	seedCascadeFixture(t, db, did, repo, digest)
+
+	if err := UpsertTag(db, &Tag{
+		DID:        did,
+		Repository: repo,
+		Tag:        "v1",
+		Digest:     digest,
+		CreatedAt:  time.Now(),
+	}); err != nil {
+		t.Fatalf("UpsertTag: %v", err)
+	}
+
+	ok, err := ShouldCascadeDeleteManifest(db, did, repo, digest)
+	if err != nil {
+		t.Fatalf("ShouldCascadeDeleteManifest: %v", err)
+	}
+	if ok {
+		t.Error("expected cascade=false when another tag still points to the digest")
+	}
+}
+
+// TestShouldCascadeDeleteManifest_MultiArchChildBlocks: the digest is a child
+// of a manifest list (multi-arch parent). Even with no tags, deleting it
+// would orphan the parent's reference, so we must NOT cascade.
+func TestShouldCascadeDeleteManifest_MultiArchChildBlocks(t *testing.T) {
+	db, err := InitDB(":memory:", LibsqlConfig{})
+	if err != nil {
+		t.Fatalf("InitDB: %v", err)
+	}
+	defer db.Close()
+
+	const did = "did:plc:cascade3"
+	const repo = "myapp"
+	const childDigest = "sha256:amd64child"
+	const parentDigest = "sha256:multiarchparent"
+
+	// Insert the child manifest fixture.
+	seedCascadeFixture(t, db, did, repo, childDigest)
+
+	// Insert a separate parent (manifest list) and attach a manifest_reference
+	// from parent → child.
+	parentID, err := InsertManifest(db, &Manifest{
+		DID:           did,
+		Repository:    repo,
+		Digest:        parentDigest,
+		HoldEndpoint:  "did:web:hold.example.com",
+		SchemaVersion: 2,
+		MediaType:     "application/vnd.oci.image.index.v1+json",
+		CreatedAt:     time.Now(),
+	})
+	if err != nil {
+		t.Fatalf("InsertManifest(parent): %v", err)
+	}
+
+	if err := InsertManifestReference(db, &ManifestReference{
+		ManifestID:           parentID,
+		Digest:               childDigest,
+		Size:                 1234,
+		MediaType:            "application/vnd.oci.image.manifest.v1+json",
+		PlatformArchitecture: "amd64",
+		PlatformOS:           "linux",
+		ReferenceIndex:       0,
+	}); err != nil {
+		t.Fatalf("InsertManifestReference: %v", err)
+	}
+
+	ok, err := ShouldCascadeDeleteManifest(db, did, repo, childDigest)
+	if err != nil {
+		t.Fatalf("ShouldCascadeDeleteManifest: %v", err)
+	}
+	if ok {
+		t.Error("expected cascade=false when digest is a child of a manifest list")
+	}
+}
@@ -0,0 +1,3 @@
+description: Add registry_domain column to users table for preferred registry domain
+query: |
+  ALTER TABLE users ADD COLUMN registry_domain TEXT DEFAULT '';
@@ -10,6 +10,7 @@ type User struct {
 	Avatar         string
 	DefaultHoldDID string
 	OciClient      string
+	RegistryDomain string
 	LastSeen       time.Time
 }

@@ -442,12 +442,12 @@ func GetRepositoryMetadata(db DBTX, did string, repository string) (map[string]s
 // GetUserByDID retrieves a user by DID
 func GetUserByDID(db DBTX, did string) (*User, error) {
 	var user User
-	var avatar, defaultHoldDID, ociClient sql.NullString
+	var avatar, defaultHoldDID, ociClient, registryDomain sql.NullString
 	err := db.QueryRow(`
-        SELECT did, handle, pds_endpoint, avatar, default_hold_did, oci_client, last_seen
+        SELECT did, handle, pds_endpoint, avatar, default_hold_did, oci_client, registry_domain, last_seen
        FROM users
        WHERE did = ?
-    `, did).Scan(&user.DID, &user.Handle, &user.PDSEndpoint, &avatar, &defaultHoldDID, &ociClient, &user.LastSeen)
+    `, did).Scan(&user.DID, &user.Handle, &user.PDSEndpoint, &avatar, &defaultHoldDID, &ociClient, &registryDomain, &user.LastSeen)

 	if err == sql.ErrNoRows {
 		return nil, nil
@@ -465,6 +465,9 @@ func GetUserByDID(db DBTX, did string) (*User, error) {
 	if ociClient.Valid {
 		user.OciClient = ociClient.String
 	}
+	if registryDomain.Valid {
+		user.RegistryDomain = registryDomain.String
+	}

 	return &user, nil
 }
@@ -472,12 +475,12 @@ func GetUserByDID(db DBTX, did string) (*User, error) {
 // GetUserByHandle retrieves a user by handle
 func GetUserByHandle(db DBTX, handle string) (*User, error) {
 	var user User
-	var avatar, defaultHoldDID, ociClient sql.NullString
+	var avatar, defaultHoldDID, ociClient, registryDomain sql.NullString
 	err := db.QueryRow(`
-        SELECT did, handle, pds_endpoint, avatar, default_hold_did, oci_client, last_seen
+        SELECT did, handle, pds_endpoint, avatar, default_hold_did, oci_client, registry_domain, last_seen
        FROM users
        WHERE handle = ?
-    `, handle).Scan(&user.DID, &user.Handle, &user.PDSEndpoint, &avatar, &defaultHoldDID, &ociClient, &user.LastSeen)
+    `, handle).Scan(&user.DID, &user.Handle, &user.PDSEndpoint, &avatar, &defaultHoldDID, &ociClient, &registryDomain, &user.LastSeen)

 	if err == sql.ErrNoRows {
 		return nil, nil
@@ -495,6 +498,9 @@ func GetUserByHandle(db DBTX, handle string) (*User, error) {
 	if ociClient.Valid {
 		user.OciClient = ociClient.String
 	}
+	if registryDomain.Valid {
+		user.RegistryDomain = registryDomain.String
+	}

 	return &user, nil
 }
@@ -591,6 +597,15 @@ func UpdateUserOciClient(db DBTX, did string, ociClient string) error {
 	return err
 }

+// UpdateUserRegistryDomain updates a user's cached preferred registry domain.
+// An empty string means "use the primary (first configured) domain".
+func UpdateUserRegistryDomain(db DBTX, did string, registryDomain string) error {
+	_, err := db.Exec(`
+        UPDATE users SET registry_domain = ? WHERE did = ?
+    `, registryDomain, did)
+	return err
+}
+
 // GetUserHoldDID returns the hold DID for a user. Uses cached default_hold_did
 // if available, otherwise falls back to the most recent manifest's hold_endpoint.
 func GetUserHoldDID(db DBTX, did string) string {
@@ -1541,6 +1556,37 @@ func IsManifestReferenced(db DBTX, did, digest string) (bool, error) {
 	return count > 0, nil
 }

+// GetTagDigest returns the manifest digest that a tag points to.
+// Returns sql.ErrNoRows if the tag does not exist.
+func GetTagDigest(db DBTX, did, repository, tag string) (string, error) {
+	var digest string
+	err := db.QueryRow(`
+        SELECT digest FROM tags
+        WHERE did = ? AND repository = ? AND tag = ?
+    `, did, repository, tag).Scan(&digest)
+	return digest, err
+}
+
+// ShouldCascadeDeleteManifest returns true iff a manifest can be safely
+// deleted after a tag is removed: it has no remaining tags AND is not
+// referenced by any manifest list (multi-arch parent). Manifest-list
+// children must be preserved even when untagged, since their parent index
+// still depends on them.
+func ShouldCascadeDeleteManifest(db DBTX, did, repository, digest string) (bool, error) {
+	tagged, err := IsManifestTagged(db, did, repository, digest)
+	if err != nil {
+		return false, err
+	}
+	if tagged {
+		return false, nil
+	}
+	referenced, err := IsManifestReferenced(db, did, digest)
+	if err != nil {
+		return false, err
+	}
+	return !referenced, nil
+}
+
 // IsManifestTagged checks if a manifest has any tags
 func IsManifestTagged(db DBTX, did, repository, digest string) (bool, error) {
 	var count int
@@ -987,6 +987,64 @@ func TestGetTagsWithPlatforms(t *testing.T) {
 	_ = manifestID1
 }

+func TestUserRegistryDomain(t *testing.T) {
+	db, err := InitDB(":memory:", LibsqlConfig{})
+	if err != nil {
+		t.Fatalf("Failed to init database: %v", err)
+	}
+	defer db.Close()
+
+	user := &User{
+		DID:         "did:plc:domainpref",
+		Handle:      "pref.bsky.social",
+		PDSEndpoint: "https://bsky.social",
+		LastSeen:    time.Now(),
+	}
+	if err := UpsertUser(db, user); err != nil {
+		t.Fatalf("Failed to upsert user: %v", err)
+	}
+
+	// Default is empty string (NULL-safe scan).
+	got, err := GetUserByDID(db, user.DID)
+	if err != nil {
+		t.Fatalf("GetUserByDID failed: %v", err)
+	}
+	if got.RegistryDomain != "" {
+		t.Errorf("Expected empty registry domain by default, got %q", got.RegistryDomain)
+	}
+
+	// Update and confirm it round-trips through both lookups.
+	if err := UpdateUserRegistryDomain(db, user.DID, "buoy.cr"); err != nil {
+		t.Fatalf("UpdateUserRegistryDomain failed: %v", err)
+	}
+	byDID, err := GetUserByDID(db, user.DID)
+	if err != nil {
+		t.Fatalf("GetUserByDID failed: %v", err)
+	}
+	if byDID.RegistryDomain != "buoy.cr" {
+		t.Errorf("GetUserByDID: expected 'buoy.cr', got %q", byDID.RegistryDomain)
+	}
+	byHandle, err := GetUserByHandle(db, user.Handle)
+	if err != nil {
+		t.Fatalf("GetUserByHandle failed: %v", err)
+	}
+	if byHandle.RegistryDomain != "buoy.cr" {
+		t.Errorf("GetUserByHandle: expected 'buoy.cr', got %q", byHandle.RegistryDomain)
+	}
+
+	// Clearing the preference propagates back to empty.
+	if err := UpdateUserRegistryDomain(db, user.DID, ""); err != nil {
+		t.Fatalf("UpdateUserRegistryDomain (clear) failed: %v", err)
+	}
+	cleared, err := GetUserByDID(db, user.DID)
+	if err != nil {
+		t.Fatalf("GetUserByDID failed: %v", err)
+	}
+	if cleared.RegistryDomain != "" {
+		t.Errorf("Expected empty registry domain after clear, got %q", cleared.RegistryDomain)
+	}
+}
+
 func TestUpdateUserHandle(t *testing.T) {
 	// Create in-memory test database
 	db, err := InitDB(":memory:", LibsqlConfig{})
@@ -14,6 +14,7 @@ CREATE TABLE IF NOT EXISTS users (
    avatar TEXT,
    default_hold_did TEXT,
    oci_client TEXT DEFAULT '',
+    registry_domain TEXT DEFAULT '',
    last_seen TIMESTAMP NOT NULL,
    UNIQUE(handle)
 );
@@ -19,9 +19,10 @@ import (
 // Route registration becomes simply &Handler{base} for everything.
 type BaseUIHandler struct {
 	// Display
-	Templates   *template.Template
-	RegistryURL string // Docker registry domain (e.g., "buoy.cr" or "atcr.io")
-	SiteURL     string // Website domain (e.g., "seamark.dev" or "atcr.io")
+	Templates       *template.Template
+	RegistryURL     string   // Primary Docker registry domain (e.g., "buoy.cr" or "atcr.io")
+	RegistryDomains []string // All configured registry domains; users may pick one as their default
+	SiteURL         string   // Website domain (e.g., "seamark.dev" or "atcr.io")

 	// Database (handlers choose which to use)
 	DB         *sql.DB // Write access
@@ -2,6 +2,7 @@ package handlers

 import (
 	"net/http"
+	"slices"
 	"strings"

 	"atcr.io/pkg/appview/db"
@@ -22,17 +23,28 @@ type PageData struct {
 	CurrentPath      string   // Request path (used for OAuth return_to)
 }

+// resolveRegistryURL returns the user's preferred registry domain when it is
+// non-empty and still one of the configured domains; otherwise the primary.
+// This keeps a stale or removed preference from breaking the displayed commands.
+func resolveRegistryURL(primary string, domains []string, pref string) string {
+	if pref != "" && slices.Contains(domains, pref) {
+		return pref
+	}
+	return primary
+}
+
 // NewPageData creates a PageData struct with common fields populated from the request
 func NewPageData(r *http.Request, h *BaseUIHandler) PageData {
 	user := middleware.GetUser(r)
-	var ociClient string
+	var ociClient, registryPref string
 	if user != nil {
 		ociClient = user.OciClient
+		registryPref = user.RegistryDomain
 	}
 	return PageData{
 		User:             user,
 		Query:            r.URL.Query().Get("q"),
-		RegistryURL:      h.RegistryURL,
+		RegistryURL:      resolveRegistryURL(h.RegistryURL, h.RegistryDomains, registryPref),
 		SiteURL:          h.SiteURL,
 		ClientName:       h.ClientName,
 		ClientShortName:  h.ClientShortName,
@@ -2,6 +2,34 @@ package handlers

 import "testing"

+func TestResolveRegistryURL(t *testing.T) {
+	primary := "buoy.cr"
+	domains := []string{"buoy.cr", "atcr.io"}
+
+	tests := []struct {
+		name    string
+		domains []string
+		pref    string
+		want    string
+	}{
+		{"empty pref falls back to primary", domains, "", "buoy.cr"},
+		{"valid pref is used", domains, "atcr.io", "atcr.io"},
+		{"pref equal to primary", domains, "buoy.cr", "buoy.cr"},
+		{"stale pref not in list falls back", domains, "gone.example", "buoy.cr"},
+		{"nil domains falls back", nil, "atcr.io", "buoy.cr"},
+		{"empty domains falls back", []string{}, "atcr.io", "buoy.cr"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := resolveRegistryURL(primary, tt.domains, tt.pref)
+			if got != tt.want {
+				t.Errorf("resolveRegistryURL(%q, %v, %q) = %q, want %q", primary, tt.domains, tt.pref, got, tt.want)
+			}
+		})
+	}
+}
+
 func TestTrimRegistryURL(t *testing.T) {
 	tests := []struct {
 		name     string
@@ -1,10 +1,12 @@
 package handlers

 import (
+	"database/sql"
 	"encoding/json"
 	"errors"
 	"fmt"
 	"io"
+	"log/slog"
 	"net/http"
 	"strings"
 	"time"
@@ -42,15 +44,30 @@ func (h *DeleteTagHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	repo := req.Repo
 	tag := req.Tag

+	// Look up the digest the tag points to (and the hold that owns its blobs)
+	// before we delete anything — we need both for the cascade decision and
+	// for purging the manifest's blobs on the hold. Missing-tag means there's
+	// nothing to delete; reuse 404 semantics.
+	digest, err := db.GetTagDigest(h.ReadOnlyDB, user.DID, repo, tag)
+	if err != nil {
+		if errors.Is(err, sql.ErrNoRows) {
+			http.Error(w, "Tag not found", http.StatusNotFound)
+			return
+		}
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	var holdDID string
+	if cached, err := db.GetManifestDetail(h.ReadOnlyDB, user.DID, repo, digest); err == nil && cached != nil {
+		holdDID = cached.HoldEndpoint
+	}
+
 	// Create ATProto client with session provider (uses DoWithSession for DPoP nonce safety)
 	pdsClient := atproto.NewClientWithSessionProvider(user.PDSEndpoint, user.DID, h.Refresher)

-	// Compute rkey for tag record (repository_tag with slashes replaced)
-	rkey := fmt.Sprintf("%s_%s", repo, tag)
-	rkey = strings.ReplaceAll(rkey, "/", "-")
-
-	// Delete from PDS first
-	if err := pdsClient.DeleteRecord(r.Context(), atproto.TagCollection, rkey); err != nil {
+	// Delete the tag record from PDS first.
+	tagRKey := atproto.RepositoryTagToRKey(repo, tag)
+	if err := pdsClient.DeleteRecord(r.Context(), atproto.TagCollection, tagRKey); err != nil {
 		// Check if OAuth error - if so, invalidate sessions and return 401
 		if handleOAuthError(r.Context(), h.Refresher, user.DID, err) {
 			http.Error(w, "Authentication failed, please log in again", http.StatusUnauthorized)
@@ -66,7 +83,32 @@ func (h *DeleteTagHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	// Return empty response (HTMX will swap out the element)
+	// If this was the last tag pointing to the digest and the digest isn't a
+	// child of any manifest list, cascade-delete the manifest too — that's
+	// what the user almost always wants when they click "delete tag" on an
+	// only-tagged image. Failures here are non-fatal: the tag is already
+	// gone, so the worst case is a leftover untagged manifest the user can
+	// clean up via "Delete untagged".
+	shouldCascade, err := db.ShouldCascadeDeleteManifest(h.ReadOnlyDB, user.DID, repo, digest)
+	if err != nil {
+		slog.Warn("delete-tag: failed to evaluate cascade", "did", user.DID, "repo", repo, "digest", digest, "error", err)
+	} else if shouldCascade {
+		manifestRKey := strings.TrimPrefix(digest, "sha256:")
+		if err := pdsClient.DeleteRecord(r.Context(), atproto.ManifestCollection, manifestRKey); err != nil {
+			slog.Warn("delete-tag: cascade PDS delete failed", "did", user.DID, "digest", digest, "error", err)
+		} else {
+			if err := db.DeleteManifest(h.DB, user.DID, repo, digest); err != nil {
+				slog.Warn("delete-tag: cascade DB delete failed", "did", user.DID, "digest", digest, "error", err)
+			}
+			purgeOnHold(r.Context(), h.Refresher, user.DID, user.PDSEndpoint, holdDID, atproto.BuildManifestURI(user.DID, digest))
+		}
+	}
+
+	// Return empty response (HTMX will swap out the element). The clicked
+	// row was tag-specific, so removing it is correct whether or not the
+	// underlying manifest was also cascaded. In the "multi-arch parent
+	// preserved the manifest" case, the untagged manifest will reappear as
+	// its own row on the next page load — which matches reality.
 	w.WriteHeader(http.StatusOK)
 }

@@ -8,6 +8,7 @@ import (
 	"log/slog"
 	"net/http"
 	"net/url"
+	"slices"
 	"strings"
 	"time"

@@ -77,6 +78,7 @@ type settingsProfile struct {
 	DefaultHold        string
 	AutoRemoveUntagged bool
 	OciClient          string
+	RegistryDomain     string
 	AIAdvisorEnabled   bool
 	HasAIAdvisorAccess bool
 }
@@ -87,16 +89,17 @@ type settingsProfile struct {
 // hold_selector template from doing filter-the-same-list-twice gymnastics.
 type settingsPageData struct {
 	PageData
-	Meta          *PageMeta
-	ActiveTab     string
-	Tabs          []settingsTab
-	Profile       settingsProfile
-	ActiveHold    *HoldDisplay
-	OtherHolds    []HoldDisplay
-	MemberHolds   []HoldDisplay
-	EligibleHolds []HoldDisplay
-	WebhooksData  webhooksTemplateData
-	Subscription  SubscriptionDisplay
+	Meta            *PageMeta
+	ActiveTab       string
+	Tabs            []settingsTab
+	Profile         settingsProfile
+	RegistryDomains []string
+	ActiveHold      *HoldDisplay
+	OtherHolds      []HoldDisplay
+	MemberHolds     []HoldDisplay
+	EligibleHolds   []HoldDisplay
+	WebhooksData    webhooksTemplateData
+	Subscription    SubscriptionDisplay
 }

 // ServeHTTP redirects /settings to /settings/user.
@@ -151,8 +154,10 @@ func (h *SettingsHandler) ServeTab(tab string) http.HandlerFunc {
 				DefaultHold:        profile.DefaultHold,
 				AutoRemoveUntagged: profile.AutoRemoveUntagged,
 				OciClient:          profile.OciClient,
+				RegistryDomain:     profile.RegistryDomain,
 				AIAdvisorEnabled:   profile.AIAdvisorEnabled == nil || *profile.AIAdvisorEnabled,
 			},
+			RegistryDomains: h.RegistryDomains,
 		}
 		if h.BillingManager != nil {
 			data.Profile.HasAIAdvisorAccess = h.BillingManager.HasAIAdvisor(user.DID)
@@ -564,6 +569,58 @@ func (h *UpdateOciClientHandler) ServeHTTP(w http.ResponseWriter, r *http.Reques
 	w.WriteHeader(http.StatusNoContent)
 }

+// UpdateRegistryDomainHandler handles updating the preferred registry domain
+type UpdateRegistryDomainHandler struct {
+	BaseUIHandler
+}
+
+func (h *UpdateRegistryDomainHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	user := middleware.GetUser(r)
+	if user == nil {
+		http.Error(w, "Unauthorized", http.StatusUnauthorized)
+		return
+	}
+
+	registryDomain := r.FormValue("registry_domain")
+	// Empty means "use the primary domain". Any non-empty value must be one of
+	// the configured registry domains.
+	if registryDomain != "" && !slices.Contains(h.RegistryDomains, registryDomain) {
+		http.Error(w, "Invalid registry domain", http.StatusBadRequest)
+		return
+	}
+
+	// Store empty string when the primary (first configured) domain is selected,
+	// so the preference tracks the primary even if the admin reorders domains.
+	if len(h.RegistryDomains) > 0 && registryDomain == h.RegistryDomains[0] {
+		registryDomain = ""
+	}
+
+	// Create ATProto client with session provider
+	client := atproto.NewClientWithSessionProvider(user.PDSEndpoint, user.DID, h.Refresher)
+
+	// Fetch existing profile
+	profile, err := storage.GetProfile(r.Context(), client)
+	if err != nil || profile == nil {
+		http.Error(w, "Failed to fetch profile", http.StatusInternalServerError)
+		return
+	}
+
+	profile.RegistryDomain = registryDomain
+	profile.UpdatedAt = time.Now()
+
+	if err := storage.UpdateProfile(r.Context(), client, profile); err != nil {
+		http.Error(w, "Failed to update profile: "+err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	// Cache locally
+	if h.DB != nil {
+		_ = db.UpdateUserRegistryDomain(h.DB, user.DID, registryDomain)
+	}
+
+	w.WriteHeader(http.StatusNoContent)
+}
+
 // UpdateAIAdvisorHandler handles toggling the AI Image Advisor setting
 type UpdateAIAdvisorHandler struct {
 	BaseUIHandler
@@ -542,6 +542,11 @@ func (p *Processor) ProcessSailorProfile(ctx context.Context, did string, record
 		}
 	}

+	// Cache preferred registry domain (write unconditionally so clearing it propagates)
+	if err := db.UpdateUserRegistryDomain(p.db, did, profileRecord.RegistryDomain); err != nil {
+		slog.Warn("Failed to cache registry domain preference", "component", "processor", "did", did, "registryDomain", profileRecord.RegistryDomain, "error", err)
+	}
+
 	// Skip hold processing if no default hold set
 	if profileRecord.DefaultHold == "" {
 		return nil
@@ -43,6 +43,7 @@ func setupTestDB(t *testing.T) *sql.DB {
 			avatar TEXT,
 			default_hold_did TEXT,
 			oci_client TEXT DEFAULT '',
+			registry_domain TEXT DEFAULT '',
 			last_seen TIMESTAMP NOT NULL
 		);

@@ -533,6 +534,46 @@ func TestProcessStar_InvalidRecord(t *testing.T) {
 	}
 }

+func TestProcessSailorProfile(t *testing.T) {
+	database := setupTestDB(t)
+	defer database.Close()
+
+	if _, err := database.Exec(
+		`INSERT INTO users (did, handle, pds_endpoint, last_seen) VALUES (?, ?, ?, ?)`,
+		"did:plc:profile123", "profile.test", "https://pds.example.com", time.Now()); err != nil {
+		t.Fatalf("Failed to insert user: %v", err)
+	}
+
+	p := NewProcessor(database, false, nil)
+	ctx := context.Background()
+
+	readDomain := func() string {
+		var d string
+		if err := database.QueryRow("SELECT registry_domain FROM users WHERE did = ?", "did:plc:profile123").Scan(&d); err != nil {
+			t.Fatalf("query registry_domain: %v", err)
+		}
+		return d
+	}
+
+	// A profile with registryDomain set caches it on the user row.
+	rec := []byte(`{"$type":"io.atcr.sailor.profile","registryDomain":"buoy.cr","createdAt":"2025-01-01T00:00:00Z"}`)
+	if err := p.ProcessSailorProfile(ctx, "did:plc:profile123", rec, nil); err != nil {
+		t.Fatalf("ProcessSailorProfile failed: %v", err)
+	}
+	if got := readDomain(); got != "buoy.cr" {
+		t.Errorf("expected cached registry domain 'buoy.cr', got %q", got)
+	}
+
+	// A profile with an empty registryDomain clears the cached value.
+	recEmpty := []byte(`{"$type":"io.atcr.sailor.profile","createdAt":"2025-01-01T00:00:00Z"}`)
+	if err := p.ProcessSailorProfile(ctx, "did:plc:profile123", recEmpty, nil); err != nil {
+		t.Fatalf("ProcessSailorProfile (empty) failed: %v", err)
+	}
+	if got := readDomain(); got != "" {
+		t.Errorf("expected cleared registry domain, got %q", got)
+	}
+}
+
 func TestProcessManifest_Duplicate(t *testing.T) {
 	database := setupTestDB(t)
 	defer database.Close()
@@ -33,7 +33,8 @@ type UIDependencies struct {
 	OAuthStore        *db.OAuthStore
 	Refresher         *oauth.Refresher
 	BaseURL           string
-	RegistryDomain    string // Separate OCI registry domain (e.g., "buoy.cr"); empty = same as BaseURL
+	RegistryDomain    string   // Separate OCI registry domain (e.g., "buoy.cr"); empty = same as BaseURL
+	RegistryDomains   []string // All configured registry domains; users may pick one as their default
 	DeviceStore       *db.DeviceStore
 	HealthChecker     *holdhealth.Checker
 	ReadmeFetcher     *readme.Fetcher
@@ -63,6 +64,7 @@ func RegisterUIRoutes(router chi.Router, deps UIDependencies) {
 	base := uihandlers.BaseUIHandler{
 		Templates:         deps.Templates,
 		RegistryURL:       registryURL,
+		RegistryDomains:   deps.RegistryDomains,
 		SiteURL:           siteURL,
 		DB:                deps.Database,
 		ReadOnlyDB:        deps.ReadOnlyDB,
@@ -208,6 +210,7 @@ func RegisterUIRoutes(router chi.Router, deps UIDependencies) {
 		r.Post("/api/profile/default-hold", (&uihandlers.UpdateDefaultHoldHandler{BaseUIHandler: base}).ServeHTTP)
 		r.Post("/api/profile/auto-remove-untagged", (&uihandlers.UpdateAutoRemoveUntaggedHandler{BaseUIHandler: base}).ServeHTTP)
 		r.Post("/api/profile/oci-client", (&uihandlers.UpdateOciClientHandler{BaseUIHandler: base}).ServeHTTP)
+		r.Post("/api/profile/registry-domain", (&uihandlers.UpdateRegistryDomainHandler{BaseUIHandler: base}).ServeHTTP)
 		r.Post("/api/profile/ai-advisor", (&uihandlers.UpdateAIAdvisorHandler{BaseUIHandler: base}).ServeHTTP)

 		// Subscription management
@@ -194,7 +194,7 @@ func NewAppViewServer(cfg *Config, branding *BrandingOverrides) (*AppViewServer,
 	slog.Info("Using SQLite for device storage")

 	baseURL := cfg.Server.BaseURL
-	defaultHoldDID := cfg.Server.DefaultHoldDID
+	defaultHoldDID := cfg.Server.PrimaryHoldDID()
 	testMode := cfg.Server.TestMode

 	slog.Debug("Base URL for OAuth", "base_url", baseURL)
@@ -340,6 +340,7 @@ func NewAppViewServer(cfg *Config, branding *BrandingOverrides) (*AppViewServer,
 		Refresher:         s.Refresher,
 		BaseURL:           baseURL,
 		RegistryDomain:    primaryRegistryDomain(cfg.Server.RegistryDomains),
+		RegistryDomains:   cfg.Server.RegistryDomains,
 		DeviceStore:       s.DeviceStore,
 		HealthChecker:     s.HealthChecker,
 		ReadmeFetcher:     s.ReadmeFetcher,
@@ -909,7 +910,7 @@ func (s *AppViewServer) initializeJetstream() {

 	if s.Config.Jetstream.BackfillEnabled {
 		relayEndpoints := s.Config.Jetstream.RelayEndpoints
-		defaultHoldDID := s.Config.Server.DefaultHoldDID
+		defaultHoldDID := s.Config.Server.PrimaryHoldDID()
 		testMode := s.Config.Server.TestMode

 		backfillWorker, err := jetstream.NewBackfillWorker(s.Database, relayEndpoints, defaultHoldDID, testMode, s.Refresher)
@@ -33,7 +33,7 @@
                {{ end }}
                <div class="flex flex-wrap items-center justify-center gap-2 min-w-0">
                    <h1 class="text-2xl md:text-3xl font-display font-bold tracking-tight break-all min-w-0">{{ .ViewedUser.Handle }}</h1>
-                    {{ if or (eq .SupporterBadge "Captain") (eq .SupporterBadge "owner") }}
+                    {{ if eq .SupporterBadge "Captain" }}
                        <span class="badge badge-sm supporter-badge-owner">{{ .SupporterBadge }}</span>
                    {{ else if .SupporterBadge }}
                        <span class="badge badge-sm supporter-badge">{{ .SupporterBadge }}</span>
@@ -5,13 +5,13 @@
        <p class="text-base-content/70 mt-1">Customize your experience across the site.</p>
    </div>

-    <div class="flex items-center gap-4">
-        <div>
+    <div class="flex flex-col gap-2 sm:flex-row sm:items-center sm:justify-between sm:gap-6">
+        <div class="sm:flex-1 sm:min-w-0">
            <label for="oci-client-select" class="text-sm font-medium">Preferred client</label>
            <p id="oci-client-hint" class="text-xs text-base-content/70">Sets the pull command shown on repository pages. Choose <em>Image reference only</em> to copy without a command prefix.</p>
        </div>
        {{ $oci := .Profile.OciClient }}
-        <select id="oci-client-select" aria-describedby="oci-client-hint" class="select select-sm select-bordered min-w-40"
+        <select id="oci-client-select" aria-describedby="oci-client-hint" class="select select-sm select-bordered w-full sm:w-56 sm:shrink-0"
                name="oci_client"
                hx-post="/api/profile/oci-client"
                hx-trigger="change"
@@ -25,6 +25,25 @@
        </select>
    </div>

+    {{ if gt (len .RegistryDomains) 1 }}
+    <div class="flex flex-col gap-2 sm:flex-row sm:items-center sm:justify-between sm:gap-6">
+        <div class="sm:flex-1 sm:min-w-0">
+            <label for="registry-domain-select" class="text-sm font-medium">Registry domain</label>
+            <p id="registry-domain-hint" class="text-xs text-base-content/70">Sets the registry domain shown in pull and push commands across the site. All domains work, this only changes what is displayed.</p>
+        </div>
+        {{ $pref := .Profile.RegistryDomain }}
+        <select id="registry-domain-select" aria-describedby="registry-domain-hint" class="select select-sm select-bordered w-full sm:w-56 sm:shrink-0"
+                name="registry_domain"
+                hx-post="/api/profile/registry-domain"
+                hx-trigger="change"
+                hx-swap="none">
+            {{ range $i, $d := .RegistryDomains }}
+            <option value="{{ $d }}"{{ if or (eq $d $pref) (and (eq $i 0) (eq $pref "")) }} selected{{ end }}>{{ $d }}</option>
+            {{ end }}
+        </select>
+    </div>
+    {{ end }}
+
    {{ if .AIAdvisorEnabled }}
    <div class="divider my-2"></div>
    <div class="flex items-start gap-3">
@@ -44,8 +63,8 @@
        <div>
            <span class="font-medium text-base-content/50">AI Image Advisor</span>
            <p class="text-xs text-base-content/70">Analyze your container images for optimization suggestions using AI.</p>
-            <p class="text-xs text-primary mt-1">
-                <a href="/settings/billing">Upgrade your plan</a> to enable this feature.
+            <p class="text-xs text-base-content/70 mt-1">
+                <a href="/settings/billing" class="text-primary link link-hover">Upgrade your plan</a> to enable this feature.
            </p>
        </div>
        {{ end }}
@@ -356,6 +356,10 @@ type SailorProfileRecord struct {
 	// "none" means image reference only (no `<client> pull ` prefix). Defaults to "docker" if empty.
 	OciClient string `json:"ociClient,omitempty"`

+	// RegistryDomain is the user's preferred registry domain for UI display.
+	// Must be one of the appview's configured registry_domains. Empty = primary (first configured).
+	RegistryDomain string `json:"registryDomain,omitempty"`
+
 	// AIAdvisorEnabled controls whether the AI Image Advisor feature is active for this user.
 	// nil = default (enabled if user has billing access), false = explicitly disabled.
 	AIAdvisorEnabled *bool `json:"aiAdvisorEnabled,omitempty"`
@@ -23,9 +23,6 @@ type Config struct {

 	// Subscription tiers with Stripe price IDs.
 	Tiers []BillingTierConfig `yaml:"tiers" comment:"Subscription tiers ordered by rank (lowest to highest)."`
-
-	// Whether hold owners get a supporter badge on their profile.
-	OwnerBadge bool `yaml:"owner_badge" comment:"Show supporter badge on hold owner profiles."`
 }

 // BillingTierConfig represents a single tier with optional Stripe pricing.
@@ -92,20 +92,10 @@ type AdminUI struct {
 	sessions   map[string]*AdminSession
 	sessionsMu sync.RWMutex

-	// scan-backfill state — runs as a background goroutine on click; the
-	// status endpoint reads this for progress polling. Only one run at a
-	// time (idempotent, so re-running is safe but pointless).
-	scanBackfill scanBackfillState
-}
-
-// scanBackfillState tracks the in-flight scan-status backfill run.
-type scanBackfillState struct {
-	mu        sync.Mutex
-	running   bool
-	startedAt time.Time
-	current   *pds.ScanBackfillResult // running totals (snapshot)
-	result    *pds.ScanBackfillResult // final result, set when running=false
-	err       string                  // last error (running ends with err set)
+	// jobs tracks long-running background admin operations (bulk crew tier
+	// remap, crew import, scan-record backfill). The kickoff handler returns a
+	// progress fragment that polls /admin/api/jobs/{key}/status. See jobs.go.
+	jobs jobRegistry
 }

 // adminContextKey is used to store session data in request context
@@ -544,11 +534,15 @@ func (ui *AdminUI) RegisterRoutes(r chi.Router) {
 		r.Get("/admin/api/relay/status", ui.handleRelayStatus)
 		r.Get("/admin/api/crew/member", ui.handleCrewMemberInfo)

-		// Scan-record backfill: kicks off a background run and returns a
-		// progress fragment that polls /status. Use Accept:application/json
-		// for a synchronous JSON response (curl-friendly).
+		// Scan-record backfill: kicks off a background job and returns a
+		// progress fragment that polls /admin/api/jobs/scan-backfill/status.
+		// Use Accept:application/json for a synchronous JSON response.
 		r.Post("/admin/api/scan-backfill", ui.handleScanBackfill)
-		r.Get("/admin/api/scan-backfill/status", ui.handleScanBackfillStatus)
+
+		// Generic background-job status, polled by progress fragments. Serves
+		// every job registered via startJob (see jobs.go) — crew tier remap,
+		// crew import, scan-record backfill.
+		r.Get("/admin/api/jobs/{key}/status", ui.handleJobStatus)

 		// Logout
 		r.Post("/admin/auth/logout", ui.handleLogout)
@@ -462,11 +462,16 @@ func (ui *AdminUI) handleCrewUpdate(w http.ResponseWriter, r *http.Request) {
 // because `to` is still validated against the live tier list.
 func (ui *AdminUI) handleCrewRemapTier(w http.ResponseWriter, r *http.Request) {
 	ctx := r.Context()
-	defer clearFlash(w)
+
+	// This endpoint is driven by htmx (hx-post on the reconciliation card), so
+	// synchronous validation failures must render a 200 fragment, not a 302 —
+	// an htmx swap can't follow a redirect cleanly.
+	renderErr := func(msg string) {
+		ui.renderTemplate(w, "partials/gc_error.html", struct{ Error string }{msg})
+	}

 	if err := r.ParseForm(); err != nil {
-		setFlash(w, r, "error", "Invalid form data")
-		http.Redirect(w, r, "/admin#crew", http.StatusFound)
+		renderErr("Invalid form data")
 		return
 	}

@@ -474,14 +479,12 @@ func (ui *AdminUI) handleCrewRemapTier(w http.ResponseWriter, r *http.Request) {
 	to := strings.TrimSpace(r.FormValue("to"))

 	if from == "" || to == "" {
-		setFlash(w, r, "error", "Both source and target tier are required")
-		http.Redirect(w, r, "/admin#crew", http.StatusFound)
+		renderErr("Both source and target tier are required")
 		return
 	}

 	if ui.quotaMgr == nil || !ui.quotaMgr.IsEnabled() {
-		setFlash(w, r, "error", "Quotas are not enabled on this hold")
-		http.Redirect(w, r, "/admin#crew", http.StatusFound)
+		renderErr("Quotas are not enabled on this hold")
 		return
 	}

@@ -495,60 +498,86 @@ func (ui *AdminUI) handleCrewRemapTier(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 	if !validTo {
-		setFlash(w, r, "error", fmt.Sprintf("Unknown target tier %q", to))
-		http.Redirect(w, r, "/admin#crew", http.StatusFound)
+		renderErr(fmt.Sprintf("Unknown target tier %q", to))
 		return
 	}

+	// List crew synchronously (fast) so we can report load failures inline and
+	// size the progress bar before detaching.
 	members, err := ui.pds.ListCrewMembers(ctx)
 	if err != nil {
 		slog.Error("Failed to list crew members for tier remap", "error", err)
-		setFlash(w, r, "error", "Failed to load crew: "+err.Error())
-		http.Redirect(w, r, "/admin#crew", http.StatusFound)
+		renderErr("Failed to load crew: " + err.Error())
 		return
 	}

-	var ok, failed int
+	total := 0
 	for _, member := range members {
-		if member.Record.Tier != from {
-			continue
+		if member.Record.Tier == from {
+			total++
 		}
-		if err := ui.pds.UpdateCrewMemberTier(ctx, member.Record.Member, to); err != nil {
-			slog.Warn("Failed to remap crew tier",
-				"did", member.Record.Member,
-				"from", from,
-				"to", to,
-				"error", err)
-			failed++
-			continue
-		}
-		ok++
-		// Throttle firehose events. Matches the convention in
-		// pkg/hold/gc/gc.go (image config backfill).
-		time.Sleep(200 * time.Millisecond)
+	}
+	if total == 0 {
+		ui.renderTemplate(w, "partials/job_result.html", jobResult{
+			Message: fmt.Sprintf("No crew were on tier %q.", from),
+		})
+		return
 	}

 	session := getSessionFromContext(ctx)
-	slog.Info("Bulk crew tier remap",
-		"from", from,
-		"to", to,
-		"updated", ok,
-		"failed", failed,
-		"by", func() string {
-			if session != nil {
-				return session.DID
-			}
-			return ""
-		}())
-
-	if ok == 0 && failed == 0 {
-		setFlash(w, r, "info", fmt.Sprintf("No crew were on tier %q", from))
-	} else if failed > 0 {
-		setFlash(w, r, "warning", fmt.Sprintf("Remapped %d crew from %q → %q (%d failed, see logs)", ok, from, to, failed))
-	} else {
-		setFlash(w, r, "success", fmt.Sprintf("Remapped %d crew from %q → %q", ok, from, to))
+	byDID := ""
+	if session != nil {
+		byDID = session.DID
 	}
-	http.Redirect(w, r, "/admin#crew", http.StatusFound)
+
+	// The per-member UpdateCrewMemberTier loop runs detached so it survives the
+	// reverse-proxy timeout that was cancelling it mid-run (~50 of N before a
+	// 504, leaving the rest as "context canceled").
+	ui.startJob("crew-remap-tier", "Remapping crew tier",
+		"partials/job_result.html", 10*time.Minute,
+		func(jobCtx context.Context, progress func(jobProgress)) (any, error) {
+			var done, ok, failed int
+			for _, member := range members {
+				if member.Record.Tier != from {
+					continue
+				}
+				done++
+				progress(jobProgress{
+					Done:    done,
+					Total:   total,
+					Message: fmt.Sprintf("Remapping %s", member.Record.Member),
+				})
+				if err := ui.pds.UpdateCrewMemberTier(jobCtx, member.Record.Member, to); err != nil {
+					slog.Warn("Failed to remap crew tier",
+						"did", member.Record.Member,
+						"from", from,
+						"to", to,
+						"error", err)
+					failed++
+					continue
+				}
+				ok++
+				// Throttle firehose events. Matches the convention in
+				// pkg/hold/gc/gc.go (image config backfill).
+				time.Sleep(200 * time.Millisecond)
+			}
+
+			slog.Info("Bulk crew tier remap",
+				"from", from, "to", to, "updated", ok, "failed", failed, "by", byDID)
+
+			msg := fmt.Sprintf("Remapped %d crew from %q → %q", ok, from, to)
+			if failed > 0 {
+				msg += fmt.Sprintf(" (%d failed, see logs)", failed)
+			}
+			return jobResult{
+				Message:      msg,
+				Failed:       failed,
+				ReloadURL:    "/admin/api/tab/crew",
+				ReloadTarget: "#tab-crew",
+			}, nil
+		})
+
+	ui.renderTemplate(w, "partials/job_progress.html", ui.jobSnapshot("crew-remap-tier"))
 }

 // handleCrewDelete removes a crew member
@@ -1,6 +1,7 @@
 package admin

 import (
+	"context"
 	"encoding/json"
 	"fmt"
 	"log/slog"
@@ -34,6 +35,15 @@ type importResult struct {
 	Reason string
 }

+// crewImportResult is the job result rendered by partials/crew_import_results.html.
+type crewImportResult struct {
+	Results []importResult
+	Added   int
+	Skipped int
+	Errors  int
+	Total   int
+}
+
 const maxImportSize = 1 << 20 // 1 MB

 // handleCrewExport exports all crew members as a JSON file download
@@ -92,22 +102,28 @@ func (ui *AdminUI) handleCrewImportForm(w http.ResponseWriter, r *http.Request)
 	ui.renderTemplate(w, "pages/crew_import.html", data)
 }

-// handleCrewImport processes an uploaded crew JSON file
+// handleCrewImport processes an uploaded crew JSON file. The upload is parsed
+// and decoded synchronously (the request body can't be read after the handler
+// returns), then the per-entry loop — which does a network handle resolution
+// plus a PDS write per member — runs as a detached background job so a large
+// file can't 504. The form is htmx-driven, so failures render 200 fragments.
 func (ui *AdminUI) handleCrewImport(w http.ResponseWriter, r *http.Request) {
 	ctx := r.Context()

+	renderErr := func(msg string) {
+		ui.renderTemplate(w, "partials/gc_error.html", struct{ Error string }{msg})
+	}
+
 	r.Body = http.MaxBytesReader(w, r.Body, maxImportSize)

 	if err := r.ParseMultipartForm(maxImportSize); err != nil {
-		setFlash(w, r, "error", "File too large (max 1 MB)")
-		http.Redirect(w, r, "/admin/crew/import", http.StatusFound)
+		renderErr("File too large (max 1 MB)")
 		return
 	}

 	file, _, err := r.FormFile("crew_file")
 	if err != nil {
-		setFlash(w, r, "error", "No file uploaded")
-		http.Redirect(w, r, "/admin/crew/import", http.StatusFound)
+		renderErr("No file uploaded")
 		return
 	}
 	defer file.Close()
@@ -115,99 +131,103 @@ func (ui *AdminUI) handleCrewImport(w http.ResponseWriter, r *http.Request) {
 	var export crewExportFile
 	dec := json.NewDecoder(file)
 	if err := dec.Decode(&export); err != nil {
-		setFlash(w, r, "error", "Invalid JSON: "+err.Error())
-		http.Redirect(w, r, "/admin/crew/import", http.StatusFound)
+		renderErr("Invalid JSON: " + err.Error())
 		return
 	}

 	if export.Version != 1 {
-		setFlash(w, r, "error", fmt.Sprintf("Unsupported export version: %d (expected 1)", export.Version))
-		http.Redirect(w, r, "/admin/crew/import", http.StatusFound)
+		renderErr(fmt.Sprintf("Unsupported export version: %d (expected 1)", export.Version))
 		return
 	}

 	if len(export.Crew) == 0 {
-		setFlash(w, r, "error", "No crew members in file")
-		http.Redirect(w, r, "/admin/crew/import", http.StatusFound)
+		renderErr("No crew members in file")
 		return
 	}

-	var results []importResult
-	for _, entry := range export.Crew {
-		result := importResult{DID: entry.DID}
-
-		if !strings.HasPrefix(entry.DID, "did:") {
-			result.Status = "error"
-			result.Reason = "Invalid DID format"
-			results = append(results, result)
-			continue
-		}
-
-		// Check if member already exists (O(1) lookup)
-		_, _, err := ui.pds.GetCrewMemberByDID(ctx, entry.DID)
-		if err == nil {
-			result.Status = "skipped"
-			result.Reason = "Already exists"
-			results = append(results, result)
-			continue
-		}
-
-		role := entry.Role
-		if role == "" {
-			role = "member"
-		}
-
-		// Resolve tier: use entry tier if specified, otherwise default from quota config
-		tier := entry.Tier
-		if tier == "" && ui.quotaMgr != nil && ui.quotaMgr.IsEnabled() {
-			tier = ui.quotaMgr.GetDefaultTier()
-		}
-
-		if _, err := ui.pds.AddCrewMember(ctx, entry.DID, role, entry.Permissions, tier); err != nil {
-			result.Status = "error"
-			result.Reason = err.Error()
-			results = append(results, result)
-			continue
-		}
-
-		result.Status = "added"
-		result.Handle = resolveHandle(ctx, entry.DID)
-		results = append(results, result)
-	}
-
-	var added, skipped, errored int
-	for _, res := range results {
-		switch res.Status {
-		case "added":
-			added++
-		case "skipped":
-			skipped++
-		case "error":
-			errored++
-		}
-	}
-
+	// export.Crew is fully in memory (request body capped at maxImportSize), so
+	// it's safe to hand to the detached job.
+	entries := export.Crew
 	session := getSessionFromContext(ctx)
-	slog.Info("Crew imported via admin panel",
-		"added", added,
-		"skipped", skipped,
-		"errors", errored,
-		"by", session.DID)
-
-	data := struct {
-		PageData
-		Results []importResult
-		Added   int
-		Skipped int
-		Errors  int
-		Total   int
-	}{
-		PageData: ui.newPageData(r, "Import Results", "crew"),
-		Results:  results,
-		Added:    added,
-		Skipped:  skipped,
-		Errors:   errored,
-		Total:    len(results),
+	byDID := ""
+	if session != nil {
+		byDID = session.DID
 	}
-	ui.renderTemplate(w, "pages/crew_import_results.html", data)
+
+	ui.startJob("crew-import", "Importing crew",
+		"partials/crew_import_results.html", 10*time.Minute,
+		func(jobCtx context.Context, progress func(jobProgress)) (any, error) {
+			results := make([]importResult, 0, len(entries))
+			for i, entry := range entries {
+				progress(jobProgress{
+					Done:    i + 1,
+					Total:   len(entries),
+					Message: fmt.Sprintf("Importing %s", entry.DID),
+				})
+
+				result := importResult{DID: entry.DID}
+
+				if !strings.HasPrefix(entry.DID, "did:") {
+					result.Status = "error"
+					result.Reason = "Invalid DID format"
+					results = append(results, result)
+					continue
+				}
+
+				// Check if member already exists (O(1) lookup)
+				if _, _, err := ui.pds.GetCrewMemberByDID(jobCtx, entry.DID); err == nil {
+					result.Status = "skipped"
+					result.Reason = "Already exists"
+					results = append(results, result)
+					continue
+				}
+
+				role := entry.Role
+				if role == "" {
+					role = "member"
+				}
+
+				// Resolve tier: use entry tier if specified, otherwise default from quota config
+				tier := entry.Tier
+				if tier == "" && ui.quotaMgr != nil && ui.quotaMgr.IsEnabled() {
+					tier = ui.quotaMgr.GetDefaultTier()
+				}
+
+				if _, err := ui.pds.AddCrewMember(jobCtx, entry.DID, role, entry.Permissions, tier); err != nil {
+					result.Status = "error"
+					result.Reason = err.Error()
+					results = append(results, result)
+					continue
+				}
+
+				result.Status = "added"
+				result.Handle = resolveHandle(jobCtx, entry.DID)
+				results = append(results, result)
+			}
+
+			var added, skipped, errored int
+			for _, res := range results {
+				switch res.Status {
+				case "added":
+					added++
+				case "skipped":
+					skipped++
+				case "error":
+					errored++
+				}
+			}
+
+			slog.Info("Crew imported via admin panel",
+				"added", added, "skipped", skipped, "errors", errored, "by", byDID)
+
+			return crewImportResult{
+				Results: results,
+				Added:   added,
+				Skipped: skipped,
+				Errors:  errored,
+				Total:   len(results),
+			}, nil
+		})
+
+	ui.renderTemplate(w, "partials/job_progress.html", ui.jobSnapshot("crew-import"))
 }
@@ -12,14 +12,14 @@ import (
 	"atcr.io/pkg/hold/pds"
 )

-// handleScanBackfill kicks off a scan-status backfill in a background
-// goroutine and returns a progress fragment that polls
-// /admin/api/scan-backfill/status for updates. Idempotent — clicking again
-// while a run is in flight just shows the current progress.
+// handleScanBackfill kicks off a scan-status backfill as a background job and
+// returns a progress fragment that polls /admin/api/jobs/scan-backfill/status.
+// Idempotent — clicking again while a run is in flight just shows the current
+// progress (startJob returns false).
 //
 // Why background: reverse proxies typically cap upstream HTTP timeouts at
-// 10–60s, which would cancel a synchronous request mid-loop. Detaching the
-// work from the request context lets it run to completion.
+// 10–60s, which would cancel a synchronous request mid-loop. The job runs under
+// its own detached context (see jobs.go), so it survives the request ending.
 //
 // JSON callers (Accept: application/json) get a synchronous run instead —
 // useful for curl + scripting.
@@ -50,119 +50,36 @@ func (ui *AdminUI) handleScanBackfill(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	// HTML path — kick off the background run if one isn't already going.
-	st := &ui.scanBackfill
-	st.mu.Lock()
-	alreadyRunning := st.running
-	if !alreadyRunning {
-		st.running = true
-		st.startedAt = time.Now()
-		st.current = &pds.ScanBackfillResult{}
-		st.result = nil
-		st.err = ""
-	}
-	st.mu.Unlock()
+	started := ui.startJob("scan-backfill", "Backfilling scan records",
+		"partials/scan_backfill_result.html", 10*time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			res, err := ui.pds.BackfillScanStatus(ctx, scanBackfillLogger, func(snap *pds.ScanBackfillResult) {
+				progress(jobProgress{
+					Done: snap.Scanned,
+					Message: fmt.Sprintf("Scanned %d · rewrites %d (%d skipped, %d failed)",
+						snap.Scanned, snap.Rewritten, snap.MarkedSkipped, snap.MarkedFailed),
+				})
+			})
+			if err != nil {
+				return nil, err
+			}
+			slog.Info("scan-status backfill complete",
+				"scanned", res.Scanned,
+				"already_tagged", res.AlreadyTagged,
+				"marked_skipped", res.MarkedSkipped,
+				"marked_failed", res.MarkedFailed,
+				"rewritten", res.Rewritten,
+			)
+			return res, nil
+		})

-	if !alreadyRunning {
+	if started {
 		slog.Info("scan-status backfill started via admin panel", "by", session.DID)
-		go ui.runScanBackfill()
 	} else {
 		slog.Debug("scan-status backfill already in progress; returning current state")
 	}

-	ui.renderTemplate(w, "partials/scan_backfill_progress.html", ui.snapshotScanBackfill())
-}
-
-// handleScanBackfillStatus is polled by the progress fragment. Returns the
-// progress fragment again if running, the result fragment when done, or an
-// error fragment if something went wrong.
-func (ui *AdminUI) handleScanBackfillStatus(w http.ResponseWriter, r *http.Request) {
-	snap := ui.snapshotScanBackfill()
-	if snap.Running {
-		ui.renderTemplate(w, "partials/scan_backfill_progress.html", snap)
-		return
-	}
-	if snap.Error != "" {
-		ui.renderTemplate(w, "partials/gc_error.html", struct{ Error string }{snap.Error})
-		return
-	}
-	if snap.Result == nil {
-		// Initial state, before any run — render an empty placeholder.
-		_, _ = w.Write([]byte(""))
-		return
-	}
-	ui.renderTemplate(w, "partials/scan_backfill_result.html", snap.Result)
-}
-
-// runScanBackfill is the goroutine body. Updates the shared state as the
-// backfill progresses and stores the final result (or error) when it ends.
-func (ui *AdminUI) runScanBackfill() {
-	st := &ui.scanBackfill
-	defer func() {
-		st.mu.Lock()
-		st.running = false
-		st.mu.Unlock()
-	}()
-
-	// Generous independent timeout — the loop is single-threaded and large
-	// holds with thousands of legacy records can take a few minutes.
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Minute)
-	defer cancel()
-
-	res, err := ui.pds.BackfillScanStatus(ctx, scanBackfillLogger, func(snap *pds.ScanBackfillResult) {
-		// Copy so we don't keep a pointer the loop will mutate.
-		c := *snap
-		st.mu.Lock()
-		st.current = &c
-		st.mu.Unlock()
-	})
-	st.mu.Lock()
-	if err != nil {
-		st.err = err.Error()
-		slog.Error("scan-status backfill failed", "error", err)
-	} else {
-		st.result = res
-		slog.Info("scan-status backfill complete",
-			"scanned", res.Scanned,
-			"already_tagged", res.AlreadyTagged,
-			"marked_skipped", res.MarkedSkipped,
-			"marked_failed", res.MarkedFailed,
-			"rewritten", res.Rewritten,
-		)
-	}
-	st.mu.Unlock()
-}
-
-// scanBackfillSnapshot is the shape exposed to templates.
-type scanBackfillSnapshot struct {
-	Running   bool
-	StartedAt time.Time
-	Current   *pds.ScanBackfillResult // populated while running
-	Result    *pds.ScanBackfillResult // populated when complete
-	Error     string
-}
-
-// snapshotScanBackfill returns a copy of the current state — safe to render
-// without holding the mutex.
-func (ui *AdminUI) snapshotScanBackfill() scanBackfillSnapshot {
-	st := &ui.scanBackfill
-	st.mu.Lock()
-	defer st.mu.Unlock()
-
-	snap := scanBackfillSnapshot{
-		Running:   st.running,
-		StartedAt: st.startedAt,
-		Error:     st.err,
-	}
-	if st.current != nil {
-		c := *st.current
-		snap.Current = &c
-	}
-	if st.result != nil {
-		r := *st.result
-		snap.Result = &r
-	}
-	return snap
+	ui.renderTemplate(w, "partials/job_progress.html", ui.jobSnapshot("scan-backfill"))
 }

 // scanBackfillLogger formats the printf-style messages from BackfillScanStatus
@@ -0,0 +1,183 @@
+package admin
+
+import (
+	"context"
+	"net/http"
+	"sync"
+	"time"
+
+	"github.com/go-chi/chi/v5"
+)
+
+// Long-running admin operations (bulk crew tier remaps, crew imports,
+// scan-record backfills) must NOT run as a synchronous loop bound to the HTTP
+// request context. A reverse proxy times out such a request around the 10s mark
+// and cancels r.Context(), which aborts the work mid-flight ("context canceled"
+// from the blockstore). Instead they run here as a detached background job: the
+// kickoff handler returns a progress fragment immediately, the work runs under
+// its own context.Background() timeout, and the page polls
+// /admin/api/jobs/{key}/status until it finishes.
+//
+// This is the same shape hand-rolled in gc.startBackground; the admin package
+// keeps its own copy because gc must not import admin.
+
+// jobProgress is the live progress of a running job. It is a value type copied
+// under the job's lock — never hand a pointer the job loop mutates to a template.
+type jobProgress struct {
+	Done    int
+	Total   int // 0 = indeterminate (spinner only, no progress bar)
+	Message string
+}
+
+// jobResult is the generic success payload rendered by partials/job_result.html.
+// Jobs with richer output (crew import, scan backfill) register their own result
+// template and return a different struct instead.
+type jobResult struct {
+	Message      string // summary line
+	Failed       int    // >0 renders the alert as a warning rather than success
+	ReloadURL    string // optional htmx affordance to reload a tab after the job
+	ReloadTarget string // CSS selector the ReloadURL swaps into
+}
+
+// jobState is the state machine for one background job, keyed by a stable string
+// (e.g. "crew-remap-tier"). Only one run per key at a time.
+type jobState struct {
+	mu             sync.Mutex
+	title          string
+	resultTemplate string
+	started        bool
+	running        bool
+	startedAt      time.Time
+	progress       jobProgress
+	result         any
+	err            string
+}
+
+// jobRegistry holds one jobState per key. The zero value is ready to use.
+type jobRegistry struct {
+	mu   sync.Mutex
+	jobs map[string]*jobState
+}
+
+// get returns the jobState for key, creating an empty one on first access.
+func (r *jobRegistry) get(key string) *jobState {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	if r.jobs == nil {
+		r.jobs = make(map[string]*jobState)
+	}
+	st, ok := r.jobs[key]
+	if !ok {
+		st = &jobState{}
+		r.jobs[key] = st
+	}
+	return st
+}
+
+// jobSnapshot is the read-only view rendered to templates.
+type jobSnapshot struct {
+	Key            string
+	Title          string
+	ResultTemplate string
+	Started        bool
+	Running        bool
+	StartedAt      time.Time
+	Progress       jobProgress
+	Result         any
+	Error          string
+}
+
+// startJob launches fn in a detached goroutine under its own timeout and returns
+// true. If a job with this key is already running it returns false and leaves the
+// in-flight run untouched (the caller should just render the current snapshot).
+//
+// fn publishes progress via the passed callback and returns a result value
+// (rendered by resultTemplate) or an error (rendered by partials/gc_error.html).
+// fn must use the ctx it is given — that ctx carries the detached timeout, not
+// the request deadline.
+func (ui *AdminUI) startJob(key, title, resultTemplate string, timeout time.Duration,
+	fn func(ctx context.Context, progress func(jobProgress)) (any, error)) bool {
+	st := ui.jobs.get(key)
+
+	st.mu.Lock()
+	if st.running {
+		st.mu.Unlock()
+		return false
+	}
+	st.running = true
+	st.started = true
+	st.startedAt = time.Now()
+	st.title = title
+	st.resultTemplate = resultTemplate
+	st.progress = jobProgress{}
+	st.result = nil
+	st.err = ""
+	st.mu.Unlock()
+
+	publish := func(p jobProgress) {
+		st.mu.Lock()
+		st.progress = p
+		st.mu.Unlock()
+	}
+
+	go func() {
+		defer func() {
+			st.mu.Lock()
+			st.running = false
+			st.mu.Unlock()
+		}()
+
+		ctx, cancel := context.WithTimeout(context.Background(), timeout)
+		defer cancel()
+
+		res, err := fn(ctx, publish)
+
+		st.mu.Lock()
+		if err != nil {
+			st.err = err.Error()
+		} else {
+			st.result = res
+		}
+		st.mu.Unlock()
+	}()
+
+	return true
+}
+
+// jobSnapshot returns a copy of the current state for key — safe to render
+// without holding the lock.
+func (ui *AdminUI) jobSnapshot(key string) jobSnapshot {
+	st := ui.jobs.get(key)
+	st.mu.Lock()
+	defer st.mu.Unlock()
+	return jobSnapshot{
+		Key:            key,
+		Title:          st.title,
+		ResultTemplate: st.resultTemplate,
+		Started:        st.started,
+		Running:        st.running,
+		StartedAt:      st.startedAt,
+		Progress:       st.progress,
+		Result:         st.result,
+		Error:          st.err,
+	}
+}
+
+// handleJobStatus is polled by the progress fragment. It renders the progress
+// fragment while running, the error fragment on failure, or the job's registered
+// result template on success. A never-started key renders an empty body.
+func (ui *AdminUI) handleJobStatus(w http.ResponseWriter, r *http.Request) {
+	key := chi.URLParam(r, "key")
+	snap := ui.jobSnapshot(key)
+
+	switch {
+	case !snap.Started:
+		_, _ = w.Write([]byte(""))
+	case snap.Running:
+		ui.renderTemplate(w, "partials/job_progress.html", snap)
+	case snap.Error != "":
+		ui.renderTemplate(w, "partials/gc_error.html", struct{ Error string }{snap.Error})
+	default:
+		ui.renderTemplate(w, snap.ResultTemplate, snap.Result)
+	}
+}
@@ -0,0 +1,309 @@
+package admin
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/go-chi/chi/v5"
+)
+
+// waitDone polls until the job is finished (Started && !Running). Because the
+// goroutine sets running=false in a defer that runs AFTER the result is stored,
+// observing !Running guarantees the result/error is visible — no race.
+func waitDone(t *testing.T, ui *AdminUI, key string) jobSnapshot {
+	t.Helper()
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		snap := ui.jobSnapshot(key)
+		if snap.Started && !snap.Running {
+			return snap
+		}
+		time.Sleep(time.Millisecond)
+	}
+	t.Fatalf("job %q did not finish within deadline", key)
+	return jobSnapshot{}
+}
+
+func TestStartJob_HappyPath(t *testing.T) {
+	ui := &AdminUI{}
+	release := make(chan struct{})
+
+	started := ui.startJob("k", "Doing thing", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			<-release
+			return jobResult{Message: "all done"}, nil
+		})
+	if !started {
+		t.Fatal("startJob returned false for a fresh key")
+	}
+
+	// While the fn is blocked, the job is observably running.
+	snap := ui.jobSnapshot("k")
+	if !snap.Started || !snap.Running {
+		t.Fatalf("expected started+running, got started=%v running=%v", snap.Started, snap.Running)
+	}
+	if snap.Title != "Doing thing" {
+		t.Errorf("title = %q, want %q", snap.Title, "Doing thing")
+	}
+
+	close(release)
+	snap = waitDone(t, ui, "k")
+
+	if snap.Running {
+		t.Error("job still running after completion")
+	}
+	if snap.Error != "" {
+		t.Errorf("unexpected error: %q", snap.Error)
+	}
+	res, ok := snap.Result.(jobResult)
+	if !ok {
+		t.Fatalf("result type = %T, want jobResult", snap.Result)
+	}
+	if res.Message != "all done" {
+		t.Errorf("result message = %q, want %q", res.Message, "all done")
+	}
+}
+
+func TestStartJob_DoubleStartGuard(t *testing.T) {
+	ui := &AdminUI{}
+	release := make(chan struct{})
+
+	if !ui.startJob("k", "first", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			<-release
+			return jobResult{Message: "first"}, nil
+		}) {
+		t.Fatal("first startJob returned false")
+	}
+
+	// Second start while the first is in flight must be rejected and must not
+	// clobber the running job's title.
+	if ui.startJob("k", "second", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			return jobResult{Message: "second"}, nil
+		}) {
+		t.Fatal("second startJob returned true while a run was in flight")
+	}
+	if snap := ui.jobSnapshot("k"); snap.Title != "first" {
+		t.Errorf("running job title clobbered: %q", snap.Title)
+	}
+
+	close(release)
+	snap := waitDone(t, ui, "k")
+	if res := snap.Result.(jobResult); res.Message != "first" {
+		t.Errorf("result message = %q, want from first run", res.Message)
+	}
+}
+
+func TestStartJob_RerunAfterCompletion(t *testing.T) {
+	ui := &AdminUI{}
+
+	ui.startJob("k", "run1", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			progress(jobProgress{Done: 5, Total: 5})
+			return jobResult{Message: "run1"}, nil
+		})
+	waitDone(t, ui, "k")
+
+	// A second start with the same key succeeds and resets progress/result/err.
+	release := make(chan struct{})
+	if !ui.startJob("k", "run2", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			<-release
+			return jobResult{Message: "run2"}, nil
+		}) {
+		t.Fatal("re-run startJob returned false after completion")
+	}
+	snap := ui.jobSnapshot("k")
+	if snap.Progress != (jobProgress{}) {
+		t.Errorf("progress not reset on re-run: %+v", snap.Progress)
+	}
+	if snap.Result != nil {
+		t.Errorf("result not reset on re-run: %v", snap.Result)
+	}
+	close(release)
+	waitDone(t, ui, "k")
+}
+
+func TestStartJob_ProgressPublishingIsCopied(t *testing.T) {
+	ui := &AdminUI{}
+	publishedOne := make(chan struct{})
+	proceed := make(chan struct{})
+	publishedTwo := make(chan struct{})
+	release := make(chan struct{})
+
+	ui.startJob("k", "t", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			progress(jobProgress{Done: 1, Total: 3, Message: "one"})
+			publishedOne <- struct{}{}
+			<-proceed // wait until the test has snapshotted the first value
+			progress(jobProgress{Done: 2, Total: 3, Message: "two"})
+			publishedTwo <- struct{}{}
+			<-release
+			return jobResult{}, nil
+		})
+
+	// Synchronize on the first publish, then snapshot it (fn is parked on proceed).
+	<-publishedOne
+	first := ui.jobSnapshot("k")
+	if first.Progress.Done != 1 || first.Progress.Message != "one" {
+		t.Fatalf("first progress = %+v", first.Progress)
+	}
+
+	// Let the job publish again; the earlier snapshot must be an independent copy.
+	proceed <- struct{}{}
+	<-publishedTwo
+	if first.Progress.Done != 1 || first.Progress.Message != "one" {
+		t.Errorf("earlier snapshot mutated: %+v", first.Progress)
+	}
+	second := ui.jobSnapshot("k")
+	if second.Progress.Done != 2 || second.Progress.Message != "two" {
+		t.Errorf("second progress = %+v", second.Progress)
+	}
+
+	close(release)
+	waitDone(t, ui, "k")
+}
+
+func TestStartJob_ErrorPath(t *testing.T) {
+	ui := &AdminUI{}
+	ui.startJob("k", "t", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			return nil, errors.New("boom")
+		})
+	snap := waitDone(t, ui, "k")
+	if snap.Error != "boom" {
+		t.Errorf("error = %q, want %q", snap.Error, "boom")
+	}
+	if snap.Result != nil {
+		t.Errorf("result = %v, want nil on error", snap.Result)
+	}
+}
+
+func TestStartJob_RespectsTimeout(t *testing.T) {
+	ui := &AdminUI{}
+	ui.startJob("k", "t", "partials/job_result.html", 10*time.Millisecond,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			<-ctx.Done() // detached timeout fires here, not the request deadline
+			return nil, ctx.Err()
+		})
+	snap := waitDone(t, ui, "k")
+	if snap.Error != context.DeadlineExceeded.Error() {
+		t.Errorf("error = %q, want %q", snap.Error, context.DeadlineExceeded.Error())
+	}
+}
+
+func TestJobSnapshot_UnknownKey(t *testing.T) {
+	ui := &AdminUI{}
+	snap := ui.jobSnapshot("never-run")
+	if snap.Started || snap.Running {
+		t.Errorf("unknown key reported started=%v running=%v", snap.Started, snap.Running)
+	}
+}
+
+func TestStartJob_DistinctKeysAreIndependent(t *testing.T) {
+	ui := &AdminUI{}
+	releaseA := make(chan struct{})
+
+	ui.startJob("a", "A", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			<-releaseA
+			return jobResult{Message: "a"}, nil
+		})
+	ui.startJob("b", "B", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			return jobResult{Message: "b"}, nil
+		})
+
+	// b finishes while a is still blocked.
+	bSnap := waitDone(t, ui, "b")
+	if bSnap.Result.(jobResult).Message != "b" {
+		t.Errorf("b result = %v", bSnap.Result)
+	}
+	if aSnap := ui.jobSnapshot("a"); !aSnap.Running {
+		t.Error("job a should still be running while b finished")
+	}
+
+	close(releaseA)
+	aSnap := waitDone(t, ui, "a")
+	if aSnap.Result.(jobResult).Message != "a" {
+		t.Errorf("a result = %v", aSnap.Result)
+	}
+}
+
+// TestStartJob_OutlivesCaller is the core regression: the work must complete
+// even after the calling scope (the HTTP handler) has returned.
+func TestStartJob_OutlivesCaller(t *testing.T) {
+	ui := &AdminUI{}
+
+	// kickoff mimics a handler that returns immediately after starting the job.
+	kickoff := func() {
+		ui.startJob("k", "t", "partials/job_result.html", time.Minute,
+			func(ctx context.Context, progress func(jobProgress)) (any, error) {
+				time.Sleep(20 * time.Millisecond)
+				return jobResult{Message: "finished after handler returned"}, nil
+			})
+	}
+	kickoff() // handler scope ends here
+
+	snap := waitDone(t, ui, "k")
+	if snap.Result.(jobResult).Message != "finished after handler returned" {
+		t.Errorf("job did not complete after caller returned: %v", snap.Result)
+	}
+}
+
+func TestHandleJobStatus_Dispatch(t *testing.T) {
+	tmpls, err := parseTemplates()
+	if err != nil {
+		t.Fatalf("parseTemplates: %v", err)
+	}
+	ui := &AdminUI{templates: tmpls}
+
+	status := func(key string) (int, string) {
+		req := httptest.NewRequest(http.MethodGet, "/admin/api/jobs/"+key+"/status", nil)
+		rctx := chi.NewRouteContext()
+		rctx.URLParams.Add("key", key)
+		req = req.WithContext(context.WithValue(req.Context(), chi.RouteCtxKey, rctx))
+		rec := httptest.NewRecorder()
+		ui.handleJobStatus(rec, req)
+		return rec.Code, rec.Body.String()
+	}
+
+	// not-started -> empty body
+	if code, body := status("nope"); code != http.StatusOK || strings.TrimSpace(body) != "" {
+		t.Errorf("not-started: code=%d body=%q", code, body)
+	}
+
+	// running -> progress fragment (polls the status endpoint)
+	release := make(chan struct{})
+	ui.startJob("running", "Working", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			<-release
+			return jobResult{Message: "done"}, nil
+		})
+	if _, body := status("running"); !strings.Contains(body, "/admin/api/jobs/running/status") {
+		t.Errorf("running fragment missing poll URL: %q", body)
+	}
+	close(release)
+	waitDone(t, ui, "running")
+
+	// done -> result template
+	if _, body := status("running"); !strings.Contains(body, "done") {
+		t.Errorf("result fragment missing message: %q", body)
+	}
+
+	// error -> gc_error fragment
+	ui.startJob("failed", "Working", "partials/job_result.html", time.Minute,
+		func(ctx context.Context, progress func(jobProgress)) (any, error) {
+			return nil, errors.New("kaput")
+		})
+	waitDone(t, ui, "failed")
+	if _, body := status("failed"); !strings.Contains(body, "alert-error") || !strings.Contains(body, "kaput") {
+		t.Errorf("error fragment wrong: %q", body)
+	}
+}
@@ -11,7 +11,11 @@
 {{define "page-content"}}
 <div class="card bg-base-100 shadow-sm">
    <div class="card-body">
-        <form action="/admin/crew/import" method="POST" enctype="multipart/form-data" class="max-w-lg">
+        <form hx-post="/admin/crew/import"
+              hx-encoding="multipart/form-data"
+              hx-target="#import-status"
+              hx-swap="innerHTML"
+              class="max-w-lg">
            {{ csrfInput .CSRFToken }}
            <fieldset class="fieldset mb-6">
                <label class="fieldset-label font-medium" for="crew_file">Crew Export File</label>
@@ -37,6 +41,8 @@
                <a href="/admin#crew" class="btn btn-ghost">Cancel</a>
            </div>
        </form>
+
+        <div id="import-status" class="mt-6"></div>
    </div>
 </div>
 {{end}}
@@ -1,14 +1,4 @@
-{{define "page-header"}}
-<div class="flex justify-between items-center min-h-12 mb-6">
-    <h1 class="text-2xl font-bold">Import Results</h1>
-    <a href="/admin#crew" class="btn btn-ghost gap-2">
-        {{ icon "arrow-left" "size-4" }}
-        Back to Crew
-    </a>
-</div>
-{{end}}
-
-{{define "page-content"}}
+{{define "partials/crew_import_results.html"}}
 <div class="grid grid-cols-1 sm:grid-cols-3 gap-4 mb-6">
    <div class="stat bg-base-100 shadow-sm rounded-lg">
        <div class="stat-title">Added</div>
@@ -63,6 +53,6 @@

 <div class="flex gap-3 mt-6">
    <a href="/admin#crew" class="btn btn-primary">Done</a>
-    <a href="/admin/crew/import" class="btn btn-ghost">Import Another</a>
+    <a href="/admin/crew/import" class="btn btn-ghost">Import another</a>
 </div>
 {{end}}
@@ -0,0 +1,22 @@
+{{define "partials/job_progress.html"}}
+<div hx-get="/admin/api/jobs/{{.Key}}/status"
+     hx-trigger="load delay:1s"
+     hx-swap="outerHTML"
+     class="alert alert-info">
+    <span class="loading loading-spinner loading-sm" aria-hidden="true"></span>
+    <div class="text-sm w-full">
+        <p class="font-medium">{{ .Title }}...</p>
+        {{ if .Progress.Message }}
+        <p class="text-base-content/70 mt-1">
+            {{ .Progress.Message }}
+            {{ if .Progress.Total }}<span class="font-mono">({{ .Progress.Done }}/{{ .Progress.Total }})</span>{{ else if .Progress.Done }}<span class="font-mono">({{ .Progress.Done }})</span>{{ end }}
+        </p>
+        {{ else }}
+        <p class="text-base-content/70 mt-1">Starting...</p>
+        {{ end }}
+        {{ if .Progress.Total }}
+        <progress class="progress progress-info w-full mt-2" value="{{ .Progress.Done }}" max="{{ .Progress.Total }}"></progress>
+        {{ end }}
+    </div>
+</div>
+{{end}}
@@ -0,0 +1,15 @@
+{{define "partials/job_result.html"}}
+<div class="alert {{ if .Failed }}alert-warning{{ else }}alert-success{{ end }}" role="status">
+    {{ if .Failed }}{{ icon "triangle-alert" "size-5" }}{{ else }}{{ icon "check-circle" "size-5 shrink-0" }}{{ end }}
+    <span>{{ .Message }}</span>
+    {{ if .ReloadURL }}
+    <button type="button" class="btn btn-sm btn-ghost"
+            hx-get="{{ .ReloadURL }}"
+            hx-target="{{ .ReloadTarget }}"
+            hx-swap="innerHTML">
+        {{ icon "refresh-cw" "size-4" }}
+        Reload
+    </button>
+    {{ end }}
+</div>
+{{end}}
@@ -1,20 +0,0 @@
-{{define "partials/scan_backfill_progress.html"}}
-<div hx-get="/admin/api/scan-backfill/status"
-     hx-trigger="load delay:1s"
-     hx-swap="outerHTML"
-     class="alert alert-info">
-    <span class="loading loading-spinner loading-sm" aria-hidden="true"></span>
-    <div class="text-sm">
-        <p class="font-medium">Backfilling scan records...</p>
-        {{ if .Current }}
-        <p class="text-base-content/70 mt-1">
-            Scanned <span class="font-mono">{{ .Current.Scanned }}</span> records ·
-            rewrites: <span class="font-mono">{{ .Current.Rewritten }}</span>
-            ({{ .Current.MarkedSkipped }} skipped, {{ .Current.MarkedFailed }} failed)
-        </p>
-        {{ else }}
-        <p class="text-base-content/70 mt-1">Starting...</p>
-        {{ end }}
-    </div>
-</div>
-{{end}}
@@ -48,7 +48,10 @@
                        <td><code class="text-sm">{{.Name}}</code></td>
                        <td class="text-right tabular-nums">{{.Count}}</td>
                        <td colspan="2">
-                            <form action="/admin/crew/remap-tier" method="POST" class="flex gap-2 items-center">
+                            <form hx-post="/admin/crew/remap-tier"
+                                  hx-target="#remap-status-{{.Name}}"
+                                  hx-swap="innerHTML"
+                                  class="flex gap-2 items-center">
                                {{ csrfInput $.CSRFToken }}
                                <input type="hidden" name="from" value="{{.Name}}">
                                <select name="to" required class="select select-bordered select-sm w-full max-w-xs">
@@ -59,6 +62,7 @@
                                </select>
                                <button type="submit" class="btn btn-warning btn-sm">Apply</button>
                            </form>
+                            <div id="remap-status-{{.Name}}" class="mt-2"></div>
                        </td>
                    </tr>
                    {{end}}