Merge pull request 'Release v26.05.25.3 — ML embedder SigLIP fix, import-UX, extension publish' (#13) from dev into main

backend/app/api/import_admin.py

@@ -47,6 +47,8 @@ async def status():
         if active:
             payload["active_batch"] = {
                 "id": active.id,
+                "source_path": active.source_path,
+                "scan_mode": active.scan_mode,
                 "total_files": active.total_files,
                 "imported": active.imported,
                 "skipped": active.skipped,

backend/app/services/ml/embedder.py

@@ -34,10 +34,21 @@ class Embedder:
         if self._model is not None:
             return
         import torch
-        from transformers import AutoModel, AutoProcessor
+        from transformers import AutoModel, SiglipImageProcessor
 
         self._torch = torch
-        self._processor = AutoProcessor.from_pretrained(str(self._model_dir))
+        # FC's embedder only does IMAGE inference — never text. AutoProcessor
+        # loads the full processor including SiglipTokenizer, which requires
+        # the sentencepiece library at import time even if we never call it.
+        # SiglipImageProcessor loads ONLY preprocessor_config.json (image
+        # side) and skips the tokenizer config entirely. Operator hit the
+        # ImportError 2026-05-25 once the ml-worker started actually running
+        # tag_and_embed; switching to the image-only loader avoids the
+        # tokenizer dep without adding ~30 MB of unused C++ build to the
+        # lean ml-worker image.
+        self._processor = SiglipImageProcessor.from_pretrained(
+            str(self._model_dir)
+        )
         self._model = AutoModel.from_pretrained(str(self._model_dir))
         self._model.eval()
 

extension/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "FabledCurator",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "description": "Export cookies from supported platforms to FabledCurator and add creators as sources in one click.",
 
   "browser_specific_settings": {

extension/package.json

@@ -1,6 +1,6 @@
 {
   "name": "fabledcurator-extension",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "private": true,
   "description": "Firefox extension for FabledCurator",
   "scripts": {

frontend/src/components/settings/ImportTriggerPanel.vue

@@ -7,7 +7,8 @@
           indeterminate color="accent" size="20"
         />
         <span>
-          Scanning {{ store.activeBatch.source_path }} —
+          {{ store.activeBatch.scan_mode === 'deep' ? 'Deep scanning' : 'Scanning' }}
+          {{ store.activeBatch.source_path || '/import' }} —
           imported {{ store.activeBatch.imported }},
           skipped {{ store.activeBatch.skipped }},
           failed {{ store.activeBatch.failed }} /
@@ -24,8 +25,12 @@
 
       <p class="text-body-2 mb-3">
         <span v-if="!store.activeBatch">
-          Run a quick scan of the import directory. Deep scan (pHash dedup,
-          archives) lands in FC-2d.
+          <strong>Quick scan</strong> walks <code>/import</code> and enqueues
+          new files only (skips paths already on a non-failed ImportTask).
+          <strong>Deep scan</strong> additionally chains a phash backfill
+          across the existing library — use after bulk-imports to catch
+          near-duplicates that slipped through. Both modes route non-media
+          + sidecar pairs through PostAttachment capture.
         </span>
         <span v-else>
           An active batch is in progress. Wait for it to finish, or click
@@ -34,15 +39,26 @@
         </span>
       </p>
 
-      <v-btn
-        color="primary" rounded="pill"
-        :disabled="!!store.activeBatch"
-        :loading="busy"
-        @click="trigger"
-      >
-        <v-icon start>mdi-magnify-scan</v-icon>
-        Quick scan
-      </v-btn>
+      <div class="d-flex flex-wrap" style="gap: 12px;">
+        <v-btn
+          color="primary" rounded="pill"
+          :disabled="!!store.activeBatch"
+          :loading="busy === 'quick'"
+          @click="trigger('quick')"
+        >
+          <v-icon start>mdi-magnify-scan</v-icon>
+          Quick scan
+        </v-btn>
+        <v-btn
+          color="secondary" rounded="pill" variant="tonal"
+          :disabled="!!store.activeBatch"
+          :loading="busy === 'deep'"
+          @click="trigger('deep')"
+        >
+          <v-icon start>mdi-magnify-plus-outline</v-icon>
+          Deep scan
+        </v-btn>
+      </div>
 
       <v-alert v-if="store.triggerError" type="error" variant="tonal" class="mt-3" closable>
         {{ store.triggerError }}
@@ -56,12 +72,12 @@ import { ref } from 'vue'
 import { useImportStore } from '../../stores/import.js'
 
 const store = useImportStore()
-const busy = ref(false)
+const busy = ref(null)
 const clearing = ref(false)
 
-async function trigger() {
-  busy.value = true
-  try { await store.triggerScan() } catch {} finally { busy.value = false }
+async function trigger(mode) {
+  busy.value = mode
+  try { await store.triggerScan(mode) } catch {} finally { busy.value = null }
 }
 
 async function onClearStuck() {

frontend/src/stores/import.js

@@ -52,25 +52,31 @@ export const useImportStore = defineStore('import', () => {
     }
   }
 
-  async function triggerScan() {
+  async function triggerScan(mode = 'quick') {
+    if (!['quick', 'deep', 'verify'].includes(mode)) {
+      throw new Error(`unsupported scan mode: ${mode}`)
+    }
     triggerError.value = null
     try {
-      await api.post('/api/import/trigger', { body: { mode: 'quick' } })
+      await api.post('/api/import/trigger', { body: { mode } })
       // Acknowledge immediately so the click isn't invisible. scan_directory
       // can finalize the batch synchronously when every file in /import is
       // already on a non-failed ImportTask (operator-flagged 2026-05-25:
       // 233k existing tasks → all paths in skip-set → files_seen=0 →
       // batch flashes 'running' for <100ms then 'complete' before the
       // first refreshStatus() lands; UI never sees the active state).
-      window.__fcToast?.({ text: 'Scan triggered', type: 'success' })
+      const label = mode === 'deep'
+        ? 'Deep scan triggered (pHash backfill chained)'
+        : mode === 'verify' ? 'Library verify triggered' : 'Quick scan triggered'
+      window.__fcToast?.({ text: label, type: 'success' })
       await refreshStatus()
       // Re-poll twice over ~5s to catch quick-finalize transitions and
-      // surface a result toast either way.
+      // surface a result toast either way. Skip the "no new files" hint
+      // for 'verify' since it doesn't walk the import_root.
       setTimeout(async () => {
         await refreshStatus()
-        if (!activeBatch.value) {
-          // Either scan completed with zero new files, or it never visibly
-          // started. Fetch the freshest task to differentiate.
+        await loadTasks(true)
+        if (!activeBatch.value && mode !== 'verify') {
           window.__fcToast?.({
             text: 'Scan complete — no new files (everything already on an ImportTask row)',
             type: 'info',

frontend/src/views/SettingsView.vue

@@ -1,6 +1,16 @@
 <template>
   <v-container fluid class="py-6">
-    <v-tabs v-model="tab" color="accent" class="mb-4">
+    <!-- Sticky tabs: operator-flagged 2026-05-25 — long Import / Maintenance
+         panels pushed the tab strip out of the viewport, forcing a scroll-
+         to-top just to change tab. AppShell's TopNav is 64px sticky, so the
+         tab strip lives directly under it. Background uses the theme surface
+         token so it visually merges with the page rather than the
+         translucent v-tabs default. -->
+    <v-tabs
+      v-model="tab" color="accent" class="mb-4"
+      style="position: sticky; top: 64px; z-index: 4;
+             background: rgb(var(--v-theme-surface));"
+    >
       <v-tab value="overview">Overview</v-tab>
       <v-tab value="activity">Activity</v-tab>
       <v-tab value="import">Import</v-tab>
@@ -28,11 +38,14 @@
       </v-window-item>
 
       <v-window-item value="import">
+        <!-- Order: trigger → recent tasks → filters. Tasks sit directly
+             below the trigger so operator sees hit/miss feedback without
+             scrolling past the filter card (operator-flagged 2026-05-25). -->
         <ImportTriggerPanel />
         <v-divider class="my-6" />
-        <ImportFiltersForm />
-        <v-divider class="my-6" />
         <ImportTaskList />
+        <v-divider class="my-6" />
+        <ImportFiltersForm />
       </v-window-item>
 
       <v-window-item value="maintenance">