Sources: LLM deep-preview, source search, duplicate-add guard

Three admin Sources upgrades:
- Deep preview: a per-candidate "🔬 Deep preview" button runs the REAL
  classifier on an 8-item sample (the same model that judges live articles),
  versus the fast keyword heuristic the add/Re-preview path uses. Preview now
  carries `classified`, surfaced as a "model-checked" vs "quick estimate"
  badge — so the acceptance % is no longer ambiguously heuristic. conn is
  released during the ~30-60s model pass; postJSON has no client timeout.
- Search: free-text box over the sources table (name / category / feed URL /
  homepage), folded into the existing status filter, with a live match count
  and empty state. Makes "is this already added?" a glance.
- Duplicate-add guard: sources.find_existing_feed() + feed_key() normalize
  scheme/www/trailing-slash/case, so re-adding a feed that's already a live
  source or a queued candidate is refused with a 409 naming where it lives
  (DB already enforced exact-URL uniqueness; this catches the near-miss
  variants and overwrite-on-promote footgun).

Tests: test_candidate_deep_preview_and_dedup (deep flag wires the model +
uses the small sample; exact/www/slash/case variants all 409). 224 pytest +
11 vitest green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

frontend/src/routes/admin/+page.svelte

@@ -192,15 +192,21 @@
   let paused = $derived(sources.filter((s) => sstatus(s) === 'paused').length);
   let retired = $derived(sources.filter((s) => sstatus(s) === 'retired').length);
 
-  // Sources table filter.
+  // Sources table filter + free-text search (name / category / feed URL).
   let srcFilter = $state('all');
+  let srcSearch = $state('');
   let shownSources = $derived(
-    srcFilter === 'healthy' ? sources.filter((s) => sstatus(s) === 'active' && !s.failures && !s.review_flag)
+    (srcFilter === 'healthy' ? sources.filter((s) => sstatus(s) === 'active' && !s.failures && !s.review_flag)
       : srcFilter === 'resting' ? sources.filter((s) => sstatus(s) === 'active' && s.failures > 0)
         : srcFilter === 'flagged' ? sources.filter((s) => s.review_flag)
           : srcFilter === 'paused' ? sources.filter((s) => sstatus(s) === 'paused')
             : srcFilter === 'retired' ? sources.filter((s) => sstatus(s) === 'retired')
               : sources
+    ).filter((s) => {
+      const q = srcSearch.trim().toLowerCase();
+      if (!q) return true;
+      return `${s.name} ${s.category || ''} ${s.feed_url || ''} ${s.homepage_url || ''}`.toLowerCase().includes(q);
+    })
   );
 
   // Lifecycle (status: active/paused/retired) keeps `active` mirrored server-side;
@@ -276,6 +282,13 @@
     try { Object.assign(c, await postJSON(`/api/admin/candidates/${c.id}/preview`)); }
     catch (e) { c._err = e?.message || 'Re-preview failed.'; }
   }
+  // Deep preview runs the REAL classifier on a small sample (~30-60s) — the
+  // model's true acceptance view, not the fast heuristic estimate.
+  async function deepPreview(c) {
+    c._err = ''; c._deep = true;
+    try { Object.assign(c, await postJSON(`/api/admin/candidates/${c.id}/preview?deep=true`), { _deep: false }); }
+    catch (e) { c._err = e?.message || 'Deep preview failed.'; c._deep = false; }
+  }
   async function promoteCandidate(c) {
     c._err = '';
     try {
@@ -560,6 +573,7 @@
                 {#if c.preview}
                   <div class="cprev">
                     {c.preview.accepted ?? 0}/{c.preview.sampled ?? 0} sampled would pass{#if c.preview.acceptance_rate != null} · {Math.round(c.preview.acceptance_rate * 100)}% accept{/if}{#if c.preview.recent_7d != null} · {c.preview.recent_7d} in last 7d{/if}
+                    {#if c.preview.classified}<span class="vbadge model" title="Scored by the real classifier — the true acceptance view">model-checked</span>{:else}<span class="vbadge fast" title="Fast keyword heuristic — an estimate. Run Deep preview for the model's real verdict.">quick estimate</span>{/if}
                     {#if c.preview.examples_accepted?.length}<div class="cex">e.g. {c.preview.examples_accepted.slice(0, 3).join(' · ')}</div>{/if}
                   </div>
                 {/if}
@@ -569,6 +583,7 @@
                   <label class="cchk"><input type="checkbox" bind:checked={c._activate} /> Activate immediately</label>
                   <button class="csend" onclick={() => promoteCandidate(c)}>Promote{c._activate ? '' : ' as paused'}</button>
                   <button class="act" onclick={() => repreviewCandidate(c)}>Re-preview</button>
+                  <button class="act" title="Run the real model on a sample (~30-60s)" onclick={() => deepPreview(c)} disabled={c._deep}>{c._deep ? 'Deep-checking…' : '🔬 Deep preview'}</button>
                   <button class="act del" onclick={() => rejectCandidate(c)}>Reject</button>
                 </div>
               </li>
@@ -579,10 +594,16 @@
 
       <h2>Sources <a class="exportlink" href="/api/admin/export/sources.csv" download>export CSV ↓</a></h2>
       <p class="sub2">{healthy} healthy · {resting} resting · {flagged} flagged · {paused} paused · {retired} retired · {sources.length} total</p>
-      <div class="filterchips">
-        {#each [['all', 'All'], ['healthy', 'Healthy'], ['resting', 'Resting'], ['flagged', 'Flagged'], ['paused', 'Paused'], ['retired', 'Retired']] as [key, label] (key)}
-          <button class="chip" class:on={srcFilter === key} onclick={() => (srcFilter = key)}>{label}</button>
-        {/each}
+      <div class="srctools">
+        <div class="filterchips">
+          {#each [['all', 'All'], ['healthy', 'Healthy'], ['resting', 'Resting'], ['flagged', 'Flagged'], ['paused', 'Paused'], ['retired', 'Retired']] as [key, label] (key)}
+            <button class="chip" class:on={srcFilter === key} onclick={() => (srcFilter = key)}>{label}</button>
+          {/each}
+        </div>
+        <div class="srcsearch">
+          <input type="search" placeholder="Search name, category, or URL…" bind:value={srcSearch} autocapitalize="off" autocomplete="off" spellcheck="false" />
+          {#if srcSearch.trim()}<span class="srccount">{shownSources.length} match{shownSources.length === 1 ? '' : 'es'}</span>{/if}
+        </div>
       </div>
       <div class="tablewrap">
         <table class="srctable">
@@ -649,6 +670,8 @@
                   </td>
                 </tr>
               {/if}
+            {:else}
+              <tr><td colspan="10" class="srcempty">{srcSearch.trim() ? `No sources match “${srcSearch.trim()}”.` : 'No sources in this view.'}</td></tr>
             {/each}
           </tbody>
         </table>
@@ -1056,6 +1079,17 @@
   }
   .filterchips .chip:hover { border-color: var(--accent); }
   .filterchips .chip.on { background: var(--accent); border-color: var(--accent); color: #fff; }
+  /* Sources filter row + search */
+  .srctools { display: flex; align-items: center; justify-content: space-between; gap: 12px; flex-wrap: wrap; margin: 0 0 14px; }
+  .srctools .filterchips { margin: 0; }
+  .srcsearch { display: inline-flex; align-items: center; gap: 8px; }
+  .srcsearch input {
+    font: inherit; font-size: 0.84rem; padding: 6px 12px; border: 1px solid var(--line);
+    border-radius: 999px; background: var(--surface); color: var(--ink); width: 230px;
+  }
+  .srcsearch input:focus { outline: none; border-color: var(--accent); }
+  .srccount { font-size: 0.78rem; color: var(--muted); white-space: nowrap; }
+  .srcempty { text-align: center; color: var(--muted); font-style: italic; padding: 22px 10px; }
 
   /* Add a source + candidate queue */
   .addsrc { background: var(--surface); border: 1px solid var(--line); border-radius: 14px; padding: 14px 16px; margin-bottom: 6px; }
@@ -1070,6 +1104,11 @@
   .curl { font-size: 0.76rem; color: var(--muted); word-break: break-all; margin-top: 2px; }
   .cprev { font-size: 0.84rem; color: var(--ink); margin-top: 7px; }
   .cprev .cex { color: var(--muted); font-size: 0.8rem; margin-top: 2px; font-style: italic; }
+  /* Heuristic-vs-model preview badge */
+  .vbadge { display: inline-block; margin-left: 8px; padding: 1px 8px; border-radius: 999px;
+            font-size: 0.68rem; font-weight: 600; text-transform: uppercase; letter-spacing: 0.04em; cursor: help; }
+  .vbadge.model { background: #e3efe4; color: #3f7048; }
+  .vbadge.fast { background: var(--line); color: var(--muted); }
   .cactions { display: flex; gap: 9px; align-items: center; flex-wrap: wrap; margin-top: 10px; }
   .cactions .ccat { font: inherit; font-size: 0.8rem; padding: 5px 9px; border: 1px solid var(--line); border-radius: 8px; background: var(--bg); color: var(--ink); width: 150px; }
   .cactions .cchk { font-size: 0.8rem; color: var(--muted); display: inline-flex; align-items: center; gap: 5px; }

goodnews/api.py

@@ -1143,10 +1143,20 @@ def create_app() -> FastAPI:
             rows = sources.list_candidates(conn)
         return [_candidate_dict(r) for r in rows]
 
-    def _preview_or_502(url: str) -> dict:
-        # SSRF-safe fetch (admin-pasted URL is untrusted); heuristic-only (fast).
+    def _preview_or_502(url: str, deep: bool = False) -> dict:
+        # SSRF-safe fetch (admin-pasted URL is untrusted). Default is the fast
+        # heuristic; deep=True also runs the real LLM classifier on a small sample
+        # (slower, ~5-7s/item — the true acceptance view, not an estimate).
+        client = None
+        if deep:
+            try:
+                client = LocalModelClient.from_env()
+            except Exception:  # noqa: BLE001 — fall back to heuristic if the model is down
+                client = None
         try:
-            return feeds.preview_feed(url, sample=20, fetcher=feeds.safe_fetch_feed)
+            return feeds.preview_feed(
+                url, sample=(8 if deep else 20), fetcher=feeds.safe_fetch_feed, client=client
+            )
         except Exception as exc:  # noqa: BLE001 — surface a readable reason
             raise HTTPException(status_code=502, detail=f"Couldn't preview that feed: {exc}")
 
@@ -1157,6 +1167,16 @@ def create_app() -> FastAPI:
             raise HTTPException(status_code=422, detail="feed_url is required")
         with get_conn() as conn:  # gate BEFORE the outbound fetch
             _require_admin(conn, request)
+            # Don't re-add a feed that's already live or already queued (catches
+            # http/https · www · trailing-slash variants, not just exact dups).
+            existing = sources.find_existing_feed(conn, url)
+            if existing:
+                where = "already a source" if existing["kind"] == "source" else "already in the candidate queue"
+                raise HTTPException(
+                    status_code=409,
+                    detail=f"“{existing['name']}” is {where} ({existing['status']}). "
+                           "Find it below — Re-preview it there if you want a fresh read.",
+                )
         preview = _preview_or_502(url)  # no DB connection held during network I/O
         with get_conn() as conn:
             _require_admin(conn, request)
@@ -1164,14 +1184,14 @@ def create_app() -> FastAPI:
         return _candidate_dict(row)
 
     @app.post("/api/admin/candidates/{cid}/preview")
-    def admin_candidate_repreview(cid: int, request: Request) -> dict:
+    def admin_candidate_repreview(cid: int, request: Request, deep: bool = False) -> dict:
         with get_conn() as conn:
             _require_admin(conn, request)
             cand = conn.execute("SELECT feed_url FROM source_candidates WHERE id = ?", (cid,)).fetchone()
             if not cand:
                 raise HTTPException(status_code=404, detail="candidate not found")
             url = cand["feed_url"]
-        preview = _preview_or_502(url)
+        preview = _preview_or_502(url, deep=deep)  # conn released during the (slow) model pass
         with get_conn() as conn:
             _require_admin(conn, request)
             row = sources.save_candidate(conn, url, preview=preview)

goodnews/sources.py

@@ -67,6 +67,37 @@ def upsert_sources(conn: sqlite3.Connection, source_defs: list[dict]) -> int:
     return count
 
 
+# --- Duplicate detection (catch the same feed added twice) --------------------
+
+
+def feed_key(url: str) -> str:
+    """A loose comparison key for spotting the same feed added twice despite
+    trivial differences (scheme, www, trailing slash, case). Compare-only — the
+    feed_url is always STORED exactly as entered; this just powers dup warnings."""
+    try:
+        p = urlsplit((url or "").strip().lower())
+        host = p.netloc.removeprefix("www.")
+        path = p.path.rstrip("/")
+        return host + path + (("?" + p.query) if p.query else "")
+    except Exception:  # noqa: BLE001 — never let a weird URL break add
+        return (url or "").strip().lower()
+
+
+def find_existing_feed(conn: sqlite3.Connection, url: str) -> dict | None:
+    """Is this feed already a live source or a pending candidate? Matches on the
+    loose key, so http/https + www + trailing-slash variants are all caught."""
+    key = feed_key(url)
+    for r in conn.execute("SELECT id, name, feed_url, status FROM sources"):
+        if feed_key(r["feed_url"]) == key:
+            return {"kind": "source", "id": r["id"], "name": r["name"], "status": r["status"]}
+    for r in conn.execute(
+        "SELECT id, name, feed_url, status FROM source_candidates WHERE status NOT IN ('rejected','promoted')"
+    ):
+        if feed_key(r["feed_url"]) == key:
+            return {"kind": "candidate", "id": r["id"], "name": r["name"] or r["feed_url"], "status": r["status"]}
+    return None
+
+
 # --- Supervised source candidates (staging before the real sources table) ----
 
 

tests/test_admin.py

@@ -141,6 +141,31 @@ def test_candidate_suggest_promote_paused(tmp_path, monkeypatch):
     assert any(s["name"] == "Good Feed" for s in tc.get("/api/admin/stats").json()["sources"])
 
 
+def test_candidate_deep_preview_and_dedup(tmp_path, monkeypatch):
+    app, api = _make(tmp_path, monkeypatch, admin_email="boss@x.com")
+
+    def fake_preview(url, **k):
+        # Echo back whether the LLM client was wired in + the sample size used.
+        return {"url": url, "sampled": k.get("sample"), "accepted": 4,
+                "classified": k.get("client") is not None}
+    monkeypatch.setattr(api.feeds, "preview_feed", fake_preview)
+    # Deep preview builds a model client; stub it so we never touch the real LAN model.
+    monkeypatch.setattr(api, "LocalModelClient", type("C", (), {"from_env": staticmethod(lambda: object())}))
+    tc = _signin(app, api, "boss@x.com")
+
+    cand = tc.post("/api/admin/candidates", json={"feed_url": "https://news.test/feed"}).json()
+    assert cand["preview"]["classified"] is False  # add uses the fast heuristic
+    # Deep preview runs the real classifier on the smaller sample.
+    deep = tc.post(f"/api/admin/candidates/{cand['id']}/preview?deep=true").json()
+    assert deep["preview"]["classified"] is True and deep["preview"]["sampled"] == 8
+    # Dedup: exact + trivial variants (scheme / www / trailing slash / case) are refused.
+    assert tc.post("/api/admin/candidates", json={"feed_url": "https://news.test/feed"}).status_code == 409
+    assert tc.post("/api/admin/candidates", json={"feed_url": "http://www.news.test/feed/"}).status_code == 409
+    # Once promoted to a live source, re-adding is still refused.
+    tc.post(f"/api/admin/candidates/{cand['id']}/promote", json={})
+    assert tc.post("/api/admin/candidates", json={"feed_url": "https://NEWS.test/feed"}).status_code == 409
+
+
 def test_candidate_reject_and_gating(tmp_path, monkeypatch):
     app, api = _make(tmp_path, monkeypatch, admin_email="boss@x.com")
     monkeypatch.setattr(api.feeds, "preview_feed", lambda url, **k: {"url": url, "sampled": 1, "accepted": 0})