Facility data hygiene: merge duplicates, drop junk-named facilities

Cleans up the crawl-generated facility table that surfaced garbage on /Facilities
(«بیمارستان هستم», «... از مدجابز», bare «کلینیک», «سازمان برنامه جنوبی» x3):

- FacilityMatcher.IsJunkName: shared detector for non-names — bare type words, cores
  made only of filler/verb tokens, and leaked crawl-source/placeholder text. Added
  داروخانه/آسایشگاه to the generic type words so bare ones are caught and dedupe better.
- HeuristicListingParser.ExtractFacilityName now rejects junk candidates (and emoji), so
  new ingests fall back to the shared placeholder instead of forging a fake facility.
- IngestionService.MergeAndCleanFacilitiesAsync (+ admin button): folds junk facilities
  into the placeholder and merges Persian-fuzzy duplicates into one keeper, repointing
  their shifts/jobs first. Hard guard: only purely crawl-generated, unmanaged facilities
  are removed — employer-owned and verified facilities are never touched.

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

src/JobsMedical.Web/Pages/Admin/Index.cshtml

@@ -76,6 +76,15 @@
                 فقط آگهی‌هایی که با صافیِ فعلی «خارج از حوزه» تشخیص داده می‌شوند (نه صرفاً ناقص) و استخدام‌های تکراری بایگانی می‌شوند (وضعیت «بایگانی»، نه حذف). آگهی‌های معتبر دست‌نخورده‌اند، پس آدرسِ ایندکس‌شده‌شان تغییر نمی‌کند؛ صفحهٔ موارد بایگانی‌شده ۴۱۰ Gone می‌دهد تا گوگل تمیز حذفشان کند.
             </p>
 
+            <form method="post" onsubmit="return confirm('مراکز درمانیِ تکراری ادغام و مراکزِ بی‌نام/نامعتبر (مثل «بیمارستان هستم» یا «از مدجابز») حذف می‌شوند؛ آگهی‌هایشان به مرکزِ معتبر یا «نامشخص» منتقل می‌شود. مراکزِ ثبت‌شده توسط کارفرما یا تأییدشده دست‌نخورده می‌مانند. ادامه؟');">
+                <button type="submit" asp-page-handler="CleanFacilities" class="btn btn-primary btn-block" style="margin-top:10px;">
+                    🏥 ادغام مراکز تکراری و حذف مراکز بی‌نام
+                </button>
+            </form>
+            <p class="muted" style="font-size:11px; margin:6px 0 0;">
+                مراکز تکراری (با تطبیقِ فارسی) در یک رکورد ادغام و مراکزِ بدونِ نامِ واقعی به «نامشخص» منتقل می‌شوند. آگهی‌ها حفظ می‌شوند؛ فقط مراکزِ جمع‌آوری‌شده و مدیریت‌نشده پاک می‌شوند.
+            </p>
+
             <hr style="border:none; border-top:1px solid var(--line); margin:16px 0;" />
 
             <h3>افزودن دستی</h3>

src/JobsMedical.Web/Pages/Admin/Index.cshtml.cs

@@ -145,6 +145,18 @@ public class IndexModel : PageModel
         return RedirectToPage();
     }
 
+    /// <summary>
+    /// Clean up the crawl-generated facility table: merge Persian-fuzzy duplicate facilities and fold
+    /// junk-named ones («بیمارستان هستم»، «... از مدجابز»، bare «کلینیک») into the shared placeholder,
+    /// repointing their listings first. Employer-owned / verified facilities are never touched.
+    /// </summary>
+    public async Task<IActionResult> OnPostCleanFacilitiesAsync()
+    {
+        var (merged, cleaned) = await _ingest.MergeAndCleanFacilitiesAsync();
+        IngestMessage = $"پاک‌سازی مراکز: {merged} مرکزِ تکراری ادغام و {cleaned} مرکزِ بی‌نام/نامعتبر حذف شد (آگهی‌هایشان به مرکزِ معتبر یا «نامشخص» منتقل شد). مراکز ثبت‌شده توسط کارفرما/تأییدشده دست‌نخورده ماند.";
+        return RedirectToPage();
+    }
+
     private async Task LoadAsync()
     {
         Queue = await _db.RawListings

src/JobsMedical.Web/Services/ListingParser.cs

@@ -218,12 +218,17 @@ public class HeuristicListingParser : IListingParser
             {
                 if (NameStops.Contains(w)) break;
                 if (Regex.IsMatch(w, @"\d")) break;        // numbers/phones aren't names
+                if (!w.Any(char.IsLetter)) break;          // emoji / punctuation («📍») isn't a name
                 if (w.Length == 1) break;                  // stray letters
                 picked.Add(w);
                 if (picked.Count >= 3) break;
             }
             if (picked.Count == 0) continue;               // bare keyword (e.g. just «بیمارستان») isn't useful
-            return (kw + " " + string.Join(" ", picked)).Trim();
+            var candidate = (kw + " " + string.Join(" ", picked)).Trim();
+            // Reject names that are only filler/verb/source noise («بیمارستان هستم», «... از مدجابز») —
+            // a real name couldn't be extracted, so fall back to the shared placeholder downstream.
+            if (Scraping.FacilityMatcher.IsJunkName(candidate)) continue;
+            return candidate;
         }
         return null;
     }

src/JobsMedical.Web/Services/Scraping/FacilityMatcher.cs

@@ -17,7 +17,7 @@ public static class FacilityMatcher
     {
         "بیمارستان", "زایشگاه", "پلی کلینیک", "پلیکلینیک", "درمانگاه", "کلینیک",
         "مرکز درمانی", "مرکز جراحی", "مجتمع پزشکی", "مجتمع درمانی", "مرکز", "مجتمع",
-        "آزمایشگاه", "مطب", "تخصصی", "فوق تخصصی", "فوقتخصصی", "عمومی", "دکتر", "دی کلینیک",
+        "آزمایشگاه", "داروخانه", "آسایشگاه", "مطب", "تخصصی", "فوق تخصصی", "فوقتخصصی", "عمومی", "دکتر", "دی کلینیک",
     };
 
     /// <summary>Lower-cased, Arabic→Persian folded, punctuation-stripped, whitespace-collapsed.</summary>
@@ -47,6 +47,42 @@ public static class FacilityMatcher
         return Regex.Replace(n, @"\s+", " ").Trim();
     }
 
+    // Filler/verb/locator tokens that are never a real facility name — the parser sweeps these in
+    // when an ad has no named facility («بیمارستان هستم», «مطب نیازمندیم سه», «کلینیک های فقط منطقه»).
+    private static readonly string[] JunkCoreWords =
+    {
+        "هستم", "هستیم", "هستش", "میشوم", "میشم", "بشوم", "میباشد", "باشد", "میباشم",
+        "نیازمندیم", "نیازمند", "نیازمندم", "داریم", "دارم", "میخواهیم", "میخوام",
+        "حتی", "تعدادی", "فقط", "منطقه", "واقع", "های", "مبتدی", "محترم", "خوب",
+        "سه", "دو", "یک", "چند", "این", "آن", "همکار", "نیرو",
+    };
+
+    // Crawl-source names that must never appear as a public facility («مرکز درمانی (از مدجابز)»),
+    // plus the shared placeholder text.
+    private static readonly string[] SourceMarkers =
+    {
+        "مدجابز", "مدجاب", "از تلگرام", "از دیوار", "از بله", "از کانال", "ثبت نشده", "نامشخص",
+    };
+
+    /// <summary>
+    /// True when a name is NOT a usable facility name: a bare type word («بیمارستان»), a name whose
+    /// distinctive core is only filler/verb tokens («بیمارستان هستم» → «هستم»), or a leaked crawl
+    /// source / placeholder («... از مدجابز», «نامشخص»). Such an ad has no real named facility and
+    /// should fall back to the shared placeholder instead of forging a fake one.
+    /// </summary>
+    public static bool IsJunkName(string? name)
+    {
+        var normalized = Normalize(name);
+        if (normalized.Length == 0) return true;
+        if (SourceMarkers.Any(m => normalized.Contains(Normalize(m)))) return true;
+
+        var core = Core(name);
+        if (core.Length == 0) return true;   // bare type word only («بیمارستان»، «کلینیک»)
+
+        var tokens = core.Split(' ', StringSplitOptions.RemoveEmptyEntries);
+        return tokens.All(t => t.Length <= 1 || JunkCoreWords.Contains(t));
+    }
+
     /// <summary>True when two names almost certainly denote the same facility.</summary>
     public static bool IsSame(string? a, string? b)
     {

src/JobsMedical.Web/Services/Scraping/IngestionService.cs

@@ -418,6 +418,79 @@ public class IngestionService
         return archived;
     }
 
+    /// <summary>
+    /// Clean up the crawl-generated facility table: (1) fold listings of junk-named facilities
+    /// («بیمارستان هستم», «... از مدجابز», bare «کلینیک») into the shared placeholder and delete the
+    /// junk record; (2) merge Persian-fuzzy duplicates («سازمان برنامه جنوبی» ×3) into one keeper,
+    /// repointing their shifts/jobs. HARD GUARD: only ever removes facilities that are purely
+    /// crawl-generated (no owner, not verified, Unverified) and never the placeholder — employer- and
+    /// admin-managed facilities are untouched. Listings are always repointed first, so no ad is lost.
+    /// Returns (merged, cleaned).
+    /// </summary>
+    public async Task<(int merged, int cleaned)> MergeAndCleanFacilitiesAsync(CancellationToken ct = default)
+    {
+        var facilities = await _db.Facilities.ToListAsync(ct);
+        var placeholder = facilities.FirstOrDefault(f => f.Name == UnknownFacilityName);
+
+        var jobCounts = await _db.JobOpenings.GroupBy(j => j.FacilityId)
+            .Select(g => new { g.Key, C = g.Count() }).ToDictionaryAsync(x => x.Key, x => x.C, ct);
+        var shiftCounts = await _db.Shifts.GroupBy(s => s.FacilityId)
+            .Select(g => new { g.Key, C = g.Count() }).ToDictionaryAsync(x => x.Key, x => x.C, ct);
+        int Listings(int id) => jobCounts.GetValueOrDefault(id) + shiftCounts.GetValueOrDefault(id);
+
+        // Removable = purely crawl-generated and unmanaged. Never the placeholder, an owned, or a
+        // verified facility (those carry real employer data / verification).
+        bool Removable(Facility f) => f.OwnerUserId is null && !f.IsVerified
+            && f.Verification == VerificationStatus.Unverified
+            && (placeholder is null || f.Id != placeholder.Id);
+
+        async Task AbsorbAsync(int fromId, int toId)
+        {
+            await _db.Shifts.Where(s => s.FacilityId == fromId)
+                .ExecuteUpdateAsync(u => u.SetProperty(s => s.FacilityId, toId), ct);
+            await _db.JobOpenings.Where(j => j.FacilityId == fromId)
+                .ExecuteUpdateAsync(u => u.SetProperty(j => j.FacilityId, toId), ct);
+            await _db.Facilities.Where(f => f.Id == fromId).ExecuteDeleteAsync(ct);   // cascades stray docs/reviews
+        }
+
+        int merged = 0, cleaned = 0;
+
+        // 1) Junk-named crawl facilities → fold into the shared placeholder.
+        if (placeholder is not null)
+            foreach (var f in facilities.Where(f => Removable(f) && FacilityMatcher.IsJunkName(f.Name)).ToList())
+            {
+                await AbsorbAsync(f.Id, placeholder.Id);
+                cleaned++;
+            }
+
+        // 2) Merge same-city Persian-fuzzy duplicates into the best keeper.
+        var remaining = await _db.Facilities.Where(f => f.Name != UnknownFacilityName).ToListAsync(ct);
+        var done = new HashSet<int>();
+        foreach (var f in remaining)
+        {
+            if (done.Contains(f.Id)) continue;
+            done.Add(f.Id);
+            var cluster = remaining.Where(o => o.Id != f.Id && !done.Contains(o.Id)
+                && o.CityId == f.CityId && FacilityMatcher.IsSame(o.Name, f.Name)).ToList();
+            if (cluster.Count == 0) continue;
+            cluster.Add(f);
+            // keeper: verified > owned > most listings > lowest id (oldest).
+            var keeper = cluster.OrderByDescending(x => x.IsVerified)
+                .ThenByDescending(x => x.OwnerUserId.HasValue)
+                .ThenByDescending(x => Listings(x.Id)).ThenBy(x => x.Id).First();
+            foreach (var dup in cluster.Where(x => x.Id != keeper.Id))
+            {
+                done.Add(dup.Id);
+                if (!Removable(dup)) continue;   // never delete an employer/verified facility
+                await AbsorbAsync(dup.Id, keeper.Id);
+                merged++;
+            }
+        }
+
+        _log.LogInformation("Facility cleanup: merged {M} duplicates, removed {C} junk facilities.", merged, cleaned);
+        return (merged, cleaned);
+    }
+
     private static string DigitsOnly(string s) => new(HtmlUtil.ToLatinDigits(s).Where(char.IsDigit).ToArray());
 
     private static (RawListingStatus status, string? reason, int confidence) Decide(