alembic/versions/0075_tag_is_system.py

@@ -0,0 +1,60 @@
+"""tag.is_system + seed the three hygiene system tags
+
+Training hygiene (operator 2026-07-03, milestone #128): rough WIPs tagged as a
+character poison that character's head and CCIP references; banners/editor
+screenshots pollute whole-image similarity. The fix keys on SYSTEM tags the
+product ships — not operator configuration — so the seed lives here.
+
+Seeding ADOPTS an existing same-(name, kind=general) tag (case-insensitive,
+matching TagService.rename's collision stance) instead of inserting a
+duplicate, so an operator who already tagged `wip` keeps their applications.
+
+Revision ID: 0075
+Revises: 0074
+Create Date: 2026-07-03
+"""
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "0075"
+down_revision: Union[str, None] = "0074"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+SYSTEM_TAG_NAMES = ("wip", "banner", "editor screenshot")
+
+
+def upgrade() -> None:
+    op.add_column(
+        "tag",
+        sa.Column(
+            "is_system", sa.Boolean(), nullable=False,
+            server_default=sa.false(),
+        ),
+    )
+    conn = op.get_bind()
+    for name in SYSTEM_TAG_NAMES:
+        adopted = conn.execute(
+            sa.text(
+                "UPDATE tag SET is_system = true "
+                "WHERE lower(name) = lower(:name) AND kind = 'general'"
+            ),
+            {"name": name},
+        )
+        if adopted.rowcount == 0:
+            conn.execute(
+                sa.text(
+                    "INSERT INTO tag (name, kind, is_system) "
+                    "VALUES (:name, 'general', true)"
+                ),
+                {"name": name},
+            )
+
+
+def downgrade() -> None:
+    # The seeded rows survive as ordinary general tags — dropping the flag is
+    # enough to disarm the mechanism, and deleting rows would orphan any
+    # operator applications made while the flag existed.
+    op.drop_column("tag", "is_system")

backend/app/api/admin.py

@@ -156,6 +156,10 @@ async def tag_delete(tag_id: int):
             )
         except LookupError:
             return _bad("not_found", status=404)
+        except ValueError as exc:
+            # System tags (#128) — the training-hygiene machinery keys on
+            # these rows.
+            return _bad("system_tag", detail=str(exc))
     return jsonify(result)
 
 

backend/app/api/tags.py

@@ -324,6 +324,7 @@ async def get_tag(tag_id: int):
                 "name": tag.name,
                 "kind": tag.kind.value,
                 "fandom_id": tag.fandom_id,
+                "is_system": tag.is_system,
             }
         )
 
@@ -390,6 +391,7 @@ async def update_tag(tag_id: int):
                 "name": tag.name,
                 "kind": tag.kind.value,
                 "fandom_id": tag.fandom_id,
+                "is_system": tag.is_system,
             }
         )
 

backend/app/models/tag.py

@@ -10,6 +10,7 @@ from datetime import datetime
 from enum import StrEnum
 
 from sqlalchemy import (
+    Boolean,
     CheckConstraint,
     Column,
     DateTime,
@@ -17,6 +18,7 @@ from sqlalchemy import (
     Integer,
     String,
     Table,
+    false,
     func,
 )
 from sqlalchemy import (
@@ -41,6 +43,12 @@ class TagKind(StrEnum):
     # to keep historic tag rows queryable.
 
 
+# The seeded system tags (migration 0075). PRESENTATION tags additionally
+# hide from whole-image similarity results — they cluster on UI chrome, not
+# content. `wip` is real art: only the training pipelines exclude it.
+SYSTEM_TAG_NAMES = ("wip", "banner", "editor screenshot")
+PRESENTATION_SYSTEM_TAGS = ("banner", "editor screenshot")
+
 image_tag = Table(
     "image_tag",
     Base.metadata,
@@ -74,6 +82,14 @@ class Tag(Base):
     fandom_id: Mapped[int | None] = mapped_column(
         ForeignKey("tag.id", ondelete="SET NULL"), nullable=True, index=True
     )
+    # System tags ship with FC (wip / banner / editor screenshot, seeded in
+    # migration 0075) and drive the training-hygiene exclusions: images
+    # carrying one are excluded from OTHER concepts' head training and from
+    # CCIP identity references. The mechanism keys on these exact rows, so
+    # they're protected from rename/merge-away/re-fandom in TagService.
+    is_system: Mapped[bool] = mapped_column(
+        Boolean, nullable=False, default=False, server_default=false()
+    )
 
     created_at: Mapped[datetime] = mapped_column(
         DateTime(timezone=True), nullable=False, server_default=func.now()

backend/app/services/cleanup_service.py

@@ -17,7 +17,7 @@ from datetime import UTC, datetime, timedelta
 from pathlib import Path
 from typing import Any
 
-from sqlalchemy import delete, func, or_, select, update
+from sqlalchemy import and_, delete, func, or_, select, update
 from sqlalchemy.orm import Session, aliased
 
 from ..models import (
@@ -203,6 +203,9 @@ def _unused_tag_conditions() -> list:
         Tag.id.not_in(used_via_series),
         Tag.id.not_in(used_via_chapter),
         Tag.id.not_in(used_via_fandom),
+        # System tags (#128) ship with zero applications and must survive a
+        # prune — the training-hygiene machinery keys on the rows.
+        Tag.is_system.is_(False),
     ]
 
 
@@ -402,6 +405,8 @@ def delete_tag(session: Session, *, tag_id: int) -> dict:
     tag = session.get(Tag, tag_id)
     if tag is None:
         raise LookupError(f"tag id not found: {tag_id}")
+    if tag.is_system:
+        raise ValueError(f"'{tag.name}' is a system tag and cannot be deleted")
     associations_count = count_tag_associations(session, tag_id=tag_id)
     info = {"id": tag.id, "name": tag.name, "kind": tag.kind.value}
     session.delete(tag)
@@ -731,7 +736,10 @@ def reset_content_tagging(session: Session, *, dry_run: bool = False) -> dict:
     heads retrain from whatever the operator re-tags. (The API route gates the
     live run behind a preview-derived confirm token for exactly this reason.)
 
-    PRESERVED: fandom + series tags and their series_page ordering. CASCADE on
+    PRESERVED: fandom + series tags and their series_page ordering, AND the
+    system hygiene tags (#128) WITH their applications — the reset re-tags
+    CONTENT concepts, while wip/banner flags describe the file itself and
+    re-flagging hundreds of banners by hand would be pure loss. CASCADE on
     image_tag / tag_alias / tag_suggestion_rejection clears each deleted tag's
     applications + metadata. Tag.fandom_id is SET NULL, so deleting character
     tags never touches the fandom rows. Irreversible except via DB backup
@@ -744,7 +752,9 @@ def reset_content_tagging(session: Session, *, dry_run: bool = False) -> dict:
        "sample_names": [first 50],
        and on live runs "deleted": total}
     """
-    predicate = Tag.kind.in_(RESETTABLE_TAG_KINDS)
+    predicate = and_(
+        Tag.kind.in_(RESETTABLE_TAG_KINDS), Tag.is_system.is_(False)
+    )
     rows = session.execute(
         select(Tag.id, Tag.name, Tag.kind).where(predicate)
     ).all()

backend/app/services/gallery_service.py

@@ -23,7 +23,7 @@ from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.orm import aliased
 
 from ..models import Artist, ImageProvenance, ImageRecord, Post, Source, Tag
-from ..models.tag import image_tag
+from ..models.tag import PRESENTATION_SYSTEM_TAGS, image_tag
 from .pagination import decode_cursor, encode_cursor
 from .tag_query import (
     fandom_join_alias,
@@ -693,9 +693,23 @@ class GalleryService:
         eff = _effective_date_col()
         stmt = select(ImageRecord, Post.post_date, eff.label("eff"))
         stmt = _outer_join_primary_post(stmt)
+        # Presentation images (banner / editor-screenshot system tags, #128)
+        # cluster on UI chrome rather than content, so near any one of them
+        # they'd fill the grid. Excluded from CANDIDATES only — the anchor
+        # itself may be a banner, and `wip` stays surfaced (real art; only
+        # the training pipelines exclude it).
+        presentation = (
+            select(image_tag.c.image_record_id)
+            .join(Tag, Tag.id == image_tag.c.tag_id)
+            .where(
+                Tag.is_system.is_(True),
+                Tag.name.in_(PRESENTATION_SYSTEM_TAGS),
+            )
+        )
         stmt = stmt.where(
             ImageRecord.siglip_embedding.is_not(None),
             ImageRecord.id != image_id,
+            ImageRecord.id.not_in(presentation),
         )
         stmt = _apply_scope(
             stmt, tag_ids=tag_ids, post_id=None,

backend/app/services/ml/ccip.py

@@ -62,6 +62,18 @@ def _single_character_images():
     )
 
 
+def _hygiene_tagged_images():
+    """Subquery of image ids carrying any SYSTEM tag (wip / banner / editor
+    screenshot). Training hygiene (#128): such images never contribute
+    reference prototypes — a faceless wip's figure region would otherwise
+    become an identity reference for the character it's tagged with."""
+    return (
+        select(image_tag.c.image_record_id)
+        .join(Tag, Tag.id == image_tag.c.tag_id)
+        .where(Tag.is_system.is_(True))
+    )
+
+
 async def _ref_signature(session: AsyncSession) -> tuple:
     n_tags = (
         await session.execute(
@@ -79,7 +91,17 @@ async def _ref_signature(session: AsyncSession) -> tuple:
             )
         )
     ).one()
-    return (n_tags, n_regs, max_id)
+    # Hygiene applications must invalidate too: tagging an image `wip` changes
+    # the reference set without touching character-tag or region counts.
+    n_hygiene = (
+        await session.execute(
+            select(func.count())
+            .select_from(image_tag)
+            .join(Tag, Tag.id == image_tag.c.tag_id)
+            .where(Tag.is_system.is_(True))
+        )
+    ).scalar_one()
+    return (n_tags, n_regs, max_id, n_hygiene)
 
 
 async def character_references(session: AsyncSession) -> dict[int, list]:
@@ -102,6 +124,9 @@ async def character_references(session: AsyncSession) -> dict[int, list]:
             .where(ImageRegion.kind.in_(_FIGURE_KINDS))
             .where(ImageRegion.ccip_embedding.is_not(None))
             .where(ImageRegion.image_record_id.in_(_single_character_images()))
+            .where(
+                ImageRegion.image_record_id.not_in(_hygiene_tagged_images())
+            )
         )
     ).all()
     refs: dict[int, list] = {}

backend/app/services/ml/heads.py

@@ -40,6 +40,7 @@ from ...models import (
 from ...models.tag import image_tag
 from .training_data import (
     _auto_apply_point,
+    _hygiene_excluded_ids,
     _ids_with_tag,
     _l2norm,
     _load_embeddings,
@@ -150,11 +151,16 @@ def train_all_heads(
     embedding_version = _embedder_version(session)
     eligible = _eligible_tag_ids(session, cfg["min_positives"])
     eligible_set = set(eligible)
+    # Computed once per run, not per head — the hygiene set is identical for
+    # every non-system concept.
+    hygiene = _hygiene_excluded_ids(session)
     trained = 0
     skipped = 0
     for i, tag_id in enumerate(eligible):
         try:
-            ok = train_head(session, tag_id, embedding_version, cfg, np)
+            ok = train_head(
+                session, tag_id, embedding_version, cfg, np, hygiene=hygiene
+            )
         except Exception:
             log.exception("train_head failed for tag %d", tag_id)
             ok = False
@@ -174,27 +180,56 @@ def train_all_heads(
     return {"n_trained": trained, "n_skipped": skipped}
 
 
+def head_training_ids(
+    session: Session, tag_id: int, cfg: dict, hygiene: set[int] | None = None,
+) -> tuple[list[int], list[int]] | None:
+    """Select (pos_ids, neg_ids) for one head. Split out of train_head and
+    kept sklearn-free so the hygiene exclusion is testable in the CI env
+    (sklearn only exists in the ml image). Returns None when the concept has
+    too few usable positives.
+
+    Training hygiene (#128): images carrying a system tag are ABSENT from
+    every other concept's training — dropped as positives AND kept out of
+    the rejection/sampled negative pool (see _hygiene_excluded_ids). A system
+    tag's own head trains on them unfiltered: its positives ARE the hygiene
+    images."""
+    tag = session.get(Tag, tag_id)
+    if tag is not None and tag.is_system:
+        hygiene = set()
+    elif hygiene is None:
+        hygiene = _hygiene_excluded_ids(session)
+
+    pos_ids = [i for i in _ids_with_tag(session, tag_id) if i not in hygiene]
+    if len(pos_ids) < cfg["min_positives"]:
+        return None
+
+    pos_set = set(pos_ids)
+    rejected = [
+        i for i in _rejected_ids(session, tag_id)
+        if i not in pos_set and i not in hygiene
+    ]
+    want_neg = max(len(pos_ids) * cfg["neg_ratio"], _EXAMPLES_MIN * 4)
+    sampled = _sample_unlabeled(
+        session, pos_set | set(rejected) | hygiene,
+        min(_UNLABELED_POOL, want_neg),
+    )
+    return pos_ids, rejected + [i for i in sampled if i not in pos_set]
+
+
 def train_head(
-    session: Session, tag_id: int, embedding_version: str, cfg: dict, np
+    session: Session, tag_id: int, embedding_version: str, cfg: dict, np,
+    hygiene: set[int] | None = None,
 ) -> bool:
     """Fit + upsert one head. Returns True if a head was written, False if the
     concept had too few usable examples to train (the row is then removed)."""
     from sklearn.linear_model import LogisticRegression
     from sklearn.model_selection import StratifiedKFold, cross_val_predict
 
-    pos_ids = _ids_with_tag(session, tag_id)
-    if len(pos_ids) < cfg["min_positives"]:
+    ids = head_training_ids(session, tag_id, cfg, hygiene)
+    if ids is None:
         session.execute(delete(TagHead).where(TagHead.tag_id == tag_id))
         return False
-
-    pos_set = set(pos_ids)
-    rejected = [i for i in _rejected_ids(session, tag_id) if i not in pos_set]
-    want_neg = max(len(pos_ids) * cfg["neg_ratio"], _EXAMPLES_MIN * 4)
-    sampled = _sample_unlabeled(
-        session, pos_set | set(rejected), min(_UNLABELED_POOL, want_neg)
-    )
-    neg_ids = rejected + [i for i in sampled if i not in pos_set]
-
+    pos_ids, neg_ids = ids
     emb = _load_embeddings(session, pos_ids + neg_ids)
     pos = [emb[i] for i in pos_ids if i in emb]
     neg = [emb[i] for i in neg_ids if i in emb]

backend/app/services/ml/training_data.py

@@ -17,10 +17,33 @@ from typing import Any
 from sqlalchemy import func, select
 from sqlalchemy.orm import Session
 
-from ...models import ImageRecord, TagSuggestionRejection
+from ...models import ImageRecord, Tag, TagSuggestionRejection
 from ...models.tag import image_tag
 
 
+def _hygiene_excluded_ids(session: Session) -> set[int]:
+    """Ids of images carrying ANY system tag (wip / banner / editor
+    screenshot — milestone #128). These images are excluded from OTHER
+    concepts' head training entirely: not positives (a rough wip tagged as a
+    character drags that head toward 'generic sketch') and not sampled or
+    rejection negatives (a wip OF character X is not evidence against X) —
+    simply absent. A system tag's OWN head trains on them unchanged; that is
+    what makes auto-flagging banners/editor screenshots work.
+
+    Item-level by design: a wip-tagged process video contributes (or
+    withholds) ALL its sampled frames, though some may show the finished
+    piece. Operator call 2026-07-03: with enough clean data this washes out —
+    no per-frame handling.
+    """
+    return set(
+        session.execute(
+            select(image_tag.c.image_record_id)
+            .join(Tag, Tag.id == image_tag.c.tag_id)
+            .where(Tag.is_system.is_(True))
+        ).scalars().all()
+    )
+
+
 def _ids_with_tag(session: Session, tag_id: int) -> list[int]:
     return [
         r[0] for r in session.execute(

backend/app/services/platforms/subscribestar.py

@@ -14,9 +14,11 @@
    `_personalization_id` age-confirmation cookie that the user can't
    easily refresh — SubscribeStar's frontend JS uses localStorage to
    suppress the age popup once dismissed. gallery-dl's own login flow
-   sidesteps this by setting `18_plus_agreement_generic=true` on
-   `.subscribestar.adult`; we mirror that for cookies captured via the
-   extension.
+   sidesteps this by setting `18_plus_agreement_generic=true`; we mirror
+   that for cookies captured via the extension — on EVERY SubscribeStar
+   domain, because cookies are domain-scoped and the wall exists on all
+   of them: an .adult-only line never rides along to a .art creator page
+   (Elasid, event #54116 — every poll 302'd to /age_confirmation_warning).
 """
 
 from .base import GD_DEFAULTS, PlatformInfo, str_id_value
@@ -29,20 +31,31 @@ def derive_post_url(data: dict) -> str | None:
     return None
 
 
+# All domains SubscribeStar serves creators from; the age wall gates each.
+_SS_DOMAINS = (".subscribestar.com", ".subscribestar.adult", ".subscribestar.art")
+
+
 def augment_cookies(netscape: str) -> str:
-    if "18_plus_agreement_generic" in netscape:
-        return netscape
     # Far-future expiry — gallery-dl's own login flow sets this with no
     # explicit expiry; the server only checks presence/value.
     expiry = 4102444800  # 2100-01-01 UTC
-    line = "\t".join([
-        ".subscribestar.adult", "TRUE", "/", "TRUE",
-        str(expiry), "18_plus_agreement_generic", "true",
-    ])
     body = netscape.rstrip("\n")
     if not body:
         body = "# Netscape HTTP Cookie File"
-    return body + "\n" + line + "\n"
+    lines = netscape.splitlines()
+    for domain in _SS_DOMAINS:
+        # Per-domain presence check: captured cookies carrying the age
+        # cookie for one domain must not suppress injection on the others.
+        if any(
+            line.startswith(domain + "\t") and "18_plus_agreement_generic" in line
+            for line in lines
+        ):
+            continue
+        body += "\n" + "\t".join([
+            domain, "TRUE", "/", "TRUE",
+            str(expiry), "18_plus_agreement_generic", "true",
+        ])
+    return body + "\n"
 
 
 INFO = PlatformInfo(
@@ -51,10 +64,11 @@ INFO = PlatformInfo(
     description="Download posts from SubscribeStar creators",
     auth_type="cookies",
     requires_auth=True,
-    url_pattern=r"^https?://(www\.)?subscribestar\.(com|adult)/",
+    url_pattern=r"^https?://(www\.)?subscribestar\.(com|adult|art)/",
     url_examples=[
         "https://subscribestar.adult/example_artist",
         "https://www.subscribestar.com/example_artist",
+        "https://subscribestar.art/example_artist",
     ],
     default_config={**GD_DEFAULTS, "content_types": ["all"]},
     derive_post_url=derive_post_url,

backend/app/services/tag_directory_service.py

@@ -107,6 +107,7 @@ class TagDirectoryService:
                 "kind": tag.kind.value if hasattr(tag.kind, "value") else tag.kind,
                 "fandom_id": tag.fandom_id,
                 "fandom_name": fandom_name,
+                "is_system": tag.is_system,
                 "image_count": int(image_count),
                 "preview_thumbnails": previews.get(tag.id, []),
             }

backend/app/services/tag_query.py

@@ -67,24 +67,28 @@ def fandom_join_alias():
 
 
 def tag_columns(fandom_alias):
-    """The canonical (id, name, kind, fandom_id, fandom_name) column set for a
-    tag select that outerjoins `fandom_alias` (from `fandom_join_alias()`)."""
+    """The canonical (id, name, kind, fandom_id, fandom_name, is_system)
+    column set for a tag select that outerjoins `fandom_alias` (from
+    `fandom_join_alias()`)."""
     return [
         Tag.id,
         Tag.name,
         Tag.kind,
         Tag.fandom_id,
         fandom_alias.c.name.label("fandom_name"),
+        Tag.is_system,
     ]
 
 
 def serialize_tag(row) -> dict:
     """Serialize a row carrying .id/.name/.kind/.fandom_id/.fandom_name to the
-    canonical tag dict. `kind` may be a TagKind enum or a plain string."""
+    canonical tag dict. `kind` may be a TagKind enum or a plain string.
+    `is_system` defaults False for callers whose selects predate the column."""
     return {
         "id": row.id,
         "name": row.name,
         "kind": row.kind.value if hasattr(row.kind, "value") else row.kind,
         "fandom_id": row.fandom_id,
         "fandom_name": row.fandom_name,
+        "is_system": bool(getattr(row, "is_system", False)),
     }

backend/app/services/tag_service.py

@@ -330,6 +330,12 @@ class TagService:
         tag = await self.session.get(Tag, tag_id)
         if tag is None:
             raise TagValidationError(f"Tag {tag_id} not found")
+        # The training-hygiene machinery keys on system tags by row; a rename
+        # would silently disarm it (milestone #128).
+        if tag.is_system:
+            raise TagValidationError(
+                f"'{tag.name}' is a system tag and cannot be renamed"
+            )
 
         # Case-insensitive clash (#701) — renaming onto a differently-cased tag
         # is still a merge, not a silent fork.
@@ -528,6 +534,14 @@ class TagService:
         source = await self.session.get(Tag, source_id)
         if source is None:
             raise TagValidationError(f"Tag {source_id} not found")
+        # Merge deletes the source row — the only tag-delete path — and the
+        # training-hygiene machinery keys on system rows (milestone #128).
+        # Merging INTO a system tag stays allowed (adopting an operator's old
+        # 'wip sketch' tag into the system 'wip' is the intended move).
+        if source.is_system:
+            raise TagValidationError(
+                f"'{source.name}' is a system tag and cannot be merged away"
+            )
         target = await self.session.get(Tag, target_id)
         if target is None:
             raise TagValidationError(f"Tag {target_id} not found")
@@ -792,6 +806,10 @@ async def normalize_existing_tags(
     rows = (
         await session.execute(
             select(Tag.id, Tag.name, Tag.kind, Tag.fandom_id)
+            # System tags (#128) are exempt: their names are the keys the
+            # hygiene machinery matches on, and canonicalization would recase
+            # 'wip' → 'Wip' (this path bypasses TagService.rename's guard).
+            .where(Tag.is_system.is_(False))
         )
     ).all()
     groups = _group_existing_tags(rows)

frontend/src/components/discovery/TagCard.vue

@@ -13,9 +13,15 @@
       <div class="fc-tagcard__name">
         <template v-if="!editing">
           {{ card.name }}
+          <v-icon
+            v-if="card.is_system" size="14" icon="mdi-shield-outline"
+            class="fc-tagcard__system"
+            title="System tag — tagged items are excluded from training other concepts"
+          />
           <span v-if="card.kind === 'character' && card.fandom_name"
                 class="fc-tagcard__fandom">— {{ card.fandom_name }}</span>
           <v-icon
+            v-if="!card.is_system"
             class="fc-tagcard__edit" size="14" icon="mdi-pencil"
             @click.stop="startEdit"
           />
@@ -58,12 +64,18 @@
               prepend-icon="mdi-tag-multiple"
               @click="$emit('aliases', card)"
             />
+            <!-- System tags (#128): merge-away and delete are refused
+                 server-side — the hygiene machinery keys on the row — so
+                 don't offer them. Aliases stay (mapping model outputs ONTO
+                 a system tag is useful). -->
             <v-list-item
+              v-if="!card.is_system"
               title="Merge with…"
               prepend-icon="mdi-call-merge"
               @click="$emit('merge-with', card)"
             />
             <v-list-item
+              v-if="!card.is_system"
               title="Delete tag"
               prepend-icon="mdi-delete"
               base-color="error"
@@ -133,6 +145,7 @@ function submit() {
 .fc-tagcard__body { padding: 8px 12px; }
 .fc-tagcard__name { font-weight: 600; }
 .fc-tagcard__fandom { font-weight: 400; opacity: 0.7; }
+.fc-tagcard__system { opacity: 0.6; margin-left: 2px; }
 .fc-tagcard__meta {
   display: flex; align-items: center; justify-content: space-between;
   margin-top: 4px;

frontend/src/components/modal/TagAutocomplete.vue

@@ -100,6 +100,13 @@ import { useSuggestionsStore } from '../../stores/suggestions.js'
 import FandomPicker from './FandomPicker.vue'
 
 const emit = defineEmits(['pick-existing', 'pick-new', 'accept-suggestion', 'cancel'])
+// The host surface's applied tags (host.current?.tags). Both dropdown sections
+// filter against this so a just-added tag drops out of the search the moment
+// the chip rail updates, not after the next modal refresh (operator-asked
+// 2026-07-03).
+const props = defineProps({
+  appliedTags: { type: Array, default: () => [] },
+})
 const store = useTagStore()
 // The image's ML (Camie) suggestions, surfaced inline in this dropdown so the
 // operator can pick a suggestion while typing instead of hunting for it in the
@@ -208,6 +215,18 @@ const createLabel = computed(() =>
 
 function scorePct (s) { return `${Math.round(s.score * 100)}%` }
 
+const appliedIds = computed(() => new Set(props.appliedTags.map(t => t.id)))
+const appliedNames = computed(() =>
+  new Set(props.appliedTags.map(t => `${t.kind}:${(t.name || '').toLowerCase()}`)),
+)
+
+// Server matches minus the image's applied tags. allowCreate/sameNameCharExists
+// keep reading the RAW hits — they reason about the tag universe (does this tag
+// exist?), not about this image.
+const visibleHits = computed(() =>
+  hits.value.filter(h => !appliedIds.value.has(h.id)),
+)
+
 // This image's suggestions that match the typed query, minus any the server
 // autocomplete already returned (same name+kind) so a tag never shows twice.
 // Sources the FULL prediction set (allByCategory, down to the store floor) — NOT
@@ -218,7 +237,7 @@ function scorePct (s) { return `${Math.round(s.score * 100)}%` }
 const suggestionHits = computed(() => {
   const q = parsedName.value.toLowerCase()
   if (!q) return []
-  const seen = new Set(hits.value.map(h => `${h.kind}:${h.name.toLowerCase()}`))
+  const seen = new Set(visibleHits.value.map(h => `${h.kind}:${h.name.toLowerCase()}`))
   const out = []
   for (const list of Object.values(suggestions.allByCategory)) {
     for (const s of list || []) {
@@ -226,7 +245,12 @@ const suggestionHits = computed(() => {
       // can show + un-reject them; keep them OUT of the type-to-add dropdown,
       // whose job is finding a tag to ADD (un-reject lives in the panel).
       if (s.rejected) continue
+      // Already on the image (matched by canonical id or name+kind): nothing
+      // to add. Covers the window between an add and the next suggestions
+      // fetch, where the stale row would otherwise still be pickable.
+      if (s.canonical_tag_id != null && appliedIds.value.has(s.canonical_tag_id)) continue
       const key = `${s.category}:${s.display_name.toLowerCase()}`
+      if (appliedNames.value.has(key)) continue
       if (!s.display_name.toLowerCase().includes(q)) continue
       if (seen.has(key)) continue
       seen.add(key)
@@ -242,7 +266,7 @@ const suggestionHits = computed(() => {
 // One ordered list backing both the rendered dropdown and keyboard nav, so the
 // highlight index maps 1:1 to a row regardless of which section it's in.
 const rows = computed(() => {
-  const r = hits.value.map(h => ({ type: 'hit', key: `h${h.id}`, hit: h }))
+  const r = visibleHits.value.map(h => ({ type: 'hit', key: `h${h.id}`, hit: h }))
   for (const s of suggestionHits.value) {
     r.push({ type: 'suggestion', key: `s${s.category}:${s.display_name}`, sugg: s })
   }

frontend/src/components/modal/TagChip.vue

@@ -10,14 +10,23 @@
       @click:close="$emit('remove', tag.id)"
     >
       <v-icon start size="x-small">{{ iconFor(tag.kind) }}</v-icon>
-      {{ tag.name }}<span
+      {{ tag.name }}<v-icon
+        v-if="tag.is_system" end size="x-small" class="fc-tag-chip__system"
+        title="System tag — tagged items are excluded from training other concepts"
+      >mdi-shield-outline</v-icon><span
         v-if="tag.fandom_id" class="fc-tag-chip__fandom"
         :title="tag.fandom_name ? `Fandom: ${tag.fandom_name}` : 'Has a fandom'"
       >→<template v-if="fandomLabel">&nbsp;{{ fandomLabel }}</template></span>
     </v-chip>
     <!-- Modal-safe kebab is baked into KebabMenu (this chip lives in the
-         teleported image modal — #711). -->
-    <KebabMenu class="fc-tag-chip__kebab" :label="`More actions for ${tag.name}`">
+         teleported image modal — #711). System tags hide it entirely: rename
+         is refused server-side (the hygiene machinery keys on the row) and
+         set-fandom never applies to their general kind. Remove (✕) stays —
+         un-tagging an image is normal use. -->
+    <KebabMenu
+      v-if="!tag.is_system"
+      class="fc-tag-chip__kebab" :label="`More actions for ${tag.name}`"
+    >
       <v-list-item @click="$emit('rename', tag)">
         <v-list-item-title>Rename…</v-list-item-title>
       </v-list-item>
@@ -64,6 +73,7 @@ function iconFor (k) { return KIND_ICONS[k] || 'mdi-tag' }
   outline: 2px solid rgb(var(--v-theme-accent));
   outline-offset: 1px;
 }
+.fc-tag-chip__system { opacity: 0.6; margin-left: 2px; }
 .fc-tag-chip__kebab { opacity: 0.7; }
 .fc-tag-chip:hover .fc-tag-chip__kebab { opacity: 1; }
 .fc-tag-chip__fandom { opacity: 0.7; font-size: 0.85em; }

frontend/src/components/modal/TagPanel.vue

@@ -15,6 +15,7 @@
 
     <TagAutocomplete
       ref="tagInputRef"
+      :applied-tags="host.current?.tags || []"
       @pick-existing="onPickExisting" @pick-new="onPickNew"
       @accept-suggestion="onAcceptSuggestion"
     />
@@ -107,17 +108,56 @@ defineExpose({ focusTagInput })
 // (operator-asked 2026-06-26; the Explore workspace leans on this hard).
 async function onRemove(tagId) {
   errorMsg.value = null
-  try { await host.removeTag(tagId); focusTagInput() }
+  try {
+    await host.removeTag(tagId)
+    // removeTag also records a per-image dismissal server-side, so reloading
+    // puts a model-suggested tag BACK in the suggestions area (flagged
+    // rejected — visible + one-click reversible) instead of it vanishing
+    // until the next modal open (operator-asked 2026-07-03).
+    if (host.currentImageId != null) {
+      suggestions.load(host.currentImageId)
+      suggestions.loadAll(host.currentImageId)
+    }
+    focusTagInput()
+  }
   catch (e) { errorMsg.value = e.message }
 }
 async function onPickExisting(hit) {
   errorMsg.value = null
-  try { await host.addExistingTag(hit.id); focusTagInput() }
+  try {
+    // Manually picking a tag the model also suggested = accepting the
+    // suggestion: the acceptance is recorded and the row drops from the
+    // panel, instead of silently bypassing the feedback loop
+    // (operator-asked 2026-07-03).
+    const pending = suggestions.findPending(hit.kind, hit.name, hit.id)
+    if (pending) {
+      await suggestions.accept(pending, { tagId: hit.id })
+      await host.reloadTags()
+    } else {
+      await host.addExistingTag(hit.id)
+    }
+    focusTagInput()
+  }
   catch (e) { errorMsg.value = e.message }
 }
 async function onPickNew(payload) {
   errorMsg.value = null
-  try { await host.createAndAdd(payload); focusTagInput() }
+  try {
+    const imageId = host.currentImageId
+    const created = await host.createAndAdd(payload)
+    // The typed name can match a raw (or canonical) suggestion; creating it
+    // by hand still counts as accepting — record it and drop the row. Runs
+    // AFTER createAndAdd so a character's fandom pick is preserved (accept's
+    // own create path knows no fandom); the accept apply is idempotent.
+    // Guard on the image not having changed mid-create (fast prev/next).
+    if (created && host.currentImageId === imageId) {
+      const pending = suggestions.findPending(payload.kind, payload.name, created.id)
+      if (pending) {
+        await suggestions.accept(pending, { tagId: created.id })
+      }
+    }
+    focusTagInput()
+  }
   catch (e) { errorMsg.value = e.message }
 }
 // A suggestion picked from the autocomplete dropdown runs the SAME path as the

frontend/src/stores/explore.js

@@ -176,6 +176,9 @@ export const useExploreStore = defineStore('explore', () => {
     }
     if (anchor.value?.id !== id) return  // walked away mid-create
     await addExistingTag(tag.id)
+    // Returned so TagPanel can match the created tag against a pending
+    // suggestion and record the acceptance (manual add == accept, 2026-07-03).
+    return tag
   }
 
   // No overlay to dismiss in the Explore workspace — the chip-body "show me

frontend/src/stores/modal.js

@@ -170,6 +170,9 @@ export const useModalStore = defineStore('modal', () => {
     }
     if (currentImageId.value !== imageId) return  // navigated away
     await addExistingTag(tag.id)
+    // Returned so TagPanel can match the created tag against a pending
+    // suggestion and record the acceptance (manual add == accept, 2026-07-03).
+    return tag
   }
 
   const isOpen = computed(() => currentImageId.value !== null)

frontend/src/stores/suggestions.js

@@ -97,7 +97,12 @@ export const useSuggestionsStore = defineStore('suggestions', () => {
     }
   }
 
-  async function accept(suggestion) {
+  // opts.tagId: the tag's known id, when the caller already created/resolved
+  // it (TagPanel's manual-add-matches-suggestion path). Passed separately —
+  // NOT spread onto the suggestion — because _dropEverywhere keys off the
+  // suggestion object, and mutating canonical_tag_id on a raw suggestion
+  // would stop it matching its own row in the lists.
+  async function accept(suggestion, { tagId: knownTagId = null } = {}) {
     // Capture imageId so a mid-flight prev/next can't reroute the
     // accept POST to a different image AND push the tag to that
     // image's allowlist.
@@ -106,7 +111,7 @@ export const useSuggestionsStore = defineStore('suggestions', () => {
     // Raw tags (creates_new_tag) have no canonical_tag_id; the backend's
     // accept endpoint needs a tag_id, so for raw tags we create the tag
     // first via the existing /api/tags endpoint, then accept by id.
-    let tagId = suggestion.canonical_tag_id
+    let tagId = knownTagId ?? suggestion.canonical_tag_id
     if (tagId == null) {
       const created = await api.post('/api/tags', {
         body: { name: suggestion.display_name, kind: suggestion.category }
@@ -169,6 +174,29 @@ export const useSuggestionsStore = defineStore('suggestions', () => {
     toast({ text: `Alias removed: ${suggestion.display_name}`, type: 'success' })
   }
 
+  // Find a live (non-rejected) pending suggestion matching a manually-picked
+  // tag — by canonical id when known, else by (kind, name). Manually adding a
+  // tag the model also suggested must be indistinguishable from accepting the
+  // suggestion (recorded + dropped from the panel), so TagPanel routes matches
+  // through accept() (operator-asked 2026-07-03). Searches the full dropdown
+  // set first, then the panel list (loadAll is best-effort and can be empty).
+  function findPending(kind, name, tagId = null) {
+    const lname = (name || '').toLowerCase()
+    for (const map of [allByCategory.value, byCategory.value]) {
+      for (const list of Object.values(map)) {
+        for (const s of list || []) {
+          if (s.rejected) continue
+          if (tagId != null && s.canonical_tag_id === tagId) return s
+          if (
+            s.category === kind &&
+            (s.display_name || '').toLowerCase() === lname
+          ) return s
+        }
+      }
+    }
+    return null
+  }
+
   async function dismiss(suggestion) {
     const imageId = currentImageId
     if (imageId == null) return
@@ -202,6 +230,7 @@ export const useSuggestionsStore = defineStore('suggestions', () => {
 
   return {
     byCategory, allByCategory, loading, error,
-    load, loadAll, accept, aliasAccept, removeAlias, dismiss, undismiss
+    load, loadAll, accept, aliasAccept, removeAlias, dismiss, undismiss,
+    findPending
   }
 })

frontend/test/suggestions.spec.js

@@ -0,0 +1,112 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest'
+import { setActivePinia, createPinia } from 'pinia'
+import { useSuggestionsStore } from '../src/stores/suggestions.js'
+
+vi.mock('../src/utils/toast.js', () => ({ toast: vi.fn() }))
+
+function stubFetch(handler) {
+  globalThis.fetch = vi.fn(async (url, init) => {
+    const { status, body } = handler(url, init)
+    return {
+      ok: status >= 200 && status < 300,
+      status,
+      statusText: String(status),
+      text: async () => (body == null ? '' : JSON.stringify(body)),
+    }
+  })
+}
+
+const sugg = (over = {}) => ({
+  canonical_tag_id: 7,
+  display_name: 'Ichigo',
+  category: 'character',
+  score: 0.9,
+  source: 'head',
+  creates_new_tag: false,
+  rejected: false,
+  ...over,
+})
+
+// Seed the store through its own load/loadAll so currentImageId is set the
+// way the panel sets it (accept() derives the image from it).
+async function seed(store, byCategory) {
+  stubFetch(() => ({ status: 200, body: { by_category: byCategory } }))
+  await store.load(1)
+  await store.loadAll(1)
+}
+
+describe('suggestions store: findPending (manual add == accept, 2026-07-03)', () => {
+  beforeEach(() => setActivePinia(createPinia()))
+  afterEach(() => vi.restoreAllMocks())
+
+  it('matches by canonical tag id', async () => {
+    const s = useSuggestionsStore()
+    await seed(s, { character: [sugg()] })
+    expect(s.findPending('general', 'unrelated', 7)?.display_name).toBe('Ichigo')
+  })
+
+  it('matches raw suggestions by kind + name, case-insensitively', async () => {
+    const s = useSuggestionsStore()
+    await seed(s, {
+      general: [sugg({
+        canonical_tag_id: null, display_name: 'Holding Sword',
+        category: 'general', creates_new_tag: true,
+      })],
+    })
+    expect(s.findPending('general', 'holding sword')).toBeTruthy()
+    // Kind must match: same name under another kind is a different tag.
+    expect(s.findPending('character', 'holding sword')).toBeNull()
+  })
+
+  it('skips rejected rows and returns null when nothing matches', async () => {
+    const s = useSuggestionsStore()
+    await seed(s, { character: [sugg({ rejected: true })] })
+    expect(s.findPending('character', 'Ichigo', 7)).toBeNull()
+    expect(s.findPending('character', 'Rukia')).toBeNull()
+  })
+})
+
+describe('suggestions store: accept with a known tag id', () => {
+  beforeEach(() => setActivePinia(createPinia()))
+  afterEach(() => vi.restoreAllMocks())
+
+  it('POSTs only the accept endpoint (no tag create) and drops the row', async () => {
+    const s = useSuggestionsStore()
+    await seed(s, { character: [sugg()] })
+    const calls = []
+    stubFetch((url, init) => {
+      calls.push({ url, method: init?.method })
+      return { status: 200, body: { accepted: true, tag_id: 7 } }
+    })
+    await s.accept(s.byCategory.character[0], { tagId: 7 })
+    expect(calls).toHaveLength(1)
+    expect(calls[0].url).toContain('/api/images/1/suggestions/accept')
+    expect(s.byCategory.character).toHaveLength(0)
+    expect(s.allByCategory.character).toHaveLength(0)
+  })
+
+  it('a RAW suggestion accepted with a known tagId skips create AND still drops its row', async () => {
+    // TagPanel's manual-create path: the tag was just created by
+    // host.createAndAdd, so accept must not create again — and the drop must
+    // key off the original raw suggestion object, not the resolved id
+    // (regression: spreading canonical_tag_id onto the suggestion changed its
+    // identity key and left the panel row behind).
+    const s = useSuggestionsStore()
+    const raw = sugg({
+      canonical_tag_id: null, display_name: 'Holding Sword',
+      category: 'general', creates_new_tag: true,
+    })
+    await seed(s, { general: [raw] })
+    const calls = []
+    stubFetch((url, init) => {
+      calls.push({ url, body: init?.body })
+      return { status: 200, body: { accepted: true, tag_id: 42 } }
+    })
+    await s.accept(s.byCategory.general[0], { tagId: 42 })
+    expect(calls).toHaveLength(1)
+    expect(calls[0].url).toContain('/api/images/1/suggestions/accept')
+    expect(JSON.parse(calls[0].body)).toEqual({ tag_id: 42 })
+    expect(s.byCategory.general).toHaveLength(0)
+    expect(s.allByCategory.general).toHaveLength(0)
+  })
+})

tests/conftest.py

@@ -166,6 +166,17 @@ def _reset_db_after_integration(request, _truncate_engine):
             rows = (_SEED_SNAPSHOT or {}).get(t.name)
             if rows:
                 conn.execute(t.insert(), rows)
+                # Restored rows carry their explicit ids while RESTART
+                # IDENTITY reset the sequence to 1, so the next ORM insert
+                # would collide on the PK (bitten by migration 0075's seeded
+                # system tags: every Tag insert failed with pk_tag id=1).
+                # Resync any serial id sequence past the restored rows.
+                if "id" in t.c:
+                    conn.exec_driver_sql(
+                        f"SELECT setval(pg_get_serial_sequence('{t.name}', 'id'), "
+                        f"(SELECT COALESCE(MAX(id), 1) FROM {t.name})) "
+                        f"WHERE pg_get_serial_sequence('{t.name}', 'id') IS NOT NULL"
+                    )
 
 
 @pytest_asyncio.fixture(autouse=True)

tests/test_api_admin.py

@@ -660,8 +660,10 @@ async def test_reset_content_tagging_apply_requires_confirm_token(client, db):
     for bad in ({"dry_run": False}, {"dry_run": False, "confirm": "deadbeef"}):
         resp = await client.post("/api/admin/tags/reset-content", json=bad)
         assert resp.status_code == 400
+    # Non-system only: the seeded hygiene tags (#128) survive a reset by
+    # design, so they'd inflate a whole-table count.
     remaining = (await db.execute(
-        select(func.count()).select_from(Tag)
+        select(func.count()).select_from(Tag).where(Tag.is_system.is_(False))
     )).scalar_one()
     assert remaining == 1
 

tests/test_cleanup_service.py

@@ -377,8 +377,11 @@ def test_prune_unused_tags_dry_run_returns_count_and_names(db_sync, tmp_path):
     assert result["count"] == 2
     assert "prune-me-1" in result["sample_names"]
     assert "prune-me-2" in result["sample_names"]
-    # Nothing actually deleted.
-    surviving = db_sync.execute(select(func.count(Tag.id))).scalar_one()
+    # Nothing actually deleted. Non-system only — the seeded hygiene tags
+    # (#128) are exempt from pruning and would inflate a whole-table count.
+    surviving = db_sync.execute(
+        select(func.count(Tag.id)).where(Tag.is_system.is_(False))
+    ).scalar_one()
     assert surviving == 3
 
 
@@ -499,8 +502,11 @@ def test_reset_content_tagging_dry_run_counts_without_deleting(db_sync, tmp_path
     assert result["count"] == 2
     assert result["by_kind"] == {"general": 1, "character": 1}
     assert result["applications"] == 2
-    # Nothing deleted — all 4 tags still present.
-    assert db_sync.execute(select(func.count(Tag.id))).scalar_one() == 4
+    # Nothing deleted — all 4 fixture tags still present (non-system count;
+    # the seeded hygiene tags #128 also survive, by design).
+    assert db_sync.execute(
+        select(func.count(Tag.id)).where(Tag.is_system.is_(False))
+    ).scalar_one() == 4
 
 
 def test_reset_content_tagging_deletes_content_keeps_fandom_series(db_sync, tmp_path):
@@ -525,8 +531,11 @@ def test_reset_content_tagging_deletes_content_keeps_fandom_series(db_sync, tmp_
     result = cleanup_service.reset_content_tagging(db_sync, dry_run=False)
     assert result["deleted"] == 2
 
-    # general + character gone; fandom + series kept.
-    kinds = db_sync.execute(select(Tag.kind)).scalars().all()
+    # general + character gone; fandom + series kept. Non-system only — the
+    # seeded hygiene tags (#128, kind=general) survive a reset by design.
+    kinds = db_sync.execute(
+        select(Tag.kind).where(Tag.is_system.is_(False))
+    ).scalars().all()
     kind_vals = {k.value if hasattr(k, "value") else str(k) for k in kinds}
     assert kind_vals == {"fandom", "series"}
     # series_page ordering survived.

tests/test_credential_service.py

@@ -116,8 +116,10 @@ async def test_get_cookies_path_subscribestar_injects_age_cookie(db, crypto, tmp
     cookie; the browser-stored cookie expires annually and can't be easily
     refreshed (the JS age popup is suppressed by localStorage). Mirror
     gallery-dl's own login-flow workaround by injecting
-    `18_plus_agreement_generic=true` on `.subscribestar.adult` whenever
-    cookies for subscribestar are materialized."""
+    `18_plus_agreement_generic=true` whenever cookies for subscribestar
+    are materialized — on EVERY SubscribeStar domain, since cookies are
+    domain-scoped and creators live on .com/.adult/.art alike (a .adult-only
+    line left .art sources stuck on the age wall — Elasid, event #54116)."""
     netscape_in = (
         "# Netscape HTTP Cookie File\n"
         ".subscribestar.adult\tTRUE\t/\tTRUE\t1700000000\tsession_id\txyz\n"
@@ -126,16 +128,18 @@ async def test_get_cookies_path_subscribestar_injects_age_cookie(db, crypto, tmp
     await svc.upsert(platform="subscribestar", credential_type="cookies", data=netscape_in)
     path = await svc.get_cookies_path("subscribestar")
     contents = path.read_text()
-    assert "18_plus_agreement_generic\ttrue" in contents
-    assert ".subscribestar.adult" in contents
+    for domain in (".subscribestar.com", ".subscribestar.adult", ".subscribestar.art"):
+        assert f"{domain}\tTRUE\t/\tTRUE\t4102444800\t18_plus_agreement_generic\ttrue" in contents
     # Original session_id cookie preserved.
     assert "session_id\txyz" in contents
 
 
 @pytest.mark.asyncio
-async def test_get_cookies_path_subscribestar_idempotent_when_present(db, crypto, tmp_path):
-    """If the operator's captured cookies ALREADY contain the age cookie
-    (e.g. a manual paste, or a re-login), don't double-inject."""
+async def test_get_cookies_path_subscribestar_idempotent_per_domain(db, crypto, tmp_path):
+    """If the operator's captured cookies ALREADY contain the age cookie for
+    a domain (e.g. a manual paste, or a re-login), don't double-inject on
+    THAT domain — but a .adult-only capture must still get the cookie added
+    for .com and .art (per-domain check, not presence-anywhere)."""
     netscape_in = (
         "# Netscape HTTP Cookie File\n"
         ".subscribestar.adult\tTRUE\t/\tTRUE\t1700000000\t18_plus_agreement_generic\ttrue\n"
@@ -145,7 +149,15 @@ async def test_get_cookies_path_subscribestar_idempotent_when_present(db, crypto
     await svc.upsert(platform="subscribestar", credential_type="cookies", data=netscape_in)
     path = await svc.get_cookies_path("subscribestar")
     contents = path.read_text()
-    assert contents.count("18_plus_agreement_generic") == 1
+    assert contents.count("18_plus_agreement_generic") == 3
+    adult_lines = [
+        line for line in contents.splitlines()
+        if line.startswith(".subscribestar.adult\t")
+        and "18_plus_agreement_generic" in line
+    ]
+    assert len(adult_lines) == 1
+    assert ".subscribestar.com\t" in contents
+    assert ".subscribestar.art\t" in contents
 
 
 @pytest.mark.asyncio

tests/test_gallery_similar.py

@@ -4,6 +4,7 @@ precomputed SigLIP image embeddings. No query-time ML inference."""
 from datetime import UTC, datetime, timedelta
 
 import pytest
+from sqlalchemy import select
 
 from backend.app.models import ImageRecord, Tag, TagKind
 from backend.app.models.tag import image_tag
@@ -72,6 +73,31 @@ async def test_similar_missing_source_returns_none(db):
     assert await svc.similar(99999, limit=10) is None
 
 
+@pytest.mark.asyncio
+async def test_similar_excludes_presentation_tagged_images(db):
+    """banner / editor-screenshot system tags (#128) hide from similar
+    RESULTS — they cluster on UI chrome, not content. A banner anchor still
+    gets results; `wip`-tagged images stay in (real art)."""
+    src = await _img(db, 1, _vec(1, 0))
+    bannered = await _img(db, 2, _vec(1, 0.02))   # nearest, but a banner
+    wipped = await _img(db, 3, _vec(1, 0.3))
+    plain = await _img(db, 4, _vec(1, 0.6))
+    banner_tag = (await db.execute(select(Tag).where(
+        Tag.is_system.is_(True), Tag.name == "banner"))).scalar_one()
+    wip_tag = (await db.execute(select(Tag).where(
+        Tag.is_system.is_(True), Tag.name == "wip"))).scalar_one()
+    await db.execute(image_tag.insert().values(
+        image_record_id=bannered.id, tag_id=banner_tag.id, source="manual"))
+    await db.execute(image_tag.insert().values(
+        image_record_id=wipped.id, tag_id=wip_tag.id, source="manual"))
+    svc = GalleryService(db)
+    res = await svc.similar(src.id, limit=10)
+    assert [i.id for i in res] == [wipped.id, plain.id]
+    # A presentation-tagged ANCHOR still answers — only candidates hide.
+    res_from_banner = await svc.similar(bannered.id, limit=10)
+    assert {i.id for i in res_from_banner} == {src.id, wipped.id, plain.id}
+
+
 @pytest.mark.asyncio
 async def test_similar_composes_with_tag_filter(db):
     src = await _img(db, 1, _vec(1, 0))

tests/test_system_tags.py

@@ -0,0 +1,152 @@
+"""System tags (milestone #128): seeded rows, protection guards, serialization.
+
+The three hygiene tags (wip / banner / editor screenshot) ship via migration
+0075 and drive the training-hygiene exclusions. They must exist after
+migration, serialize with is_system, and refuse rename/merge-away — the
+mechanism keys on these exact rows. Merge-INTO stays allowed: adopting an
+operator's old hygiene tag into the system row is the intended move.
+"""
+import pytest
+from sqlalchemy import insert, select
+
+from backend.app.models import ImageRecord, Tag, TagKind
+from backend.app.models.tag import image_tag
+from backend.app.services.cleanup_service import (
+    prune_unused_tags,
+    reset_content_tagging,
+)
+
+pytestmark = pytest.mark.integration
+
+SYSTEM_NAMES = {"wip", "banner", "editor screenshot"}
+
+
+async def _system_tag(db, name):
+    return (await db.execute(
+        select(Tag).where(Tag.is_system.is_(True), Tag.name == name)
+    )).scalar_one()
+
+
+@pytest.mark.asyncio
+async def test_seeded_system_tags_exist(db):
+    rows = (await db.execute(
+        select(Tag.name, Tag.kind).where(Tag.is_system.is_(True))
+    )).all()
+    assert {r.name for r in rows} == SYSTEM_NAMES
+    assert all(r.kind == TagKind.general for r in rows)
+
+
+@pytest.mark.asyncio
+async def test_get_tag_serializes_is_system(client, db):
+    wip = await _system_tag(db, "wip")
+    resp = await client.get(f"/api/tags/{wip.id}")
+    assert resp.status_code == 200
+    assert (await resp.get_json())["is_system"] is True
+
+
+@pytest.mark.asyncio
+async def test_rename_system_tag_refused(client, db):
+    wip = await _system_tag(db, "wip")
+    resp = await client.patch(
+        f"/api/tags/{wip.id}", json={"name": "work in progress"}
+    )
+    assert resp.status_code == 400
+    body = await resp.get_json()
+    assert "system tag" in body["error"]
+
+
+@pytest.mark.asyncio
+async def test_merge_away_system_tag_refused(client, db):
+    banner = await _system_tag(db, "banner")
+    created = await client.post(
+        "/api/tags", json={"name": "ad banner", "kind": "general"}
+    )
+    target_id = (await created.get_json())["id"]
+    resp = await client.post(
+        f"/api/tags/{banner.id}/merge", json={"target_id": target_id}
+    )
+    assert resp.status_code == 400
+    body = await resp.get_json()
+    assert "system tag" in body["error"]
+
+
+@pytest.mark.asyncio
+async def test_admin_delete_system_tag_refused(client, db):
+    wip = await _system_tag(db, "wip")
+    resp = await client.delete(f"/api/admin/tags/{wip.id}")
+    assert resp.status_code == 400
+    body = await resp.get_json()
+    assert body["error"] == "system_tag"
+    still = (await db.execute(
+        select(Tag.id).where(Tag.id == wip.id)
+    )).scalar_one_or_none()
+    assert still == wip.id
+
+
+@pytest.mark.asyncio
+async def test_prune_unused_skips_system_tags(db):
+    """The seeded system tags have zero applications and would match every
+    'unused' condition — the shared predicate must exempt them (preview AND
+    live delete, same conditions)."""
+    preview = await db.run_sync(
+        lambda s: prune_unused_tags(s, dry_run=True)
+    )
+    assert preview["count"] == 0
+    assert preview["sample_names"] == []
+    await db.run_sync(lambda s: prune_unused_tags(s, dry_run=False))
+    remaining = (await db.execute(
+        select(Tag.name).where(Tag.is_system.is_(True))
+    )).scalars().all()
+    assert set(remaining) == SYSTEM_NAMES
+
+
+@pytest.mark.asyncio
+async def test_reset_content_preserves_system_tags_and_applications(db):
+    """Reset re-tags CONTENT concepts; hygiene flags describe the file itself
+    and survive with their applications."""
+    wip = await _system_tag(db, "wip")
+    plain = Tag(name="doomed concept", kind=TagKind.general)
+    db.add(plain)
+    img = ImageRecord(
+        path="/images/r.jpg", sha256="f" * 64, size_bytes=1, mime="image/jpeg",
+        width=1, height=1, origin="imported_filesystem",
+        integrity_status="unknown",
+    )
+    db.add(img)
+    await db.flush()
+    for tid in (wip.id, plain.id):
+        await db.execute(insert(image_tag).values(
+            image_record_id=img.id, tag_id=tid, source="manual",
+        ))
+    await db.commit()
+
+    result = await db.run_sync(
+        lambda s: reset_content_tagging(s, dry_run=False)
+    )
+    assert result["deleted"] == 1  # only the plain general tag
+    names_left = (await db.execute(
+        select(Tag.name).where(Tag.kind == TagKind.general)
+    )).scalars().all()
+    assert "doomed concept" not in names_left
+    assert "wip" in names_left
+    wip_apps = (await db.execute(
+        select(image_tag.c.image_record_id)
+        .where(image_tag.c.tag_id == wip.id)
+    )).scalars().all()
+    assert wip_apps == [img.id]
+
+
+@pytest.mark.asyncio
+async def test_merge_into_system_tag_allowed(client, db):
+    wip = await _system_tag(db, "wip")
+    created = await client.post(
+        "/api/tags", json={"name": "old wip", "kind": "general"}
+    )
+    source_id = (await created.get_json())["id"]
+    resp = await client.post(
+        f"/api/tags/{source_id}/merge", json={"target_id": wip.id}
+    )
+    assert resp.status_code == 200
+    body = await resp.get_json()
+    assert body["target"]["id"] == wip.id
+    assert body["source_deleted"] is True

tests/test_tag_query.py

@@ -10,20 +10,29 @@ def test_serialize_tag_with_enum_kind():
     # ORM/column rows carry kind as a TagKind enum.
     row = SimpleNamespace(
         id=1, name="Sasuke Uchiha", kind=TagKind.character,
-        fandom_id=5, fandom_name="Naruto",
+        fandom_id=5, fandom_name="Naruto", is_system=False,
     )
     assert serialize_tag(row) == {
         "id": 1, "name": "Sasuke Uchiha", "kind": "character",
-        "fandom_id": 5, "fandom_name": "Naruto",
+        "fandom_id": 5, "fandom_name": "Naruto", "is_system": False,
     }
 
 
 def test_serialize_tag_with_string_kind_and_no_fandom():
-    # autocomplete hits already store kind as a plain string.
+    # autocomplete hits already store kind as a plain string. A row from a
+    # select that predates the is_system column serializes as non-system.
     row = SimpleNamespace(
         id=2, name="solo", kind="general", fandom_id=None, fandom_name=None,
     )
     assert serialize_tag(row) == {
         "id": 2, "name": "solo", "kind": "general",
-        "fandom_id": None, "fandom_name": None,
+        "fandom_id": None, "fandom_name": None, "is_system": False,
     }
+
+
+def test_serialize_tag_system_row():
+    row = SimpleNamespace(
+        id=3, name="wip", kind=TagKind.general,
+        fandom_id=None, fandom_name=None, is_system=True,
+    )
+    assert serialize_tag(row)["is_system"] is True

tests/test_training_hygiene.py

@@ -0,0 +1,125 @@
+"""Training hygiene (#128): system-tagged images are ABSENT from other
+concepts' training data and from CCIP reference prototypes.
+
+sklearn only exists in the ml image, so these pin head_training_ids (the
+sklearn-free selection split out of train_head) rather than a full fit —
+the exclusion lives entirely in that selection.
+"""
+import pytest
+from sqlalchemy import insert, select
+
+from backend.app.models import ImageRecord, ImageRegion, Tag, TagKind
+from backend.app.models.tag import image_tag
+from backend.app.services.ml import ccip
+from backend.app.services.ml.heads import head_training_ids
+from backend.app.services.ml.training_data import _hygiene_excluded_ids
+from backend.app.services.tag_service import TagService
+
+pytestmark = pytest.mark.integration
+
+_CFG = {"min_positives": 2, "neg_ratio": 1}
+
+
+async def _system_wip(db) -> Tag:
+    return (await db.execute(
+        select(Tag).where(Tag.is_system.is_(True), Tag.name == "wip")
+    )).scalar_one()
+
+
+async def _img(db, sha, *, embedded=True):
+    rec = ImageRecord(
+        path=f"/images/{sha}.jpg", sha256=sha, size_bytes=1, mime="image/jpeg",
+        width=1, height=1, origin="imported_filesystem",
+        integrity_status="unknown",
+        siglip_embedding=([0.1] * 1152 if embedded else None),
+    )
+    db.add(rec)
+    await db.flush()
+    return rec
+
+
+async def _apply(db, image_id, tag_id):
+    await db.execute(insert(image_tag).values(
+        image_record_id=image_id, tag_id=tag_id, source="manual",
+    ))
+
+
+@pytest.mark.asyncio
+async def test_hygiene_excluded_ids(db):
+    wip = await _system_wip(db)
+    flagged = await _img(db, "a" * 64)
+    await _img(db, "b" * 64)
+    await _apply(db, flagged.id, wip.id)
+    excluded = await db.run_sync(lambda s: _hygiene_excluded_ids(s))
+    assert excluded == {flagged.id}
+
+
+@pytest.mark.asyncio
+async def test_head_selection_drops_hygiene_from_both_sides(db):
+    """A wip-tagged image of concept X is neither a positive NOR a sampled
+    negative for X — it is absent entirely."""
+    wip = await _system_wip(db)
+    concept = await TagService(db).find_or_create("concept_x", TagKind.general)
+    clean_a = await _img(db, "a" * 64)
+    clean_b = await _img(db, "b" * 64)
+    flagged_pos = await _img(db, "c" * 64)     # X + wip: dropped positive
+    negative_pool = await _img(db, "d" * 64)   # untagged: legit negative
+    flagged_pool = await _img(db, "e" * 64)    # wip only: must not be sampled
+    for img in (clean_a, clean_b, flagged_pos):
+        await _apply(db, img.id, concept.id)
+    await _apply(db, flagged_pos.id, wip.id)
+    await _apply(db, flagged_pool.id, wip.id)
+
+    ids = await db.run_sync(lambda s: head_training_ids(s, concept.id, _CFG))
+    assert ids is not None
+    pos_ids, neg_ids = ids
+    assert set(pos_ids) == {clean_a.id, clean_b.id}
+    assert negative_pool.id in neg_ids
+    assert flagged_pos.id not in neg_ids
+    assert flagged_pool.id not in neg_ids
+
+
+@pytest.mark.asyncio
+async def test_system_tags_own_head_keeps_hygiene_positives(db):
+    """The wip head itself trains ON wip-tagged images — that's what makes
+    auto-flagging work."""
+    wip = await _system_wip(db)
+    one = await _img(db, "a" * 64)
+    two = await _img(db, "b" * 64)
+    await _img(db, "c" * 64)  # negative pool
+    await _apply(db, one.id, wip.id)
+    await _apply(db, two.id, wip.id)
+
+    ids = await db.run_sync(lambda s: head_training_ids(s, wip.id, _CFG))
+    assert ids is not None
+    pos_ids, _ = ids
+    assert set(pos_ids) == {one.id, two.id}
+
+
+@pytest.mark.asyncio
+async def test_ccip_references_skip_hygiene_images(db):
+    """A wip's figure region must never become an identity prototype, even on
+    a single-character image."""
+    ccip._REF_CACHE.update(sig=None, refs=None)
+    wip = await _system_wip(db)
+    char = await TagService(db).find_or_create("Char A", TagKind.character)
+
+    clean = await _img(db, "a" * 64)
+    flagged = await _img(db, "b" * 64)
+    for img, slot in ((clean, 0), (flagged, 1)):
+        vec = [0.0] * 768
+        vec[slot] = 1.0
+        db.add(ImageRegion(
+            image_record_id=img.id, kind="figure",
+            rx=0.0, ry=0.0, rw=1.0, rh=1.0,
+            ccip_embedding=vec, embedding_version="ccip-test",
+        ))
+        await _apply(db, img.id, char.id)
+    await _apply(db, flagged.id, wip.id)
+    await db.commit()
+
+    refs = await ccip.character_references(db)
+    vectors = refs.get(char.id, [])
+    assert len(vectors) == 1
+    assert float(vectors[0][0]) == pytest.approx(1.0)
+    assert float(vectors[0][1]) == pytest.approx(0.0)