diff --git a/crudkit/core/service.py b/crudkit/core/service.py
index 86a4542..9eb6d8b 100644
--- a/crudkit/core/service.py
+++ b/crudkit/core/service.py
@@ -1,10 +1,13 @@
 from __future__ import annotations
 
+from collections.abc import Iterable
 from typing import Any, Callable, Type, TypeVar, Generic, Optional, Protocol, runtime_checkable, cast
-from sqlalchemy import and_, func, inspect, or_, text
+from sqlalchemy import and_, func, inspect, or_, text, UnaryExpression
 from sqlalchemy.engine import Engine, Connection
-from sqlalchemy.orm import Load, Session, with_polymorphic, Mapper, RelationshipProperty
+from sqlalchemy.orm import Load, Session, with_polymorphic, Mapper, contains_eager, selectinload
 from sqlalchemy.orm.attributes import InstrumentedAttribute
+from sqlalchemy.sql import operators
+from sqlalchemy.sql.elements import ColumnElement
 
 from crudkit.core.base import Version
 from crudkit.core.spec import CRUDSpec
@@ -15,13 +18,6 @@ from crudkit.projection import compile_projection
 import logging
 log = logging.getLogger("crudkit.service")
 
-def _is_rel(model_cls, name: str) -> bool:
-    try:
-        prop = model_cls.__mapper__.relationships.get(name)
-        return isinstance(prop, RelationshipProperty)
-    except Exception:
-        return False
-
 @runtime_checkable
 class _HasID(Protocol):
     id: int
@@ -44,9 +40,65 @@ class _CRUDModelProto(_HasID, _HasTable, _HasADict, Protocol):
 
 T = TypeVar("T", bound=_CRUDModelProto)
 
+def _belongs_to_alias(col: Any, alias: Any) -> bool:
+    # Try to detect if a column/expression ultimately comes from this alias.
+    # Works for most ORM columns; complex expressions may need more.
+    t = getattr(col, "table", None)
+    selectable = getattr(alias, "selectable", None)
+    return t is not None and selectable is not None and t is selectable
+
+def _paths_needed_for_sql(order_by: Iterable[Any], filters: Iterable[Any], join_paths: tuple) -> set[tuple[str, ...]]:
+    paths: set[tuple[str, ...]] = set()
+    # Sort columns
+    for ob in order_by or []:
+        col = getattr(ob, "element", ob) # unwrap UnaryExpression
+        for path, _rel_attr, target_alias in join_paths:
+            if _belongs_to_alias(col, target_alias):
+                paths.add(tuple(path))
+    # Filter columns (best-effort)
+    # Walk simple binary expressions
+    def _extract_cols(expr: Any) -> Iterable[Any]:
+        if isinstance(expr, ColumnElement):
+            yield expr
+            for ch in getattr(expr, "get_children", lambda: [])():
+                yield from _extract_cols(ch)
+        elif hasattr(expr, "clauses"):
+            for ch in expr.clauses:
+                yield from _extract_cols(ch)
+
+    for flt in filters or []:
+        for col in _extract_cols(flt):
+            for path, _rel_attr, target_alias in join_paths:
+                if _belongs_to_alias(col, target_alias):
+                    paths.add(tuple[path])
+    return paths
+
+def _paths_from_fields(req_fields: list[str]) -> set[tuple[str, ...]]:
+    out: set[tuple[str, ...]] = set()
+    for f in req_fields:
+        if "." in f:
+            parts = tuple(f.split(".")[:-1])
+            if parts:
+                out.add(parts)
+    return out
+
 def _is_truthy(val):
     return str(val).lower() in ('1', 'true', 'yes', 'on')
 
+def _normalize_fields_param(params: dict | None) -> list[str]:
+    if not params:
+        return []
+    raw = params.get("fields")
+    if isinstance(raw, (list, tuple)):
+        out: list[str] = []
+        for item in raw:
+            if isinstance(item, str):
+                out.extend([p for p in (s.strip() for s in item.split(",")) if p])
+        return out
+    if isinstance(raw, str):
+        return [p for p in (s.strip() for s in raw.split(",")) if p]
+    return []
+
 class CRUDService(Generic[T]):
     def __init__(
         self,
@@ -86,8 +138,6 @@ class CRUDService(Generic[T]):
         Normalize order_by into (cols, desc_flags). Supports plain columns and
         col.asc()/col.desc() (UnaryExpression). Avoids boolean evaluation of clauses.
         """
-        from sqlalchemy.sql import operators
-        from sqlalchemy.sql.elements import UnaryExpression
 
         given = self._stable_order_by(root_alias, given_order_by)
 
@@ -109,7 +159,6 @@ class CRUDService(Generic[T]):
             cols.append(col)
             desc_flags.append(bool(is_desc))
 
-        from crudkit.core.types import OrderSpec
         return OrderSpec(cols=tuple(cols), desc=tuple(desc_flags))
 
     def _key_predicate(self, spec: OrderSpec, key_vals: list[Any], backward: bool):
@@ -156,49 +205,100 @@ class CRUDService(Generic[T]):
         Returns a SeekWindow with items, first/last keys, order spec, limit, and optional total.
         """
         session = self.session
-        fields = list((params or {}).get("fields", []))
-        expanded_fields, proj_opts = compile_projection(self.model, fields) if fields else ([], [])
         query, root_alias = self.get_query()
-        if proj_opts:
-            query = query.options(*proj_opts)
+
+        # Normalize requested fields and compile projection (may skip later to avoid conflicts)
+        fields = _normalize_fields_param(params)
+        expanded_fields, proj_opts = compile_projection(self.model, fields) if fields else ([], [])
 
         spec = CRUDSpec(self.model, params or {}, root_alias)
 
         filters = spec.parse_filters()
         order_by = spec.parse_sort()
 
+        # Field parsing for root load_only fallback
         root_fields, rel_field_names, root_field_names = spec.parse_fields()
 
         # Soft delete filter
-        # if self.supports_soft_delete and not _is_truthy((params or {}).get("include_deleted")):
-        #     query = query.filter(getattr(root_alias, "is_deleted") == False)
         query = self._apply_not_deleted(query, root_alias, params)
 
-        # Parse filters first
+        # Apply filters first
         if filters:
             query = query.filter(*filters)
 
-        # Includes + joins (so relationship fields like brand.name, location.label work)
+        # Includes + join paths (dotted fields etc.)
         spec.parse_includes()
-        join_paths = tuple(spec.get_join_paths())
+        join_paths = tuple(spec.get_join_paths())  # iterable of (path, relationship_attr, target_alias)
 
-        for _, relationship_attr, target_alias in spec.get_join_paths():
-            rel_attr = cast(InstrumentedAttribute, relationship_attr)
-            target = cast(Any, target_alias)
-            query = query.join(target, rel_attr.of_type(target), isouter=True)
+        # Decide which relationship *names* are required for SQL (filters/sort) vs display-only
+        def _belongs_to_alias(col: Any, alias: Any) -> bool:
+            t = getattr(col, "table", None)
+            selectable = getattr(alias, "selectable", None)
+            return t is not None and selectable is not None and t is selectable
 
-        # Fields/projection: load_only for root columns, eager loads for relationships
+        # 1) which relationship aliases are referenced by sort/filter
+        sql_hops: set[str] = set()
+        for path, relationship_attr, target_alias in join_paths:
+            # If any ORDER BY column comes from this alias, mark it
+            for ob in (order_by or []):
+                col = getattr(ob, "element", ob)  # unwrap UnaryExpression
+                if _belongs_to_alias(col, target_alias):
+                    sql_hops.add(relationship_attr.key)
+                    break
+            # If any filter expr touches this alias, mark it (best effort)
+            if relationship_attr.key not in sql_hops:
+                def _walk_cols(expr: Any):
+                    # Primitive walker for ColumnElement trees
+                    from sqlalchemy.sql.elements import ColumnElement
+                    if isinstance(expr, ColumnElement):
+                        yield expr
+                        for ch in getattr(expr, "get_children", lambda: [])():
+                            yield from _walk_cols(ch)
+                    elif hasattr(expr, "clauses"):
+                        for ch in expr.clauses:
+                            yield from _walk_cols(ch)
+                for flt in (filters or []):
+                    if any(_belongs_to_alias(c, target_alias) for c in _walk_cols(flt)):
+                        sql_hops.add(relationship_attr.key)
+                        break
+
+        # 2) first-hop relationship names implied by dotted projection fields
+        proj_hops: set[str] = {f.split(".", 1)[0] for f in fields if "." in f}
+
+        # Root column projection
+        from sqlalchemy.orm import Load  # local import to match your style
         only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
         if only_cols:
             query = query.options(Load(root_alias).load_only(*only_cols))
 
+        # Relationship handling per path (avoid loader strategy conflicts)
+        used_contains_eager = False
+        for path, relationship_attr, target_alias in join_paths:
+            rel_attr = cast(InstrumentedAttribute, relationship_attr)
+            name = relationship_attr.key
+            if name in sql_hops:
+                # Needed for WHERE/ORDER BY: join + hydrate from that join
+                query = query.join(target_alias, rel_attr.of_type(target_alias), isouter=True)
+                query = query.options(contains_eager(rel_attr, alias=target_alias))
+                used_contains_eager = True
+            elif name in proj_hops:
+                # Display-only: bulk-load efficiently, no join
+                query = query.options(selectinload(rel_attr))
+            else:
+                # Not needed
+                pass
+
+        # Apply projection loader options only if they won't conflict with contains_eager
+        if proj_opts and not used_contains_eager:
+            query = query.options(*proj_opts)
+
         # Order + limit
-        order_spec = self._extract_order_spec(root_alias, order_by)  # SA2-only helper
+        order_spec = self._extract_order_spec(root_alias, order_by)  # SA 2.x helper
         limit, _ = spec.parse_pagination()
         if limit is None:
             effective_limit = 50
         elif limit == 0:
-            effective_limit = None
+            effective_limit = None  # unlimited
         else:
             effective_limit = limit
 
@@ -222,16 +322,17 @@ class CRUDService(Generic[T]):
                 query = query.limit(effective_limit)
             items = list(reversed(query.all()))
 
-
         # Tag projection so your renderer knows what fields were requested
-        if expanded_fields:
-            proj = list(expanded_fields)
+        if fields:
+            proj = list(dict.fromkeys(fields))  # dedupe, preserve order
+            if "id" not in proj and hasattr(self.model, "id"):
+                proj.insert(0, "id")
         else:
             proj = []
             if root_field_names:
                 proj.extend(root_field_names)
             if root_fields:
-                proj.extend(c.key for c in root_fields)
+                proj.extend(c.key for c in root_fields if hasattr(c, "key"))
             for path, names in (rel_field_names or {}).items():
                 prefix = ".".join(path)
                 for n in names:
@@ -248,7 +349,7 @@ class CRUDService(Generic[T]):
 
         # Boundary keys for cursor encoding in the API layer
         first_key = self._pluck_key(items[0], order_spec) if items else None
-        last_key  = self._pluck_key(items[-1], order_spec) if items else None
+        last_key = self._pluck_key(items[-1], order_spec) if items else None
 
         # Optional total that’s safe under JOINs (COUNT DISTINCT ids)
         total = None
@@ -257,10 +358,11 @@ class CRUDService(Generic[T]):
             base = self._apply_not_deleted(base, root_alias, params)
             if filters:
                 base = base.filter(*filters)
-            for _, relationship_attr, target_alias in join_paths:   # reuse
-                rel_attr = cast(InstrumentedAttribute, relationship_attr)
-                target = cast(Any, target_alias)
-                base = base.join(target, rel_attr.of_type(target), isouter=True)
+            # Mirror join structure for any SQL-needed relationships
+            for path, relationship_attr, target_alias in join_paths:
+                if relationship_attr.key in sql_hops:
+                    rel_attr = cast(InstrumentedAttribute, relationship_attr)
+                    base = base.join(target_alias, rel_attr.of_type(target_alias), isouter=True)
             total = session.query(func.count()).select_from(
                 base.order_by(None).distinct().subquery()
             ).scalar() or 0
@@ -270,7 +372,6 @@ class CRUDService(Generic[T]):
         if log.isEnabledFor(logging.DEBUG):
             log.debug("QUERY: %s", str(query))
 
-        from crudkit.core.types import SeekWindow  # avoid circulars at module top
         return SeekWindow(
             items=items,
             limit=window_limit_for_body,
@@ -311,50 +412,81 @@ class CRUDService(Generic[T]):
         return [*order_by, *pk_cols]
 
     def get(self, id: int, params=None) -> T | None:
+        """Fetch a single row by id with conflict-free eager loading and clean projection."""
         query, root_alias = self.get_query()
 
-        include_deleted = False
-        root_fields = []
-        root_field_names = {}
-        rel_field_names = {}
+        # Defaults so we can build a projection even if params is None
+        root_fields: list[Any] = []
+        root_field_names: dict[str, str] = {}
+        rel_field_names: dict[tuple[str, ...], list[str]] = {}
+        req_fields: list[str] = _normalize_fields_param(params)
+
+        # Soft-delete guard
+        query = self._apply_not_deleted(query, root_alias, params)
 
         spec = CRUDSpec(self.model, params or {}, root_alias)
-        if params:
-            if self.supports_soft_delete:
-                include_deleted = _is_truthy(params.get('include_deleted'))
-        if self.supports_soft_delete and not include_deleted:
-            query = query.filter(getattr(root_alias, "is_deleted") == False)
+
+        # Optional extra filters (in addition to id); keep parity with list()
+        filters = spec.parse_filters()
+        if filters:
+            query = query.filter(*filters)
+
+        # Always filter by id
         query = query.filter(getattr(root_alias, "id") == id)
 
+        # Includes + join paths we may need
         spec.parse_includes()
+        join_paths = tuple(spec.get_join_paths())
 
-        for _, relationship_attr, target_alias in spec.get_join_paths():
-            rel_attr = cast(InstrumentedAttribute, relationship_attr)
-            target = cast(Any, target_alias)
-            query = query.join(target, rel_attr.of_type(target), isouter=True)
-
+        # Field parsing to enable root load_only
         if params:
             root_fields, rel_field_names, root_field_names = spec.parse_fields()
 
-        req_fields = list((params or {}).get("fields", []))
-        expanded_fields, proj_opts = compile_projection(self.model, req_fields) if req_fields else ([], [])
-        if proj_opts:
-            query = query.options(*proj_opts)
+        # Decide which relationship paths are needed for SQL vs display-only
+        # For get(), there is no ORDER BY; only filters might force SQL use.
+        sql_paths = _paths_needed_for_sql(order_by=None, filters=filters, join_paths=join_paths)
+        proj_paths = _paths_from_fields(req_fields)
 
+        # Root column projection
         only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
         if only_cols:
             query = query.options(Load(root_alias).load_only(*only_cols))
 
+        # Relationship handling per path: avoid loader strategy conflicts
+        used_contains_eager = False
+        for path, relationship_attr, target_alias in join_paths:
+            rel_attr = cast(InstrumentedAttribute, relationship_attr)
+            ptuple = tuple(path)
+            if ptuple in sql_paths:
+                # Needed in WHERE: join + hydrate from the join
+                query = query.join(target_alias, rel_attr.of_type(target_alias), isouter=True)
+                query = query.options(contains_eager(rel_attr, alias=target_alias))
+                used_contains_eager = True
+            elif ptuple in proj_paths:
+                # Display-only: bulk-load efficiently
+                query = query.options(selectinload(rel_attr))
+            else:
+                pass
+
+        # Projection loader options compiled from requested fields.
+        # Skip if we used contains_eager to avoid strategy conflicts.
+        expanded_fields, proj_opts = compile_projection(self.model, req_fields) if req_fields else ([], [])
+        if proj_opts and not used_contains_eager:
+            query = query.options(*proj_opts)
+
         obj = query.first()
 
-        if expanded_fields:
-            proj = list(expanded_fields)
+        # Emit exactly what the client requested (plus id), or a reasonable fallback
+        if req_fields:
+            proj = list(dict.fromkeys(req_fields))  # dedupe, preserve order
+            if "id" not in proj and hasattr(self.model, "id"):
+                proj.insert(0, "id")
         else:
             proj = []
             if root_field_names:
                 proj.extend(root_field_names)
             if root_fields:
-                proj.extend(c.key for c in root_fields)
+                proj.extend(c.key for c in root_fields if hasattr(c, "key"))
             for path, names in (rel_field_names or {}).items():
                 prefix = ".".join(path)
                 for n in names:
@@ -374,40 +506,60 @@ class CRUDService(Generic[T]):
         return obj or None
 
     def list(self, params=None) -> list[T]:
+        """Offset/limit listing with smart relationship loading and clean projection."""
         query, root_alias = self.get_query()
 
-        root_fields = []
-        root_field_names = {}
-        rel_field_names = {}
+        # Defaults so we can reference them later even if params is None
+        root_fields: list[Any] = []
+        root_field_names: dict[str, str] = {}
+        rel_field_names: dict[tuple[str, ...], list[str]] = {}
+        req_fields: list[str] = _normalize_fields_param(params)
 
         if params:
-            if self.supports_soft_delete:
-                include_deleted = _is_truthy(params.get('include_deleted'))
-                if not include_deleted:
-                    query = query.filter(getattr(root_alias, "is_deleted") == False)
+            query = self._apply_not_deleted(query, root_alias, params)
 
             spec = CRUDSpec(self.model, params or {}, root_alias)
             filters = spec.parse_filters()
             order_by = spec.parse_sort()
             limit, offset = spec.parse_pagination()
+
+            # Includes + join paths we might need
             spec.parse_includes()
+            join_paths = tuple(spec.get_join_paths())
 
-            for _, relationship_attr, target_alias in spec.get_join_paths():
-                rel_attr = cast(InstrumentedAttribute, relationship_attr)
-                target = cast(Any, target_alias)
-                query = query.join(target, rel_attr.of_type(target), isouter=True)
-
-            if params:
-                root_fields, rel_field_names, root_field_names = spec.parse_fields()
-
-            only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
-            if only_cols:
-                query = query.options(Load(root_alias).load_only(*only_cols))
+            # Field parsing for load_only on root columns
+            root_fields, rel_field_names, root_field_names = spec.parse_fields()
 
             if filters:
                 query = query.filter(*filters)
 
-            # MSSQL: requires ORDER BY when using OFFSET (and SQLA will use OFFSET for limit+offset).
+            # Determine which relationship paths are needed for SQL vs display-only
+            sql_paths = _paths_needed_for_sql(order_by, filters, join_paths)
+            proj_paths = _paths_from_fields(req_fields)
+
+            # Root column projection
+            only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
+            if only_cols:
+                query = query.options(Load(root_alias).load_only(*only_cols))
+
+            # Relationship handling per path
+            used_contains_eager = False
+            for path, relationship_attr, target_alias in join_paths:
+                rel_attr = cast(InstrumentedAttribute, relationship_attr)
+                ptuple = tuple(path)
+                if ptuple in sql_paths:
+                    # Needed for WHERE/ORDER BY: join + hydrate from the join
+                    query = query.join(target_alias, rel_attr.of_type(target_alias), isouter=True)
+                    query = query.options(contains_eager(rel_attr, alias=target_alias))
+                    used_contains_eager = True
+                elif ptuple in proj_paths:
+                    # Display-only: no join, bulk-load efficiently
+                    query = query.options(selectinload(rel_attr))
+                else:
+                    # Not needed at all; do nothing
+                    pass
+
+            # MSSQL requires ORDER BY when OFFSET is used (SQLA uses OFFSET for limit/offset)
             paginating = (limit is not None) or (offset is not None and offset != 0)
             if paginating and not order_by and self.backend.requires_order_by_for_offset:
                 order_by = self._default_order_by(root_alias)
@@ -415,27 +567,37 @@ class CRUDService(Generic[T]):
             if order_by:
                 query = query.order_by(*order_by)
 
-            # Only apply offset/limit when not None.
+            # Only apply offset/limit when not None and not zero
             if offset is not None and offset != 0:
                 query = query.offset(offset)
             if limit is not None and limit > 0:
                 query = query.limit(limit)
 
-        req_fields = list((params or {}).get("fields", []))
-        expanded_fields, proj_opts = compile_projection(self.model, req_fields) if req_fields else ([], [])
-        if proj_opts:
-            query = query.options(*proj_opts)
+            # Projection loader options compiled from requested fields.
+            # Skip if we used contains_eager to avoid loader-strategy conflicts.
+            expanded_fields, proj_opts = compile_projection(self.model, req_fields) if req_fields else ([], [])
+            if proj_opts and not used_contains_eager:
+                query = query.options(*proj_opts)
+
+        else:
+            # No params means no filters/sorts/limits; still honor projection loaders if any
+            expanded_fields, proj_opts = compile_projection(self.model, req_fields) if req_fields else ([], [])
+            if proj_opts:
+                query = query.options(*proj_opts)
 
         rows = query.all()
 
-        if expanded_fields:
-            proj = list(expanded_fields)
+        # Emit exactly what the client requested (plus id), or a reasonable fallback
+        if req_fields:
+            proj = list(dict.fromkeys(req_fields))  # dedupe while preserving order
+            if "id" not in proj and hasattr(self.model, "id"):
+                proj.insert(0, "id")
         else:
             proj = []
             if root_field_names:
                 proj.extend(root_field_names)
             if root_fields:
-                proj.extend(c.key for c in root_fields)
+                proj.extend(c.key for c in root_fields if hasattr(c, "key"))
             for path, names in (rel_field_names or {}).items():
                 prefix = ".".join(path)
                 for n in names: