# UNDF: UNDF-2026-000000542
# sqlalchemy-0003: evaluated_keys list → set in _apply_evaluators()

## Severity
MEDIUM

## Location
`lib/sqlalchemy/orm/bulk_persistence.py` — `_apply_evaluators()` (approx. line 1873)

## Description
`evaluated_keys = list(value_evaluators.keys())` converts a dict's key view
to a plain list.  The list is then used for:

1. `c.key not in evaluated_keys` — O(K) membership scan inside a set
   comprehension over `prefetch_cols` (one test per column)
2. `.difference(evaluated_keys)` — called on a set; this is O(P×K) rather
   than the O(P) it would be with a set argument

Fix: `evaluated_keys = set(value_evaluators)` (or just use `value_evaluators`
directly for membership, since dict `in` is O(1)).

## Defective code (lines 1873–1889)
```python
evaluated_keys = list(value_evaluators.keys())   # ← plain list

to_prefetch = {
    c
    for c in prefetch_cols
    if c.key in effective_params
    and c in mapper._columntoproperty
    and c.key not in evaluated_keys              # ← O(K) linear scan per column
}
to_expire = {
    mapper._columntoproperty[c].key
    for c in postfetch_cols
    if c in mapper._columntoproperty
}.difference(evaluated_keys)                     # ← set.difference(list) is O(P×K)
```

## Fix
```diff
--- a/lib/sqlalchemy/orm/bulk_persistence.py
+++ b/lib/sqlalchemy/orm/bulk_persistence.py
@@ -1873,1 +1873,1 @@
-        evaluated_keys = list(value_evaluators.keys())
+        evaluated_keys = set(value_evaluators)
```

## Complexity
- Before: O(C×K) for the `to_prefetch` comprehension, O(P×K) for `.difference()`
- After:  O(C) + O(P)  — both operations drop to O(1) membership