diff --git a/docs/concepts.rst b/docs/concepts.rst
index 0da1d14..566fc67 100644
--- a/docs/concepts.rst
+++ b/docs/concepts.rst
@@ -104,7 +104,8 @@ reading is a fact about the domain the names come from: that makes it a
 square brackets — ``John [Johnny] Smith`` — instead of quotes is a
 fact about that one data source's export format, not about the
 language of the names in it — that's a
-:class:`~nameparser.Policy` (``nickname_delimiters={('[', ']')}``).
+:class:`~nameparser.Policy`
+(``nickname_delimiters=frozenset({('[', ']')})``).
 One particular report
 wanting names formatted as "Family, Given" while every other consumer
 of the same parsed data wants "Given Family" is a fact about where the
diff --git a/docs/customize.rst b/docs/customize.rst
index 6a9a7b4..a34341b 100644
--- a/docs/customize.rst
+++ b/docs/customize.rst
@@ -335,10 +335,10 @@ this section is how to switch them off, which you can do separately:
 
     >>> parse("김민준").family                    # both defaults on
     '김'
-    >>> positional = Parser(policy=Policy(script_orders={}))
+    >>> positional = Parser(policy=Policy(script_orders=()))
     >>> positional.parse("김민준").family         # still split
     '민준'
-    >>> unsplit = Parser(policy=Policy(segment_scripts=()))
+    >>> unsplit = Parser(policy=Policy(segment_scripts=frozenset()))
     >>> unsplit.parse("김민준").family            # one token, not split
     '김민준'
 
@@ -363,8 +363,9 @@ because splitting Han text requires knowing Chinese from Japanese;
 :doc:`locales` covers the opt-in ``zh`` pack that supplies them.
 
 The Japanese behaviors ride these same two fields, so they need no
-switches of their own: ``script_orders={}`` clears the kana-licensed
-entry along with the Han and Hangul ones, and ``segment_scripts=()``
+switches of their own: ``script_orders=()`` clears the kana-licensed
+entry along with the Han and Hangul ones, and
+``segment_scripts=frozenset()``
 deactivates every script at once, which also stops a parser consulting
 whatever segmenter it was given. The segmenter has an off-switch as
 well — ``Parser(segmenter=None)``, which is the default; see
@@ -389,13 +390,21 @@ off.
 
 .. note::
 
-   Both fields are annotated with their canonical *storage* type
+   Every field here is annotated with its canonical *storage* type
    rather than with everything the constructor accepts — the same as
-   ``capitalization_exceptions``. Under mypy the readable spellings
-   above (``script_orders={...}``, ``segment_scripts=(...)``) need a
-   ``# type: ignore[arg-type]``; ``script_orders=()`` and
-   ``segment_scripts=frozenset(...)`` check clean and mean the same
-   thing.
+   ``capitalization_exceptions``, and for the same reason: the
+   annotation is what you get back when you READ the attribute, which
+   is the commoner operation.
+
+   The constructor is deliberately wider. It takes any mapping for
+   ``script_orders``, any iterable of ``Script`` for
+   ``segment_scripts``, and plain strings wherever a ``Role`` is
+   wanted (``Role`` is a ``StrEnum`` precisely so that works). A
+   dataclass cannot express those two types separately, so the
+   examples in this guide use the spellings that check clean under
+   mypy — ``()`` and ``frozenset(...)`` rather than ``{}`` and a bare
+   set literal. The wider spellings parse identically; they just need
+   a ``# type: ignore[arg-type]`` if you run a type checker.
 
 Nicknames, maiden names, and brackets
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -407,7 +416,7 @@ one-liner is the whole recipe:
 
 .. doctest::
 
-    >>> policy = Policy(maiden_delimiters={("(", ")")})
+    >>> policy = Policy(maiden_delimiters=frozenset({("(", ")")}))
     >>> Parser(policy=policy).parse("Jane (Jones) Smith").maiden
     'Jones'
 
diff --git a/docs/migrate.rst b/docs/migrate.rst
index 5474858..b24f13b 100644
--- a/docs/migrate.rst
+++ b/docs/migrate.rst
@@ -427,7 +427,7 @@ now.
 
 ``Constants`` has no switch for any of this — the v1 configuration
 surface is frozen for 2.x — so the way out is the 2.0 API:
-``Parser(policy=Policy(script_orders={}, segment_scripts=()))``
+``Parser(policy=Policy(script_orders=(), segment_scripts=frozenset()))``
 restores 1.4's reading of every shape above that turns on order or
 splitting. The middle dots are the exception: both the katakana dot
 and the Chinese interpunct ``·`` (U+00B7, dividing a transcription
diff --git a/docs/modules.rst b/docs/modules.rst
index 41755f3..b19b918 100644
--- a/docs/modules.rst
+++ b/docs/modules.rst
@@ -140,9 +140,10 @@ because only these three orders have defined assignment semantics.
    the names no entry matches. The values are drawn from the same
    three constants above, and the same restriction applies. Build on it for
    additive customization —
-   ``script_orders=dict(DEFAULT_SCRIPT_ORDERS) | {Script.HAN:
-   GIVEN_FIRST}`` — and pass ``script_orders={}`` to opt out entirely
-   and get the purely positional read back. Latin-script and
+   ``script_orders=(*DEFAULT_SCRIPT_ORDERS, (Script.HAN, GIVEN_FIRST))``,
+   where a later entry for a script REPLACES an earlier one, so
+   appending is how you override — and pass ``script_orders=()`` to
+   opt out entirely and get the purely positional read back. Latin-script and
    mixed-script names are never affected either way.
 
 Delimiter defaults
diff --git a/tests/v2/test_policy.py b/tests/v2/test_policy.py
index b37aef6..a0c8ec6 100644
--- a/tests/v2/test_policy.py
+++ b/tests/v2/test_policy.py
@@ -565,6 +565,31 @@ def test_script_orders_default_and_canonical_storage() -> None:
     assert Policy(script_orders={}).script_orders == ()  # type: ignore[arg-type]
 
 
+def test_a_later_script_orders_entry_replaces_an_earlier_one() -> None:
+    """Appending is how you override a single script, and the docs now
+    say so (modules.rst, the script_orders entry).
+
+    The behavior falls out of _validated_script_orders building a dict
+    before sorting, so it was emergent rather than contracted -- and
+    prose that tells readers to append is prose that depends on it.
+    The type-clean spelling for "everything as shipped, but Han reads
+    given-first" is exactly this, which is why it is worth pinning.
+    """
+    overridden = Policy(
+        script_orders=(*DEFAULT_SCRIPT_ORDERS, (Script.HAN, GIVEN_FIRST)))
+    assert dict(overridden.script_orders)[Script.HAN] == GIVEN_FIRST
+    # the other shipped entries survive untouched
+    assert dict(overridden.script_orders)[Script.HANGUL] == FAMILY_FIRST
+    assert dict(overridden.script_orders)[Script.HIRAGANA] == FAMILY_FIRST
+    # one entry per script, whatever the input carried
+    keys = [s for s, _ in overridden.script_orders]
+    assert len(keys) == len(set(keys)) == len(DEFAULT_SCRIPT_ORDERS)
+    # and the mapping spelling it replaces still means the same thing
+    assert overridden == Policy(
+        script_orders=dict(DEFAULT_SCRIPT_ORDERS)  # type: ignore[arg-type]
+        | {Script.HAN: GIVEN_FIRST})
+
+
 def test_script_orders_validates_keys_and_values() -> None:
     with pytest.raises(ValueError, match="han, hangul"):
         Policy(script_orders={"klingon": FAMILY_FIRST})  # type: ignore[arg-type]