diff --git a/alfred/domain/release/ports/__init__.py b/alfred/domain/release/ports/__init__.py
new file mode 100644
index 0000000..b4ae2ec
--- /dev/null
+++ b/alfred/domain/release/ports/__init__.py
@@ -0,0 +1,10 @@
+"""Domain ports for the release domain.
+
+Protocol-based abstractions that decouple ``parse_release`` and
+``ParsedRelease`` from any concrete knowledge-base loader. The
+infrastructure layer provides the adapter that satisfies this contract.
+"""
+
+from .knowledge import ReleaseKnowledge
+
+__all__ = ["ReleaseKnowledge"]
diff --git a/alfred/domain/release/ports/knowledge.py b/alfred/domain/release/ports/knowledge.py
new file mode 100644
index 0000000..272e7ef
--- /dev/null
+++ b/alfred/domain/release/ports/knowledge.py
@@ -0,0 +1,52 @@
+"""ReleaseKnowledge port — the read-only query surface that
+``parse_release`` and ``ParsedRelease`` need from the release knowledge
+base, expressed as a structural Protocol so the domain never imports any
+concrete loader.
+
+The concrete YAML-backed implementation lives in
+``alfred/infrastructure/knowledge/release_kb.py``. Tests can supply any
+object that satisfies this shape (e.g. a simple dataclass).
+"""
+
+from __future__ import annotations
+
+from typing import Protocol
+
+
+class ReleaseKnowledge(Protocol):
+    """Read-only snapshot of release-name parsing knowledge."""
+
+    # --- Token sets used by the tokenizer / matchers ---
+
+    resolutions: set[str]
+    sources: set[str]
+    codecs: set[str]
+    language_tokens: set[str]
+    forbidden_chars: set[str]
+    hdr_extra: set[str]
+
+    # --- Structured knowledge (loaded from YAML as dicts) ---
+
+    audio: dict
+    video_meta: dict
+    editions: dict
+    media_type_tokens: dict
+
+    # --- Tokenizer separators ---
+
+    separators: list[str]
+
+    # --- File-extension sets (used by application/infra modules that work
+    #     directly with filesystem paths, e.g. media-type detection, video
+    #     lookup). Domain parsing itself doesn't touch these. ---
+
+    video_extensions: set[str]
+    non_video_extensions: set[str]
+    subtitle_extensions: set[str]
+    metadata_extensions: set[str]
+
+    # --- Filesystem sanitization (Option B: pre-sanitize at parse time) ---
+
+    def sanitize_for_fs(self, text: str) -> str:
+        """Strip filesystem-forbidden characters from ``text``."""
+        ...
diff --git a/alfred/infrastructure/knowledge/release_kb.py b/alfred/infrastructure/knowledge/release_kb.py
new file mode 100644
index 0000000..5d4a790
--- /dev/null
+++ b/alfred/infrastructure/knowledge/release_kb.py
@@ -0,0 +1,83 @@
+"""YamlReleaseKnowledge — concrete adapter for the ``ReleaseKnowledge``
+domain port.
+
+Loads every release-knowledge YAML once at construction time and exposes
+the parsed snapshots as plain attributes. The application layer builds a
+single instance at boot and passes it down to ``parse_release`` and to
+``ParsedRelease`` builder methods.
+
+A few extras (``video_extensions``, ``non_video_extensions``,
+``subtitle_extensions``, ``metadata_extensions``) are not part of the
+domain port — they are consumed by application/infra modules that handle
+filesystem-level concerns.
+"""
+
+from __future__ import annotations
+
+from .release import (
+    load_audio,
+    load_codecs,
+    load_editions,
+    load_forbidden_chars,
+    load_hdr_extra,
+    load_language_tokens,
+    load_media_type_tokens,
+    load_metadata_extensions,
+    load_non_video_extensions,
+    load_resolutions,
+    load_separators,
+    load_sources,
+    load_sources_extra,
+    load_subtitle_extensions,
+    load_video,
+    load_video_extensions,
+    load_win_forbidden_chars,
+)
+
+
+class YamlReleaseKnowledge:
+    """Single object holding every parsed-release knowledge constant.
+
+    Built once at application boot. Read-only at runtime — call sites
+    treat it as a snapshot. To pick up newly learned tokens without a
+    restart, build a fresh instance and swap it in at the call sites.
+    """
+
+    def __init__(self) -> None:
+        # Domain-port surface
+        self.resolutions: set[str] = load_resolutions()
+        self.sources: set[str] = load_sources() | load_sources_extra()
+        self.codecs: set[str] = load_codecs()
+        self.language_tokens: set[str] = load_language_tokens()
+        self.forbidden_chars: set[str] = load_forbidden_chars()
+        self.hdr_extra: set[str] = load_hdr_extra()
+
+        self.audio: dict = load_audio()
+        self.video_meta: dict = load_video()
+        self.editions: dict = load_editions()
+        self.media_type_tokens: dict = load_media_type_tokens()
+
+        self.separators: list[str] = load_separators()
+
+        # File-extension sets (used by application/infra modules, not by
+        # the parser itself — kept here so there is a single ownership
+        # point for release knowledge).
+        self.video_extensions: set[str] = load_video_extensions()
+        self.non_video_extensions: set[str] = load_non_video_extensions()
+        self.subtitle_extensions: set[str] = load_subtitle_extensions()
+        # Metadata + subtitle extensions are both ignored when deciding
+        # the media type of a folder (neither is a conclusive signal for
+        # movie/tv/other), so we expose the union under the historical
+        # name.
+        self.metadata_extensions: set[str] = (
+            load_metadata_extensions() | self.subtitle_extensions
+        )
+
+        # Translation table for stripping Windows-forbidden chars.
+        self._win_forbidden_table = str.maketrans(
+            "", "", "".join(load_win_forbidden_chars())
+        )
+
+    def sanitize_for_fs(self, text: str) -> str:
+        """Strip Windows-forbidden characters from ``text``."""
+        return text.translate(self._win_forbidden_table)