feat(generate): render long-form video digest + collapsible raw evidence

src/xbrain/generate.py

@@ -440,9 +440,27 @@ def _video_digest_lines(source: ContentSourceSuccess, strings: Strings) -> list[
         return [f"> {strings.silent_video}", ""]
     heading = source.title or source.url
     lines = [f"## {strings.video_digest_header}: {heading}", ""]
+    if not source.digest:
+        # Fallback (no long-form digest yet): the raw transcript + frame embeds
+        # render inline, exactly as before — so this render change is safe to ship
+        # before any digest exists (an empty `digest` is the default).
+        if has_text:
+            lines += [source.text, ""]
+        lines += _slide_embed_lines(source.frames)
+        return lines
+    # With a digest, it is the readable headline of the section; the raw transcript
+    # + frame slides are demoted into a collapsible `<details>` below — the evidence
+    # stays in the note without the 40-frame wall of noise up top. Blank lines around
+    # the inner content let Obsidian render the markdown/embeds inside the HTML block.
+    lines += [source.digest, ""]
+    evidence: list[str] = []
     if has_text:
-        lines += [source.text, ""]
-    lines += _slide_embed_lines(source.frames)
+        evidence += [source.text, ""]
+    evidence += _slide_embed_lines(source.frames)
+    if evidence:
+        lines += ["<details>", f"<summary>{strings.video_evidence_header}</summary>", ""]
+        lines += evidence
+        lines += ["</details>", ""]
     return lines
 
 

src/xbrain/i18n.py

@@ -32,6 +32,7 @@ class Strings:
     also_relevant: str  # "También relevante" / "Also relevant"
     video_digest_header: str  # "Resumen del vídeo" / "Video digest" (#44)
     silent_video: str  # the one-line no-speech marker (#44)
+    video_evidence_header: str  # collapsible label for the raw frames + transcript
 
 
 _STRINGS: dict[str, Strings] = {
@@ -43,6 +44,7 @@ class Strings:
         also_relevant="Also relevant",
         video_digest_header="Video digest",
         silent_video="🔇 Silent video (no speech detected).",
+        video_evidence_header="Frames + transcript",
     ),
     "Spanish": Strings(
         topics_label="Temas",
@@ -52,6 +54,7 @@ class Strings:
         also_relevant="También relevante",
         video_digest_header="Resumen del vídeo",
         silent_video="🔇 Vídeo sin voz (sin transcripción).",
+        video_evidence_header="Frames y transcripción",
     ),
 }
 

src/xbrain/models.py

@@ -646,6 +646,13 @@ class ContentSourceSuccess(BaseModel):
     # one-time backward-compatible additive churn as `has_speech`/`language`. Each
     # `VideoFrame` embeds into the note like a downloaded photo (see `VideoFrame`).
     frames: list[VideoFrame] = Field(default_factory=list)
+    # Long-form readable digest of an `x_video` source — a "what it is · key
+    # points · why it matters" synthesis of the transcript (`text`) + `frames`,
+    # written by `xbrain video-digest`. Optional + additive (defaults to `""`), so
+    # every EXISTING record LOADS unchanged — a pre-digest `x_video` source (and
+    # every article source) simply carries an empty digest, and `generate` falls
+    # back to rendering the raw transcript + frames. `""` = "no digest yet".
+    digest: str = ""
     # Ordered body blocks (text + inline images) for `kind="x_article"` sources
     # captured as a structured Article (#39). Optional + additive (defaults to
     # `[]`), so every EXISTING record LOADS unchanged — a pre-#39 `x_article`

tests/test_generate.py

@@ -576,6 +576,7 @@ def _video_item(
     has_speech: bool = True,
     title: str | None = "The Talk",
     frames: list | None = None,
+    digest: str = "",
 ) -> Item:
     """An enriched video bookmark carrying an `x_video` transcript source (#44)."""
     return Item(
@@ -603,6 +604,7 @@ def _video_item(
                     text=text,
                     has_speech=has_speech,
                     frames=frames or [],
+                    digest=digest,
                 )
             ],
         ),
@@ -653,6 +655,60 @@ def test_generate_video_digest_spanish_header(tmp_path: Path):
     assert "Cuerpo de la transcripción." in note
 
 
+def test_generate_renders_video_digest_prose_when_present(tmp_path: Path):
+    """When the `x_video` source carries a long-form `digest`, it renders as the
+    body of the `## Video digest` section — the readable headline of the video."""
+    generate(
+        {"7": _video_item(text="raw transcript body", digest="This talk explains scaling laws.")},
+        tmp_path,
+    )
+    note = next((tmp_path / "items").glob("*.md")).read_text(encoding="utf-8")
+    assert "## Video digest: The Talk" in note
+    assert "This talk explains scaling laws." in note
+
+
+def test_generate_video_digest_demotes_raw_to_details_when_digest_present(tmp_path: Path):
+    """With a digest, the raw transcript + frame embeds are demoted into a collapsible
+    `<details>` below the digest prose — evidence kept, noise hidden. The digest prose
+    sits ABOVE the `<details>`, the transcript + embeds INSIDE it."""
+    generate(
+        {
+            "7": _video_item(
+                text="raw transcript body", frames=_frames(), digest="A concise digest."
+            )
+        },
+        tmp_path,
+    )
+    note = next((tmp_path / "items").glob("*.md")).read_text(encoding="utf-8")
+    assert "<details>" in note and "</details>" in note
+    digest_idx = note.index("A concise digest.")
+    details_idx = note.index("<details>")
+    assert digest_idx < details_idx  # digest prose is the headline, above the raw dump
+    inside = note[details_idx : note.index("</details>")]
+    assert "raw transcript body" in inside
+    assert "![[_media/7/frames/0.png]]" in inside
+
+
+def test_generate_video_digest_no_details_when_digest_absent(tmp_path: Path):
+    """Fallback: with no digest, the section is unchanged — transcript rendered inline,
+    no `<details>` wrapper. Safe to ship the render change before any digest exists."""
+    generate({"7": _video_item(text="inline transcript body")}, tmp_path)
+    note = next((tmp_path / "items").glob("*.md")).read_text(encoding="utf-8")
+    assert "inline transcript body" in note
+    assert "<details>" not in note
+
+
+def test_generate_video_digest_evidence_label_is_localised(tmp_path: Path):
+    """The `<details>` summary label follows the configured output language."""
+    generate(
+        {"7": _video_item(text="cuerpo", digest="Un resumen conciso.")},
+        tmp_path,
+        output_language="Spanish",
+    )
+    note = next((tmp_path / "items").glob("*.md")).read_text(encoding="utf-8")
+    assert "transcripción" in note.lower()  # the ES evidence label
+
+
 # ----------------------------------------------------------- x_video slide frames (#44 PR4)
 
 

tests/test_i18n.py

@@ -44,6 +44,7 @@ def test_strings_dataclass_has_exactly_the_expected_fields() -> None:
         "also_relevant",
         "video_digest_header",
         "silent_video",
+        "video_evidence_header",
     }