fix(inbox): a segment title is declared, never marked derived

A segment's title comes from the plan, so a human adjudicated it. But
structure derivation runs over the segment body, finds no title key and
no usable heading, falls back to a stem, and adds "title" to derived.
The concept then emitted a stated fact under an inferred marker, and a
consumer that distrusts derived fields would distrust exactly the thing
a human decided. An over-marked field is the same defect class as an
unmarked heuristic: the marker is only worth something if it is
accurate in both directions.

Scoped to title alone, and pinned that way by test: number stays in
derived on a segment, because nothing about segmentation makes an
inferred document number declared. Without a segment a derived title is
still marked, so only a plan makes a title declared.

The SEGMENTED_V1 golden moves, which is the intended consequence and
the only golden that may. The four existing goldens are byte-identical
to baseline 770d8d4, measured against the sha rather than inspected.

Co-Authored-By: Claude <claude-opus-5>

examples/ingest-golden-segmented/expected-bundle/index.md

@@ -2,5 +2,5 @@
 bundle_id: b-golden-segmented
 ---
 
-- [Veiledning for eksempelbundel](veiledning.md) — derived: [title]
+- [Veiledning for eksempelbundel](veiledning.md)
 - [krav (index)](krav/index.md)

examples/ingest-golden-segmented/expected-bundle/krav/1-1/foerste-krav.md

@@ -6,7 +6,6 @@ source_sha256: 6906ec0acbcfc246e825bda9863c716eb5611b465020e8204eeb448c32343f7d
 ingested_at: 2026-08-30T09:00:00Z
 generated: true
 parent: s0
-derived: [title]
 bundle_id: b-golden-segmented
 segment_id: s1
 source_offset: [94, 176]

examples/ingest-golden-segmented/expected-bundle/krav/1-1/index.md

@@ -1 +1 @@
-- [Foerste krav](foerste-krav.md) — parent: s0?; derived: [title]
+- [Foerste krav](foerste-krav.md) — parent: s0?

examples/ingest-golden-segmented/expected-bundle/krav/1-2/andre-krav.md

@@ -6,7 +6,6 @@ source_sha256: 6906ec0acbcfc246e825bda9863c716eb5611b465020e8204eeb448c32343f7d
 ingested_at: 2026-08-30T09:00:00Z
 generated: true
 parent: s0
-derived: [title]
 bundle_id: b-golden-segmented
 segment_id: s2
 source_offset: [176, 253]

examples/ingest-golden-segmented/expected-bundle/krav/1-2/index.md

@@ -1 +1 @@
-- [Andre krav](andre-krav.md) — parent: s0?; derived: [title]
+- [Andre krav](andre-krav.md) — parent: s0?

examples/ingest-golden-segmented/expected-bundle/veiledning.md

@@ -5,7 +5,6 @@ source_file: veiledning.md
 source_sha256: 6906ec0acbcfc246e825bda9863c716eb5611b465020e8204eeb448c32343f7d
 ingested_at: 2026-08-30T09:00:00Z
 generated: true
-derived: [title]
 bundle_id: b-golden-segmented
 segment_id: s0
 source_offset: [0, 94]

src/llm_ingestion_okf/inbox.py

@@ -22,7 +22,7 @@ from __future__ import annotations
 import hashlib
 import unicodedata
 from collections.abc import Callable, Mapping
-from dataclasses import dataclass
+from dataclasses import dataclass, replace
 from pathlib import Path, PurePosixPath
 
 from .errors import IngestError, MaterializationError, SegmentationError, SourceError
@@ -168,6 +168,16 @@ def render_inbox_concept(
         "generated": "true",
     }
     if structure is not None and profile.index.facets is not None:
+        if segmented:
+            # A segment's title comes from the PLAN, so it is DECLARED by the
+            # adjudicator -- but derivation runs over the segment BODY, finds
+            # no title key and no usable heading, falls back, and marks it
+            # derived. The concept then emits a stated fact under an inferred
+            # marker, and a consumer that distrusts derived fields distrusts
+            # exactly the thing a human decided. Scoped to `title` alone: every
+            # other field here really was inferred from the body, and blunting
+            # the marker would be the opposite defect.
+            structure = replace(structure, derived=structure.derived - {"title"})
         frontmatter.update(structure_frontmatter(structure, profile.index.facets.keys))
     if segmented:
         assert segment is not None

tests/test_segmented_identity.py

@@ -29,6 +29,7 @@ from llm_ingestion_okf.errors import MaterializationError
 from llm_ingestion_okf.inbox import GateDecision, process_inbox, render_inbox_concept
 from llm_ingestion_okf.profiles import DEFAULT, SEGMENTED_V1, STRUCTURED_V1
 from llm_ingestion_okf.segmentation import SegmentEntry, parse_segmentation_plan
+from llm_ingestion_okf.structure import derive_document_structure
 
 INGESTED_AT = "2026-07-25T12:00:00Z"
 
@@ -213,7 +214,8 @@ def render(**overrides: object) -> str:
         "profile": SEGMENTED_V1,
     }
     arguments.update(overrides)
-    return render_inbox_concept(DOCUMENT, **arguments)  # type: ignore[arg-type]
+    text = arguments.pop("text", DOCUMENT)
+    return render_inbox_concept(text, **arguments)  # type: ignore[arg-type]
 
 
 def frontmatter_of(document: str) -> dict[str, str]:
@@ -336,3 +338,66 @@ def test_two_bundles_hold_colliding_paths_and_disjoint_identity_values() -> None
     assert [frontmatter_of(document)["bundle_id"] for document in one] == ["b-1", "b-1"]
     assert [frontmatter_of(document)["bundle_id"] for document in other] == ["b-2", "b-2"]
     assert set(one).isdisjoint(set(other))
+
+
+# --- `derived` never marks a title the adjudicator declared ---------------
+#
+# The second measured defect. A segment's `title` comes from the PLAN, which a
+# human adjudicated; but structure derivation runs over the segment BODY, finds
+# no `title` key and no usable heading, falls back, and adds `title` to
+# `derived`. The concept then emits the declared title and marks it inferred.
+# The error points the wrong way: a consumer that distrusts derived fields ends
+# up distrusting a fact the producer stated. Marking is only worth anything if
+# it is accurate in both directions, so an over-marked field is the same defect
+# class as an unmarked heuristic.
+
+
+def structure_of(text: str, source_file: str = "veiledning.md"):
+    return derive_document_structure(text, source_file=source_file)
+
+
+def test_a_segment_title_is_never_marked_derived() -> None:
+    body = "## 1.1 Foerste krav\n\nEt krav.\n"
+    assert "title" in structure_of(body).derived  # the premise, measured
+    keys = frontmatter_of(
+        render(
+            text=body,
+            title="Foerste krav",
+            structure=structure_of(body),
+            segment=segment_entry(),
+            bundle_id="b-1",
+        )
+    )
+    assert keys["title"] == "Foerste krav"
+    assert "derived" not in keys
+
+
+def test_a_genuinely_derived_field_is_still_marked_on_a_segment() -> None:
+    """Scoped to `title` alone -- the fix must not blunt the marker.
+
+    `number` here IS inferred from the body, and nothing about segmentation
+    makes it declared, so it must survive in `derived`.
+    """
+    body = "# N500 Vegtunneler\n\nSe N100.\n"
+    derived = structure_of(body, "n500.md").derived
+    assert {"title", "number"} <= derived  # the premise, measured
+    keys = frontmatter_of(
+        render(
+            text=body,
+            title="Vegtunneler",
+            structure=structure_of(body, "n500.md"),
+            segment=segment_entry(),
+            bundle_id="b-1",
+        )
+    )
+    assert "title" not in keys["derived"]
+    assert "number" in keys["derived"]
+
+
+def test_without_a_segment_a_derived_title_is_still_marked() -> None:
+    """The negative control. Only a PLAN makes a title declared."""
+    body = "## 1.1 Foerste krav\n\nEt krav.\n"
+    keys = frontmatter_of(
+        render(text=body, title="Foerste krav", structure=structure_of(body), profile=STRUCTURED_V1)
+    )
+    assert keys["derived"] == "[title]"