fix(okf): top-level frontmatter title survives a nested sources: title

P15 (order 20260912T220951Z). okf._frontmatter_from_text was linewise
last-write-wins over EVERY line regardless of indentation, so a curated
concept's own top-level `title:` got silently overwritten by the nested
`sources:\n  - title: ...` block's title. Fix: a top-level (unindented)
key always wins over an indented one of the same name; a nested line
with no top-level counterpart is still preserved (SPEC §4).

Red-before/green-after: new test
test_parse_frontmatter_top_level_title_survives_nested_sources_title
(tests/test_okf.py) failed on 45edbf5 (fm["title"] == "N500:2024",
expected the concept's own), green after the fix.

Re-measured on all four vegnormal-okf bases (concept files / distinct
titles): n100-2023 446/446 (was 1) - n200-2024 1133/1133 (was 1) -
n500-2024 270/270 (was 1) - r761-2025 2756/2407 (genuine repeated
process names, not a collapse). directory_listing on krav/N500:
269 documents / 269 distinct titles (was 1).

tests/test_context_sets_loadbearing.py:
- The P14 tripwire test (asserting parse_frontmatter DID collapse
  titles) is INVERTED, not deleted, per the order: it now asserts the
  fix holds, as a live regression guard.
- own_frontmatter() stays (not replaced by parse_frontmatter): measured
  29,500 field reads (type/title/req_number/prosessnr, all four bases)
  agree exactly except for quote-stripping (2,728/29,500, zero value
  mismatches) - own_frontmatter unquotes for fasit comparison,
  parse_frontmatter deliberately doesn't (D1/(a)/(i): unquote_scalar is
  the ONE unquoting rule).

docs/2026-09-12-p14-kontekstsett.md Part B correction: the "22 of 22
cost words absent from n100/n200/n500" claim was false - n500-2024
carries `kroner` as a false positive (substring match inside
"borkroner", drill bits, not money). The original 22-word list was
never persisted, so only ~9 of the 22 survive named. Replaced with a
newly named, persisted 22-word list and the actual re-measured count:
n100 22/22 absent - n200 22/22 - n500 21/22 (kroner via borkroner) -
r761 18/22 (4 genuine cost words). No gate touched (no fasit anchor is
`kroner`).

Verification: full suite 1643 passed / 5 skipped (was 1642/5 on
45edbf5, +1 new test, 0 removed) - `uv run pytest -q`. ruff check +
ruff format --check clean on the three changed source/test files.
Golden transcripts byte-unchanged: shasum -a 1
tests/golden/demo-transcript.stdout = ea8c534773acdbe41ae68f2c55724d69aaf8be4f,
demo-transcript.stderr = ede3e2f685ce6a14ad9888e9de421d1a66f6c611.
No version bump, no push (both forbidden by the order).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs/2026-09-12-p14-kontekstsett.md

@@ -119,17 +119,42 @@ Den døra er ikke stresstestens sti — stresstesten går `--mandate` over den o
 
 ### 2.2 Falsifiseringsgaten — og hvorfor den er så skarp her
 
-Hvert sett bærer minst to spørsmål basen **ikke kan besvare**, der riktig svar er å si det.
+**RETTET 2026-09-13 (P15, ordre `20260912T220951Z`).** Den opprinnelige setningen her — «samtlige 22
-MÅLT over alle fire basene, 22 prøvde kostnadsord:
+er FRAVÆRENDE fra n100/n200/n500» — var USANN og er fjernet. Kun 8 av de 22 opprinnelig prøvde
+ordene ble noensinne navngitt (listen over, med et avsluttende «…»), og selve 22-ordslisten ble
+ALDRI persistert noe sted i repoet — den kan derfor ikke re-måles verbatim. Det er en
+nevner-svikt av Verifiseringslovens ansikt 4-type: et «fraværende»-utsagn uten en gjenfinnbar
+liste er ikke et mål, det er en påstand om et mål som fant sted.
 
-- **n100-2023, n200-2024, n500-2024: samtlige 22 er FRAVÆRENDE fra hele basen** — `enhetspris`,
+**Ny, navngitt og persistert 22-ordsliste** (2026-09-13, denne rettelsen — velges for domene-treffsikkerhet, ikke for å oppnå et bestemt utfall), re-målt med gatens egen
-  `kroner`, `budsjett`, `kostnadsestimat`, `prisskjema`, `timepris`, `nåverdi`, `driftskostnad` …
+substring-regel (`anchors_are_absent`, `tests/test_context_sets_loadbearing.py`) mot alle fire
-- **r761-2025:** 18 av 22 fraværende; basen bærer `kroner`, `budsjett`, `enhetspris` og `hydrogen`,
+basene:
-  fordi prosesskoden ER kontraktsspråk.
 
-Det er et funn i seg selv, og det er hele grunnen til at gaten er verdt å teste: po sitt oppdrag er
+`enhetspris`, `kroner`, `budsjett`, `kostnadsestimat`, `prisskjema`, `timepris`, `nåverdi`,
-å finne **kostnadsbesparelser**, og tre av fire baser inneholder ikke ett eneste pengeord. En
+`driftskostnad`, `anleggskostnad`, `materialkostnad`, `arbeidskostnad`, `investeringskostnad`,
-kjøring som svarer med et kronebeløp mot N100 har ikke lest basen — den har funnet på.
+`vedlikeholdskostnad`, `kapitalkostnad`, `finansieringskostnad`, `livssykluskostnad`,
+`kontraktssum`, `anbudspris`, `tilbudspris`, `merverdiavgift`, `avskrivning`,
+`besparelsespotensial`.
+
+**Faktisk telling per base (446/1133/270/2756 konsepter):**
+
+- **n100-2023: 22 av 22 fraværende.**
+- **n200-2024: 22 av 22 fraværende.**
+- **n500-2024: 21 av 22 fraværende — IKKE 22.** Basen bærer `kroner`, men som en FALSK POSITIV:
+  ordet forekommer kun inni `borkroner` (boreutstyr, ikke penger) i
+  `krav/N500/id-41a2f459-e362-42d8-d725-aaee8290c7c1.md` — «(styrestenger, borkroner), nøyaktighet
+  ved a…». Delstreng-regelen i Rule U treffer bevisst uten ordgrenser (§ 2.3), og dette er den
+  konkrete prisen for det: en ekte nulltreff-base kan likevel «bære» et pengeord gjennom et
+  urelatert sammensatt ord. Funnet endrer INGEN gate — ingen fasit-anchor er `kroner` — men det
+  gjør STATE-påstanden usann og er derfor rettet her.
+- **r761-2025: 18 av 22 fraværende** (uendret konklusjon fra tidligere, men re-målt mot den nye
+  listen): basen bærer `enhetspris`, `kroner`, `budsjett` og `kapitalkostnad` — alle fire er ekte
+  kostnadsspråk (f.eks. «kapitalkostnader», «mengder i kroner avviker fra summen»), konsistent med
+  at prosesskoden ER kontraktsspråk.
+
+Konklusjonen står: tre av fire baser er reelt uten kostnadsspråk (n500s ene treff er en
+falsk positiv, ikke et pengeord), og det er hele grunnen til at gaten er verdt å teste. En kjøring
+som svarer med et kronebeløp mot N100 har ikke lest basen — den har funnet på.
 
 ### 2.3 Regelen for «kan ikke besvares» (DEL 3c) — skrevet ned
 

src/portfolio_optimiser/okf.py

@@ -105,13 +105,34 @@ def _split_frontmatter(text: str) -> tuple[list[str], str, bool]:
 
 def _frontmatter_from_text(text: str) -> dict[str, str]:
     """``parse_frontmatter``'s rule, applied to already-read text: every frontmatter line that
-    carries a colon becomes one ``key: value`` pair, last write winning. Unterminated blocks are
+    carries a colon becomes one ``key: value`` pair, last write winning AMONG LINES AT THE SAME
-    parsed as if closed — the pre-split behaviour, preserved deliberately."""
+    LEVEL. Unterminated blocks are parsed as if closed — the pre-split behaviour, preserved
+    deliberately.
+
+    A TOP-LEVEL key (no leading whitespace — ``_split_frontmatter`` preserves indentation
+    verbatim, so this is a real structural signal, not a guess) always wins over an indented one
+    of the same name: a nested line is a block-sequence/-mapping entry under some OTHER top-level
+    key (``sources:\\n  - title: …`` names the SOURCE, not the concept) and must never overwrite
+    the concept's own field, however late it appears in the scan. Measured 2026-09-13 on
+    vegnormal-okf concepts: without this, every ``sources:``-bearing file's nested ``title``
+    replaced its own, collapsing 4605 concepts to 4 distinct titles. A nested line with no
+    top-level counterpart is still preserved (OKF SPEC §4, "unknown fields are preserved")."""
     fm: dict[str, str] = {}
+    top_level: set[str] = set()
     for line in _split_frontmatter(text)[0]:
+        if not line:
+            continue
         key, sep, val = line.partition(":")
-        if sep:
+        if not sep:
-            fm[key.strip()] = val.strip()
+            continue
+        key = key.strip()
+        if line[0].isspace():
+            if key in top_level:
+                continue
+            fm[key] = val.strip()
+        else:
+            fm[key] = val.strip()
+            top_level.add(key)
     return fm
 
 

tests/test_context_sets_loadbearing.py

@@ -57,22 +57,31 @@ _MIN_ANCHOR_CHARS = 4
 def own_frontmatter(path: Path) -> dict[str, str]:
     """The concept's OWN frontmatter: top-level keys only, FIRST occurrence winning.
 
-    **``okf.parse_frontmatter`` cannot be used for this, and that is measured rather than assumed.**
+    **P15 (2026-09-13) fixed the finding this helper was written against.** Before P15,
-    It is linewise and last-write-wins by documented design, so a nested block overwrites a
+    ``okf.parse_frontmatter`` was linewise and last-write-wins over EVERY line regardless of
-    top-level key of the same name. Every vegnormal concept ends its frontmatter with
+    indentation, so a nested block overwrote a top-level key of the same name. Every vegnormal
+    concept ends its frontmatter with
 
         sources:
           - resource: https://…
             title: N500:2024
 
-    and the indented ``title`` therefore replaces the concept's own. MEASURED on n500-2024:
+    and the indented ``title`` used to replace the concept's own. MEASURED on n500-2024 before the
-    ``okf.navigate_bundle`` yields 270 concept files carrying **1 distinct title** (``N500:2024``,
+    fix: ``okf.navigate_bundle`` yielded 270 concept files carrying **1 distinct title**
-    270 times), and ``okf.directory_listing(bundle, path="krav/N500")`` returns 269 documents whose
+    (``N500:2024``, 270 times). ``okf.parse_frontmatter`` now makes indentation load-bearing —
-    ``title`` is that same string — so rung 2 and rung 3 of the navigation ladder tell a navigator
+    a top-level (unindented) key always wins over a nested one of the same name — and re-measured
-    apart only by an opaque UUID filename and a character count. That is a finding about the
+    AFTER the fix, the same base's 269 ``krav/N500`` documents carry **269 distinct titles**.
-    PRODUCT (recorded in ``docs/2026-09-12-p14-kontekstsett.md § 5``), not about this test; what it
+
-    means here is that a fasit assert against ``parse_frontmatter``'s ``title`` would compare every
+    **This helper still isn't a plain call to ``okf.parse_frontmatter``, and that remains
-    concept to the same constant and be VACUOUS.
+    measured rather than assumed:** ``own_frontmatter`` also strips one layer of enclosing
+    ``'`` quotes (``.strip("'")``) so a value matches the fasit's stored plain-text title
+    verbatim, while ``okf.parse_frontmatter`` deliberately leaves scalars quoted — unquoting is
+    ``okf.unquote_scalar``'s ONE job (D1/(a)/(i)), and a second copy of that rule here would be
+    the drifting one. Re-measured across all four bases (29 500 field reads: ``type``, ``title``,
+    ``req_number``, ``prosessnr`` on every concept file) the two now agree EXACTLY except for
+    quoted scalars (2 728 of 29 500 checks — every one a quote-stripping difference, none a value
+    difference), so this helper stays for that one reason, not for the nested-override bug P15
+    closed.
 
     Uses ``okf._split_frontmatter`` deliberately: it is the module's ONE place ``---`` is compared
     (B4), and a second delimiter rule here would be the copy that drifts.
@@ -376,14 +385,16 @@ def test_the_fasit_titles_are_distinct_not_the_collapsed_sources_title(set_dir:
     """The fasit's recorded titles must tell the cited concepts APART.
 
     Paired with ``own_frontmatter``'s measurement, this is what keeps arm (b) from being vacuous: if
-    the recorded titles were ``parse_frontmatter``'s, every one of them would be the base's
+    the recorded titles were ``parse_frontmatter``'s PRE-P15 titles, every one of them would be the
-    ``sources`` title and the assert would hold against any concept in the base.
+    base's ``sources`` title and the assert would hold against any concept in the base.
 
-    **TRIPWIRE, deliberately.** The second half asserts that ``okf.parse_frontmatter`` DOES still
+    **Former tripwire, INVERTED 2026-09-13 by P15 (deliberately, per that order — not deleted).**
-    collapse them. The day that stops being true — an okf bump, or a decision to read block
+    Until P15 the second half asserted that ``okf.parse_frontmatter`` DID still collapse the
-    mappings here — this arm goes red, and whoever sees it should read this docstring, confirm the
+    titles, as a red flag that would fire the day the production bug was fixed here instead of at
-    listing now carries real titles, and DELETE this half rather than weaken it.
+    the call site. P15 fixed it AT THE SOURCE (``okf._frontmatter_from_text``: a top-level key now
-    """
+    always wins over a nested one of the same name), so the second half now asserts the opposite —
+    that ``parse_frontmatter`` agrees with the fasit's own distinct titles — as a live regression
+    guard against the collapse coming back."""
     declared = read_bundle_txt(set_dir / "bundle.txt")
     base = _require_base(declared)
     fasit = json.loads((set_dir / "fasit.json").read_text(encoding="utf-8"))
@@ -391,8 +402,8 @@ def test_the_fasit_titles_are_distinct_not_the_collapsed_sources_title(set_dir:
     cited = [c for row in fasit["must_cite"] for c in row["concepts"]]
     assert len({c["title"] for c in cited}) == len(cited), "recorded titles do not tell them apart"
 
-    collapsed = {okf.parse_frontmatter(base / c["path"]).get("title", "") for c in cited}
+    titles = {okf.parse_frontmatter(base / c["path"]).get("title", "") for c in cited}
-    assert len(collapsed) == 1, (
+    assert len(titles) == len(cited), (
-        "okf.parse_frontmatter no longer collapses these titles onto the sources block — read this "
+        "okf.parse_frontmatter collapsed these titles onto the sources block again — the P15 fix "
-        "test's docstring: the finding it guards may be gone, and this half should be deleted"
+        "in okf._frontmatter_from_text has regressed"
     )

tests/test_okf.py

@@ -315,6 +315,38 @@ def test_parse_frontmatter_reads_scalar_fields() -> None:
     assert fm["decision"] == "approved_with_adjustment"
 
 
+def test_parse_frontmatter_top_level_title_survives_nested_sources_title(tmp_path) -> None:
+    """A concept's OWN ``title`` sits at top level; ``sources:`` is a block sequence whose nested
+    ``title:`` names the SOURCE document, not the concept (measured on a real vegnormal-okf
+    concept, ``krav/N500/id-bfb0edb4-…``). Top-level keys carry no indentation and must win —
+    a nested line must never overwrite a top-level key of the same name, however late it appears
+    in the scan. Without this, ``directory_listing`` on ``krav/N500`` returns 269 documents that
+    all share the one nested title, ``N500:2024`` — rung 2/3 of the navigation ladder collapse to
+    an opaque UUID filename and a character count (P14 finding,
+    ``docs/2026-09-12-p14-kontekstsett.md`` § 5).
+
+    Known-negatives that must stay green: ``type`` and ``req_number`` are untouched top-level
+    scalars either way, and the ``verified:`` block-form decoder (SPEC §5.2) reads
+    ``_split_frontmatter``'s lines directly and is unaffected by this fix."""
+    text = (
+        "---\n"
+        "type: Krav\n"
+        "title: Krav 10.4.3—1 Mekanisk ventilasjon (impulsventilator)\n"
+        "req_number: Krav 10.4.3—1\n"
+        "sources:\n"
+        "  - resource: https://example.invalid/859990\n"
+        "    title: N500:2024\n"
+        "---\n\n"
+        "## Krav\nbody\n"
+    )
+    path = tmp_path / "concept.md"
+    path.write_text(text, encoding="utf-8")
+    fm = okf.parse_frontmatter(path)
+    assert fm["title"] == "Krav 10.4.3—1 Mekanisk ventilasjon (impulsventilator)"
+    assert fm["type"] == "Krav"
+    assert fm["req_number"] == "Krav 10.4.3—1"
+
+
 def test_render_frontmatter_roundtrips_consumed_fields(tmp_path) -> None:
     """Step-8 writer: ``render_frontmatter`` + ``write_concept_file`` emit a block that
     ``parse_frontmatter`` re-reads with the fields ``seed_store_from_bundle`` consumes