fix(v1-gate): rows 1-2 stop at FORM OK — what files cannot show, a person has to say

The PM checkpoint of 18.09 paid M-6's price three times and row 2 still read 3 of 3 GREEN:
a whole artefact family written by hand and made internally consistent (verdict_id minted
with the product's own rule, about sixty lines of script), four REAL runs' artefacts under
a handwritten feedback file, and `<n>/outbox` symlinked out of the round to a real run
elsewhere. `RUN_ATTESTATION` stood in the output the whole time, and the row was green
anyway — which is the finding: a row that PRINTS its limit and reports GREEN has not
stated a limit, it has decorated one.

One of the three is a hole and is closed; two are not holes at all.

1. The outbox must lie in the round ITSELF (`outbox_escape`). Derived is only derived if
   the filesystem cannot redirect it: `<n>/outbox` as a symlink, or any symlinked artefact
   inside it, is refused by name.

2. Rows 1 and 2 can no longer be green from files. What the computation gives is
   `FORM_OK` ("FORM OK, IKKE BEVIST"): every check the gate knows how to make passed, the
   round counts 0 against the criterion, and the exit code still fails — it is un-green in
   every way RED is. What separates the two is what the reader must do next, not what the
   gate concludes.

3. Green is `<n>/attestering.txt`, per round: the operator's own statement, naming the
   round, the run and the date. Plain text with three keyed lines, deliberately not JSON —
   it is the one artefact in this contract a PERSON types, and a missing brace must never
   be able to read as "not confirmed". Nothing in the product writes one, and a test in
   `src/` says so: a product that can produce a witness to its own run has produced
   exactly the thing these rows cannot.

   A MISSING attestation is FORM OK: nobody has confirmed anything yet, and the gate is
   not entitled to call that a lie. A PRESENT one that names another round or run, carries
   no date, an unparsable date, or a date before the run, is RED: it is a statement about
   something this is not. Row 2 needs round 0 attested as well, since round 1 is measured
   against it — an unattested baseline is an unattested comparison.

The attestation is read raw off `outcome.json` (`_declared_run`), not through
`read_outcome`, so row 1 — which reads no run — does not inherit row 2's artefact checks
through the back door.

Nothing moved on the criterion: 0/3 · 0/3 · 3/8 · no report · 3/8 · IKKE MÅLT · 1/20,
exit 1. What changed is that rows 1-2 now name what they are waiting for, and it is not
another file.

Suite after `git add` on a clean tree: `uv run pytest -q` = 2014 passed, 5 skipped,
5 xfailed (508 s). The attack of b769537/ab00016 is the red half of this commit.

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

src/portfolio_optimiser/evals/v1_gate.py

@@ -27,7 +27,7 @@ import tempfile
 import xml.etree.ElementTree as ET
 from collections.abc import Callable, Iterable, Mapping, Sequence
 from dataclasses import asdict, dataclass, field
-from datetime import datetime
+from datetime import date, datetime
 import difflib
 import re
 from pathlib import Path
@@ -58,8 +58,26 @@ RUN_ATTESTATION = (
     "operatøren"
 )
 
+#: Rows 1-2: the file the gate can neither derive nor write, and the one thing that turns their
+#: computation into GREEN. Plain text with three keyed lines, deliberately not JSON: it is the one
+#: artefact in the contract a PERSON types by hand, and a missing brace must never be able to read
+#: as "not confirmed".
+ATTEST_FILE = "attestering.txt"
+#: What an attestation has to name: which round, which run, and when it was given.
+ATTEST_KEYS = ("runde", "kjøring", "dato")
+#: Printed on rows 1-2 on every run: what GREEN on those rows means, and whose statement it is.
+ATTEST_RULE = (
+    "grønt på rad 1-2 regner gaten ikke ut: filene kan stemme innbyrdes og likevel beskrive en "
+    "runde ingen har holdt. Målingen gir FORM OK; skrittet derfra til grønt er operatørens egen "
+    f"attestering i <runde>/{ATTEST_FILE} (runde, kjøring, dato), som gaten ALDRI skriver selv"
+)
+
 GREEN = "GRØNN"
 RED = "RØD"
+#: Rows 1-2 only: every computed check passed, and the round is still waiting for the operator's
+#: attestation. Never green, so it fails the exit code exactly as RED does — what separates it
+#: from RED is what the reader must do next, not what the gate concludes.
+FORM_OK = "FORM OK, IKKE BEVIST"
 DIAGNOSIS = "DIAGNOSE"
 #: A row whose evidence could not be read. Never green: on a failing row it fails the exit code.
 NOT_MEASURED = "IKKE MÅLT"
@@ -75,6 +93,7 @@ Rundekatalogen (--rounds-dir) har fast form. Runde n = tilbakemelding på rappor
   <rounds-dir>/<n>/outbox/           kjøring n sin EGEN utboks, kopiert hit urørt
   <rounds-dir>/<n>/report.md         rapporten bygget fra kjøring n
   <rounds-dir>/3/report.kept.md      runde 3-rapporten slik fagpersonen BEHOLDT den
+  <rounds-dir>/<n>/attestering.txt   operatørens bekreftelse på runden       (n = 0, 1, 2, 3)
 
 feedback.json:
   {"author": "<fagpersonen>", "given_at": "<ISO-8601 med tidssone>",
@@ -101,6 +120,17 @@ kjøring må stå inne for seg selv: hver EVALUERT tilnærming har <run_id>-<id>
 hvis verdict_id gaten selv minter PÅ NYTT fra forslagets egen IR (verdicts.verdict_key). En
 not_evaluated-rad har ingen av delene, og en artefakt for en tilnærming coverage ikke nevner hører
 til en annen kjøring. Dette beviser ikke at kjøringen skjedde — se attesteringen for rad 2.
+attestering.txt (ren tekst, skrevet for hånd av operatøren — gaten skriver den ALDRI):
+  runde: <n>
+  kjøring: <run_id slik rundens outcome.json oppgir den>
+  dato: <ISO-dato, ikke før kjøringen ble gjort>
+  <fritekst under, om ønskelig>
+
+Rad 1 og 2 kan ALDRI bli grønne av filer alene. Uten attesteringen er taket FORM OK, IKKE BEVIST:
+runden teller 0 mot ferdig-kriteriet og gaten feller fortsatt. En attestering som FINNES, men
+navngir en annen runde eller kjøring, mangler dato eller er datert før kjøringen, er RØD - den er
+et utsagn om noe annet. Rad 2 krever i tillegg at grunnkjøringen (runde 0) er attestert: det er
+den runde 1 måles mot. Utboksen må dessuten ligge i runden SELV - en lenke ut av runden avvises.
 Rad 4 teller innholdslinjer (ikke blanke, skillelinjer eller tabellrammer) som står uendret og i
 samme rekkefølge; fagpersonens tillegg vises som eget tall. --rounds-dir inne i repoet må være
 gitignored.
@@ -181,6 +211,111 @@ def _parse_time(value: Any) -> datetime | None:
     return stamp if stamp.tzinfo is not None else None
 
 
+def _parse_date(value: str) -> date | None:
+    try:
+        return datetime.fromisoformat(value).date()
+    except ValueError:
+        return None
+
+
+@dataclass(frozen=True)
+class Attestation:
+    """One round's attestation, read back.
+
+    ``present`` separates the two ways it can fail, because they ask different things of the
+    reader: no file at all means nobody has confirmed the round YET (the row waits at
+    ``FORM_OK``), while a file that contradicts the round is a statement about some other round
+    (``RED``). Both are un-green, and neither is a measurement."""
+
+    present: bool
+    ok: bool
+    why: str
+
+
+def _declared_run(round_dir: Path) -> tuple[str, datetime | None]:
+    """The run id and run time the round's own ``outcome.json`` DECLARES — read raw, without the
+    verification ``read_outcome`` does, so row 1 (which reads no run) does not inherit row 2's
+    artefact checks through the attestation."""
+    data, _ = _read_json(round_dir / "outcome.json")
+    if not isinstance(data, Mapping):
+        return "", None
+    return str(data.get("run_id", "")).strip(), _parse_time(data.get("ran_at"))
+
+
+def read_attestation(round_dir: Path) -> Attestation:
+    """The operator's confirmation for ONE round: that it is there, and that it is about THIS
+    round and the run this round declares.
+
+    This is the gate's one input it cannot compute, and that is the point. Measured 18.09: a
+    whole artefact family written by hand and made internally consistent — coverage, artefact
+    pairs, ``verdict_id`` minted with the product's own rule — took row 2 to 3 of 3 GREEN, as did
+    four real runs' artefacts under a handwritten feedback file. Neither is a hole ``verify_run``
+    can close: files agreeing with each other is not a witness that anything happened. So the
+    computation stops at ``FORM_OK`` and green comes from a person saying so, per round."""
+    path = round_dir / ATTEST_FILE
+    if not path.is_file():
+        return Attestation(
+            False,
+            False,
+            f"{ATTEST_FILE} mangler — operatøren har ikke bekreftet at runden ble holdt",
+        )
+    try:
+        text = path.read_text(encoding="utf-8")
+    except (OSError, ValueError) as exc:
+        return Attestation(True, False, f"{ATTEST_FILE} uleselig ({exc!r})")
+    fields: dict[str, str] = {}
+    for line in text.splitlines():
+        key, sep, value = line.partition(":")
+        if sep and key.strip().casefold() in ATTEST_KEYS:
+            fields.setdefault(key.strip().casefold(), value.strip())
+    missing = [k for k in ATTEST_KEYS if not fields.get(k)]
+    if missing:
+        return Attestation(True, False, f"{ATTEST_FILE} mangler {', '.join(missing)}")
+    if fields["runde"] != round_dir.name:
+        return Attestation(
+            True,
+            False,
+            f"{ATTEST_FILE} attesterer runde {fields['runde']!r}, ikke {round_dir.name!r}",
+        )
+    run_id, ran_at = _declared_run(round_dir)
+    if not run_id:
+        return Attestation(
+            True,
+            False,
+            f"{ATTEST_FILE} navngir kjøring {fields['kjøring']!r}, men runden oppgir ingen",
+        )
+    if fields["kjøring"] != run_id:
+        return Attestation(
+            True,
+            False,
+            f"{ATTEST_FILE} navngir kjøring {fields['kjøring']!r}, runden er kjøring {run_id!r}",
+        )
+    given = _parse_date(fields["dato"])
+    if given is None:
+        return Attestation(
+            True, False, f"{ATTEST_FILE}: dato {fields['dato']!r} er ikke en ISO-dato"
+        )
+    if ran_at is not None and given < ran_at.date():
+        return Attestation(
+            True,
+            False,
+            f"{ATTEST_FILE}: attestert {given.isoformat()}, før kjøringen {run_id} ble gjort "
+            f"({ran_at.date().isoformat()})",
+        )
+    return Attestation(True, True, "")
+
+
+def _cap(k: int, required: int, exceptions: Sequence[str], pending: Sequence[str]) -> str:
+    """Rows 1-2 stop where the files stop. Green is not something these rows can COMPUTE: a
+    tree can pass every check the gate knows how to make and still describe a round nobody held.
+    So a round that holds in form but has no attestation leaves the row at ``FORM_OK`` — un-green,
+    still failing the exit code, and counted as 0 against the criterion — while something the
+    gate CAN refuse (bad form, or an attestation about another round) is ``RED``."""
+    if k >= required:
+        return GREEN
+    return RED if exceptions else FORM_OK
+
+
 @dataclass(frozen=True)
 class Feedback:
     ids: frozenset[str]
@@ -280,20 +415,27 @@ def read_feedback(round_dir: Path, ai: tuple[str, str] | None) -> tuple[set[str]
 
 def score_rounds(rounds_dir: Path, required: int, ai: tuple[str, str] | None) -> Row:
     exceptions: list[str] = []
+    pending: list[str] = []
     k = 0
     if not rounds_dir.is_dir():
         exceptions.append(f"{rounds_dir} finnes ikke")
     else:
-        for _, (feedback, why) in sorted(read_rounds(rounds_dir, required, ai).items()):
+        for n, (feedback, why) in sorted(read_rounds(rounds_dir, required, ai).items()):
-            if feedback is not None:
+            if feedback is None:
-                k += 1
-            else:
                 exceptions.append(why)
-    status = GREEN if k == required else RED
+                continue
+            attested = read_attestation(rounds_dir / str(n))
+            if attested.ok:
+                k += 1
+            elif attested.present:
+                exceptions.append(f"runde {n}: {attested.why}")
+            else:
+                pending.append(f"runde {n}: {attested.why}")
+    status = _cap(k, required, exceptions, pending)
     reason = (
-        "form verifisert i alle runder (forfatterskap: se attestering)"
+        "form verifisert i alle runder, og hver runde attestert av operatøren"
-        if k == required
+        if status == GREEN
-        else exceptions[0]
+        else (exceptions + pending)[0]
     )
     return Row(
         "rounds",
@@ -302,7 +444,8 @@ def score_rounds(rounds_dir: Path, required: int, ai: tuple[str, str] | None) ->
         required,
         status,
         reason,
-        exceptions=tuple(exceptions),
+        exceptions=tuple(exceptions + pending),
+        attests=(ATTEST_RULE,),
     )
 
 
@@ -440,6 +583,31 @@ def verify_run(outbox: Path, run_id: str, coverage: Sequence[Mapping[str, Any]])
     return ""
 
 
+def outbox_escape(round_dir: Path) -> str:
+    """``""`` when the run the round presents lies IN the round, else why not.
+
+    The outbox is DERIVED (``RUN_OUTBOX``) rather than declared — but a derived path is only
+    derived if the filesystem cannot redirect it. Measured 18.09: ``<n>/outbox`` made a symlink
+    to a real run's directory elsewhere let a round present another run's artefacts as its own,
+    and row 2 read 3 of 3 GREEN. Every artefact the round stands on must therefore be a real file
+    under the round itself, symlink-free in both directions."""
+    outbox = round_dir / RUN_OUTBOX
+    if outbox.is_symlink():
+        return (
+            f"{RUN_OUTBOX}/ er en lenke ut av runden ({os.readlink(outbox)}) — kjøringen må "
+            f"ligge i {round_dir.name}/{RUN_OUTBOX}/ selv"
+        )
+    if not outbox.is_dir():
+        return ""
+    for entry in sorted(outbox.rglob("*")):
+        if entry.is_symlink():
+            return (
+                f"{RUN_OUTBOX}/{entry.relative_to(outbox)} er en lenke "
+                f"({os.readlink(entry)}) — kjøringens artefakter må ligge i runden selv"
+            )
+    return ""
+
+
 def read_outcome(round_dir: Path) -> tuple[Outcome | None, str]:
     """A round's outcome file, VERIFIED against the run it names — and the run is the one in the
     round's OWN ``outbox/``, never a path the file points at.
@@ -473,6 +641,9 @@ def read_outcome(round_dir: Path) -> tuple[Outcome | None, str]:
         )
     if ran_at is None:
         return None, f"{path}: ran_at er ikke et ISO-tidsstempel med tidssone"
+    escape = outbox_escape(round_dir)
+    if escape:
+        return None, f"{path}: {escape}"
     outbox = round_dir / RUN_OUTBOX
     coverage_path = outbox / f"{run_id}-coverage.json"
     payload, why = _read_json(coverage_path)
@@ -548,11 +719,18 @@ def outcomes_changed(prev: Outcome, cur: Outcome, feedback_ids: set[str]) -> tup
 
 def score_changes(rounds_dir: Path, required: int, ai: tuple[str, str] | None) -> Row:
     exceptions: list[str] = []
+    pending: list[str] = []
     k = 0
     base_path = rounds_dir / "0" / "outcome.json"
     base_outcome, base_why = read_outcome(rounds_dir / "0")
     base = f"runde 0 = {base_path}"
     base += f" (kjøring {base_outcome.run_id})" if base_outcome else f" — {base_why}"
+    # Round 1 is measured AGAINST round 0's run, so an unattested baseline is an unattested
+    # comparison and no round can count. Said ONCE here rather than repeated on every round.
+    base_attested = read_attestation(rounds_dir / "0")
+    if not base_attested.ok:
+        target = exceptions if base_attested.present else pending
+        target.append(f"runde 0 (grunnkjøringen rad 2 måler mot): {base_attested.why}")
     feedback_by_round = read_rounds(rounds_dir, required, ai) if rounds_dir.is_dir() else {}
     for n in range(1, required + 1):
         feedback, why = feedback_by_round.get(n, (None, f"runde {n}: {rounds_dir} finnes ikke"))
@@ -570,20 +748,24 @@ def score_changes(rounds_dir: Path, required: int, ai: tuple[str, str] | None) -
             )
             continue
         ok, detail = outcomes_changed(prev, cur, set(feedback.ids))
-        if ok:
+        if not ok:
-            k += 1
-        else:
             exceptions.append(f"runde {n}: {detail}")
-    status = GREEN if k == required else RED
+            continue
+        attested = read_attestation(rounds_dir / str(n))
+        if not attested.ok:
+            (exceptions if attested.present else pending).append(f"runde {n}: {attested.why}")
+            continue
+        if base_attested.ok:
+            k += 1
     return Row(
         "changes",
         "2 runder med målbar endring",
         k,
         required,
-        status,
+        _cap(k, required, exceptions, pending),
         base,
-        exceptions=tuple(exceptions),
+        exceptions=tuple(exceptions + pending),
-        attests=(RUN_ATTESTATION,),
+        attests=(RUN_ATTESTATION, ATTEST_RULE),
     )
 
 

tests/test_v1_gate.py

@@ -521,12 +521,13 @@ def _forged_rounds(root: Path) -> Path:
 
 
 def test_m6_a_handwritten_outbox_is_not_a_run(tmp_path: Path) -> None:
-    """Row 2 must not be satisfiable by files a forger wrote. Row 1 stays GREEN on the same tree —
+    """Row 2 must not be satisfiable by files a forger wrote. Row 1 reads FORM OK on the same
-    the feedback there IS well formed, and that is what makes this an attack on row 2 rather than
+    tree — the feedback there IS well formed, and that is what makes this an attack on row 2
-    a broken fixture. The control that the row can still go green is ``_green_rounds``, which
+    rather than a broken fixture; since M-7 that form is as far as row 1 can get without the
-    carries a whole run family (``test_row2_a_traced_change_counts``)."""
+    operator's attestation. The control that the row can still go green is ``_green_rounds``,
+    which carries a whole run family AND an attestation (``test_row2_a_traced_change_counts``)."""
     root = _forged_rounds(tmp_path)
-    assert gate.score_rounds(root, 3, _AI).status == gate.GREEN
+    assert gate.score_rounds(root, 3, _AI).status == gate.FORM_OK
     row = gate.score_changes(root, 3, _AI)
     assert (row.k, row.status) == (0, gate.RED), row.exceptions