fix(run): S10 del 2 — post-mortem: stopp-artefakt, SDK-isolasjon, raw-JSON-direktiv

Transkript-analyse av den stoppede live-kjøringen (10 kall, 162 250 tokens,
$0.331506): konfig-lekkasjen (setting_sources=None laster ALLE filsystem-
settings) injiserte operatørens Claude-konfig i hvert kall — ~10-15k uncachede
tokens, en påtvunget bekreftelses-preamble som gjorde ren-JSON-svar umulige,
og en checker kapret av lekkede instrukser (debatt konvergerte aldri).

- persist_stop_artifacts: stopp-event verbatim + usage/kost persisteres ALLTID
  ved BudgetExceeded (delt usage-shape med fullført-run-stien)
- build_call_options: setting_sources=[] (SDK isolation mode, verifisert mot
  installert 0.2.110-kilde), system_prompt=None → tom system-prompt; detach-
  bevis via monkeypatchet query
- _generation_prompt: krever ONLY the raw JSON object (fence-innpakning ga
  4 fullpris parse-retries)

187/187 uten nøkkel · ruff + mypy --strict rene · tre detach-bevis RØDE → grønn

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QdSfQdND84oeq2mbjueLTS

tests/test_s10_run_layer.py

@@ -15,7 +15,12 @@ from pathlib import Path
 
 import pytest
 
-from portfolio_optimiser_claude.artifacts import build_citations, persist_run_artifacts
+from portfolio_optimiser_claude.artifacts import (
+    build_citations,
+    persist_run_artifacts,
+    persist_stop_artifacts,
+)
+from portfolio_optimiser_claude.budget import BudgetExceeded
 from portfolio_optimiser_claude.contracts import (
     ModelMapContract,
     TerminationContract,
@@ -216,3 +221,57 @@ def test_persistence_is_deterministic(tmp_path: Path) -> None:
     first = {name: path.read_bytes() for name, path in _persist(tmp_path / "a").items()}
     second = {name: path.read_bytes() for name, path in _persist(tmp_path / "b").items()}
     assert first == second
+
+
+# --- stop artifacts: usage + cost persisted even when the budget stops the run ---------------
+# The S10 live run stopped structurally on the token cap (§8) and persisted
+# NOTHING — real spend without a record. A budget stop is a run outcome.
+
+
+def _persist_stop(out_dir: Path) -> dict[str, Path]:
+    return persist_stop_artifacts(
+        out_dir,
+        stop=BudgetExceeded("tokens", limit=150_000, observed=162_250),
+        termination=TerminationContract(max_rounds=12, max_tokens=150_000),
+        tokens_used=162_250,
+        rounds_used=5,
+        cost_usd=0.331506,
+    )
+
+
+def test_budget_stop_persists_stop_event_and_usage(tmp_path: Path) -> None:
+    paths = _persist_stop(tmp_path / "out")
+    assert set(paths) == {"stop", "usage"}
+    for path in paths.values():
+        assert path.is_file()
+
+
+def test_stop_artifact_mirrors_the_stop_event_verbatim(tmp_path: Path) -> None:
+    # LOAD-BEARING (§8/§11): the artifact carries the breached kind, limit and
+    # observed value exactly as raised — never recomputed, never summarised.
+    paths = _persist_stop(tmp_path / "out")
+    assert json.loads(paths["stop"].read_text(encoding="utf-8")) == {
+        "kind": "tokens",
+        "limit": 150_000,
+        "observed": 162_250,
+    }
+
+
+def test_stop_usage_artifact_has_the_completed_run_shape(tmp_path: Path) -> None:
+    # S11 reads ONE usage format: the stop path must write the same shape
+    # (usage against caps + cost) as the completed-run path.
+    paths = _persist_stop(tmp_path / "out")
+    usage = json.loads(paths["usage"].read_text(encoding="utf-8"))
+    assert usage == {
+        "cost_usd": 0.331506,
+        "max_rounds": 12,
+        "max_tokens": 150_000,
+        "rounds_used": 5,
+        "tokens_used": 162_250,
+    }
+
+
+def test_stop_persistence_is_deterministic(tmp_path: Path) -> None:
+    first = {name: path.read_bytes() for name, path in _persist_stop(tmp_path / "a").items()}
+    second = {name: path.read_bytes() for name, path in _persist_stop(tmp_path / "b").items()}
+    assert first == second