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_loop.py

@@ -22,6 +22,7 @@ from portfolio_optimiser_claude.loop import (
     generate_candidate,
     parse_checker_verdict,
     run_debate,
+    run_project,
 )
 
 # One affected item: 100 × 1000 = 100_000 NOK total; nominal feasible 30_000;
@@ -208,3 +209,21 @@ class TestParseCheckerVerdict:
 
     def test_empty_text_parses_as_absent(self) -> None:
         assert parse_checker_verdict("").decision == "absent"
+
+
+class TestGenerationPrompt:
+    def test_generation_demands_the_raw_json_object_only(self) -> None:
+        # S10 post-mortem: 3 of 4 generation replies carried plausible JSON
+        # wrapped in a markdown fence + commentary — each one a full-price
+        # blind parse-retry (§3 Step 2). The prompt must forbid the wrapping.
+        client = ScriptedClient(
+            replies=[
+                reply("reasoning"),
+                reply("VERDICT: APPROVE"),
+                reply(json.dumps(VALID_PROPOSAL)),
+            ]
+        )
+        run_project(client, "context", meter=_meter(), max_debate_rounds=1)
+        generation_prompt = client.prompts("proposer")[1]
+        assert "ONLY the raw JSON object" in generation_prompt
+        assert "no markdown fences" in generation_prompt