feat(loop): S8 — D7 agentic loop: budget meter, maker-checker gate, informed refinement

Spec §3 steps 2–5 + §8, TDD-ed offline (scripted, honesty-marked stand-in):
- budget.py: BudgetMeter over TerminationContract — provider-reported usage
  only (missing usage fails closed), structured BudgetExceeded stop event.
- loop.py: ModelClient protocol; blind parse-retry generation (never silent
  repair); round-capped debate with turn safety net and mandated VERDICT
  line; opt-in-reject checker gate (explicit REJECT overrides a validated
  outcome, validator rejection stands); most-recent-reason-verbatim informed
  refinement under max_attempts; validator_decision stamped BEFORE override,
  checker_decision as its own result field (§9, never conflated).
- 45 new tests (121 total, no API key); four detach proofs run RED and
  reverted green: checker override, informed block, surfaced checker output,
  stamp-before-override.

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

tests/_scripted.py

@@ -0,0 +1,53 @@
+"""SCRIPTED model-client stand-in for the offline test suite.
+
+Honesty rule (method-spec §1): this is a scripted stand-in, NOT a model. It
+replays canned replies (or computes them via a prompt-sensitive script
+function) deterministically, records every call, and never touches a network
+or an API key. The ONE genuine model run in the programme is S10.
+"""
+
+from __future__ import annotations
+
+from typing import Callable
+
+from portfolio_optimiser_claude.loop import ModelReply
+
+ScriptFn = Callable[[str, str], ModelReply]
+
+
+class ScriptedClient:
+    """Deterministic ``ModelClient`` stand-in — canned replies, recorded calls.
+
+    Either ``replies`` (a FIFO of ``ModelReply``) or ``script`` (a function of
+    ``(role, prompt)`` — prompt-sensitive, for load-bearing flip proofs) must
+    be given. Every call is recorded as ``(role, prompt)`` in ``calls``.
+    """
+
+    def __init__(
+        self,
+        replies: list[ModelReply] | None = None,
+        script: ScriptFn | None = None,
+    ) -> None:
+        if (replies is None) == (script is None):
+            raise ValueError("give exactly one of 'replies' or 'script'")
+        self._replies = list(replies) if replies is not None else None
+        self._script = script
+        self.calls: list[tuple[str, str]] = []
+
+    def complete(self, prompt: str, *, role: str) -> ModelReply:
+        self.calls.append((role, prompt))
+        if self._script is not None:
+            return self._script(role, prompt)
+        assert self._replies is not None
+        if not self._replies:
+            raise AssertionError("scripted client exhausted: no reply left for this call")
+        return self._replies.pop(0)
+
+    def prompts(self, role: str) -> list[str]:
+        """The recorded prompts sent to ``role``, in call order."""
+        return [prompt for r, prompt in self.calls if r == role]
+
+
+def reply(text: str, usage_tokens: int | None = 10) -> ModelReply:
+    """Shorthand for a scripted ``ModelReply`` with a small default usage."""
+    return ModelReply(text=text, usage_tokens=usage_tokens)