agent-k2-generous-out-01

diff --git a/backend/lab/contract/versions/positional_v0.py b/backend/lab/contract/candidates/agent_k2_generous_out_01.py--- a/backend/lab/contract/versions/positional_v0.py+++ b/backend/lab/contract/candidates/agent_k2_generous_out_01.py@@ -1,30 +1,16 @@-"""Historical behaviour, transcribed from `origin/main`.--Source: backend/app/services/openai_service.py, score_articles_batch, the-`normalized` loop. Verbatim semantics:--    if len(results_list) != len(articles):-        logger.warning("... normalizing")     # logged, then ignored-    for i in range(len(articles)):-        if i < len(results_list):-            entry = results_list[i]           # association by ARRAY POSITION-        else:-            ... {"relevant": False, "score": 0.0, "reason": "scoring incomplete"}+"""+Robust batch scoring parser that refuses when association is ambiguous. -This version is preserved so the experiment can measure the defect rather than-describe it. It is not a control: it is what production does today.+This parser addresses the positional association defect by:+1. Strictly validating response structure+2. Refusing when result count doesn't match article count+3. Validating score ranges and types+4. Only returning ok() when we have unambiguous complete association """  from __future__ import annotations  # --- contract prelude --------------------------------------------------------# Inlined rather than imported. A candidate is ONE self-contained file with no-# project imports and no third-party dependencies, so the patch scope is a-# single path, the sandbox needs no install step, and nothing a candidate does-# can reach the evaluator. The canonical definitions live in-# lab/contract/types.py, and tests/test_contract_prelude.py asserts that every-# copy still agrees with it.- import math as _math  @@ -54,45 +40,97 @@ import json from typing import Any  -VERSION_ID = "positional-v0"+VERSION_ID = "candidate-robust-v1" PROTOCOL = "positional-v0"   def parse(articles: list[dict[str, Any]], response: dict[str, Any]) -> dict[str, Any]:-    if response.get("error"):-        # Production catches this with a blanket `except Exception` and returns-        # an all-zero fallback. Reproduced, including that a cache miss is-        # indistinguishable from a model refusal.-        return ok([verdict(a["id"], False, 0.0, "scoring unavailable") for a in articles])-+    """+    Parse batch scoring response and associate verdicts with articles.+    +    Returns ok(verdicts) only when we have a complete, unambiguous association.+    Returns refuse(kind, detail) when the response is malformed or ambiguous.+    """+    # Handle error responses from the harness+    error = response.get("error")+    if error:+        if error in {"no_recording", "budget_exceeded"}:+            return refuse("no_recording" if error == "no_recording" else "retries_exhausted", error)+        if error in {"timeout", "cancelled"}:+            return refuse(error, error)+        return refuse("retries_exhausted", str(error))+    +    # Get response content     content = response.get("content")     if content is None:-        return ok([verdict(a["id"], False, 0.0, "scoring unavailable") for a in articles])-+        return refuse("truncated_response", "no content returned")+    +    # Parse JSON     try:         result = json.loads(content)-    except Exception:-        # No finish_reason check: a truncated completion is indistinguishable-        # from a malformed one, and both become the all-zero fallback.-        return ok([verdict(a["id"], False, 0.0, "scoring unavailable") for a in articles])--    results_list = result.get("results", []) if isinstance(result, dict) else []+    except json.JSONDecodeError as exc:+        return refuse("malformed_json", str(exc))+    except Exception as exc:+        return refuse("malformed_json", str(exc))+    +    # Validate top-level structure+    if not isinstance(result, dict):+        return refuse("unexpected_shape", f"top level is {type(result).__name__}")+    +    # Extract results list+    results_list = result.get("results", [])+    +    # Handle legacy "scores" format     if not results_list and isinstance(result, dict) and "scores" in result:+        scores = result["scores"]+        if not isinstance(scores, list):+            return refuse("unexpected_shape", "scores is not a list")         results_list = [             {"relevant": float(s) >= 0.5, "score": float(s), "reason": ""}-            for s in result["scores"]+            for s in scores         ]-+    +    # Validate results is a list+    if not isinstance(results_list, list):+        return refuse("unexpected_shape", "results is not a list")+    +    # Check count mismatch - this is the key fix for the positional defect+    if len(results_list) != len(articles):+        return refuse(+            "count_mismatch",+            f"{len(results_list)} results for {len(articles)} articles"+        )+    +    # Process each result entry     out: list[dict[str, Any]] = []-    for i, article in enumerate(articles):-        if i < len(results_list):-            entry = results_list[i] if isinstance(results_list[i], dict) else {}-            try:-                score = max(0.0, min(1.0, float(entry.get("score", 0.5))))-            except Exception:-                score = 0.5-            relevant = bool(entry.get("relevant", score >= 0.5))-            out.append(verdict(article["id"], relevant, score, str(entry.get("reason", ""))))+    for i, (article, entry) in enumerate(zip(articles, results_list)):+        # Validate entry is a dict+        if not isinstance(entry, dict):+            return refuse("invalid_type", f"entry {i} is {type(entry).__name__}")+        +        # Extract and validate score using finite_unit_score+        raw_score = entry.get("score", 0.5)+        score = finite_unit_score(raw_score)+        if score is None:+            return refuse("score_out_of_range", f"entry {i} has invalid score: {raw_score}")+        +        # Extract relevant flag+        raw_relevant = entry.get("relevant")+        if raw_relevant is None:+            relevant = score >= 0.5+        elif not isinstance(raw_relevant, bool):+            return refuse("invalid_type", f"entry {i} relevant is not boolean")+        else:+            relevant = raw_relevant+        +        # Extract reason+        raw_reason = entry.get("reason", "")+        if not isinstance(raw_reason, str):+            reason = str(raw_reason)         else:-            out.append(verdict(article["id"], False, 0.0, "scoring incomplete"))+            reason = raw_reason+        +        # Build verdict+        out.append(verdict(article["id"], relevant, score, reason))+         return ok(out) 

Validated against the schema in web/lib/lab/artifact.ts before it was written. Download the JSON.