agent-k2-generous-out-02

diff --git a/backend/lab/contract/versions/positional_v0.py b/backend/lab/contract/candidates/agent_k2_generous_out_02.py--- a/backend/lab/contract/versions/positional_v0.py+++ b/backend/lab/contract/candidates/agent_k2_generous_out_02.py@@ -1,30 +1,13 @@-"""Historical behaviour, transcribed from `origin/main`.+"""Keyed protocol parser (keyed-v2) that associates verdicts by article_id. -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"}--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 fixes the positional association defect by requiring each result+to explicitly identify its article via article_id, enabling unambiguous+association regardless of result order. """  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 +37,116 @@ import json from typing import Any  -VERSION_ID = "positional-v0"-PROTOCOL = "positional-v0"+VERSION_ID = "keyed-v2"+PROTOCOL = "keyed-v2"   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 response using keyed association by article_id.+    +    Each result entry must contain an 'article_id' field that matches one of the+    input articles. This enables unambiguous association regardless of result order.+    """+    # Handle error responses+    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 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 []-    if not results_list and isinstance(result, dict) and "scores" in result:+    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__}")++    # Get results list+    results_list = result.get("results", [])+    if not isinstance(results_list, list):+        return refuse("unexpected_shape", "results is not a list")++    # Handle legacy "scores" format by converting to keyed format+    if not results_list and "scores" in result:+        scores = result["scores"]+        if not isinstance(scores, list):+            return refuse("unexpected_shape", "scores is not a list")+        if len(scores) != len(articles):+            return refuse("count_mismatch", f"{len(scores)} scores for {len(articles)} articles")+        # Convert legacy scores to keyed results using positional mapping         results_list = [-            {"relevant": float(s) >= 0.5, "score": float(s), "reason": ""}-            for s in result["scores"]+            {"article_id": article["id"], "relevant": float(s) >= 0.5, "score": float(s), "reason": ""}+            for article, s in zip(articles, scores)         ] -    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", ""))))+    # Check count match+    if len(results_list) != len(articles):+        return refuse("count_mismatch", f"{len(results_list)} results for {len(articles)} articles")++    # Build expected ID set from input articles+    expected_ids = {article["id"] for article in articles}+    +    # Process results and validate keyed association+    verdicts = []+    seen_ids = set()+    +    for entry in results_list:+        if not isinstance(entry, dict):+            return refuse("invalid_type", f"entry is {type(entry).__name__}")+        +        # Extract article_id+        article_id = entry.get("article_id")+        if article_id is None:+            return refuse("missing_id", "result entry missing article_id field")+        +        # Validate article_id is a string+        if not isinstance(article_id, str):+            return refuse("invalid_type", "article_id is not a string")+        +        # Check for duplicates+        if article_id in seen_ids:+            return refuse("duplicate_id", f"duplicate article_id: {article_id}")+        seen_ids.add(article_id)+        +        # Check article_id is in expected set+        if article_id not in expected_ids:+            return refuse("unknown_id", f"unknown article_id: {article_id}")+        +        # Extract and validate score+        score_raw = entry.get("score", 0.5)+        score = finite_unit_score(score_raw)+        if score is None:+            return refuse("score_out_of_range", f"invalid score: {score_raw}")+        +        # Extract relevant (default based on score)+        relevant_raw = entry.get("relevant")+        if relevant_raw is None:+            relevant = score >= 0.5+        elif isinstance(relevant_raw, bool):+            relevant = relevant_raw         else:-            out.append(verdict(article["id"], False, 0.0, "scoring incomplete"))-    return ok(out)+            return refuse("invalid_type", f"relevant is not boolean: {type(relevant_raw).__name__}")+        +        # Extract reason+        reason = entry.get("reason", "")+        if not isinstance(reason, str):+            reason = str(reason)+        +        verdicts.append(verdict(article_id, relevant, score, reason))+    +    # Check all expected IDs were present+    missing = expected_ids - seen_ids+    if missing:+        return refuse("missing_id", f"missing results for articles: {sorted(missing)}")+    +    return ok(verdicts) 

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