fix(rectification): unify discriminator question contract and rank by information gain

Python and TypeScript now share a four-option probe contract, persist Focus before asking, and pick the highest-value renderable probe instead of preferring low-gain career events over D24.

Co-authored-by: Cursor <cursoragent@cursor.com>

scripts/rectification/event_probes.py

@@ -36,6 +36,7 @@ from scripts.rectification.candidate_contrast import (
     opportunity_from_probe,
 )
 from scripts.rectification.case_holdout import holdout_domain_years
+from scripts.rectification.probe_question_contract import complete_style_options
 from scripts.rectification.refinement_packet import match_level
 
 MAX_PROBES = 3
@@ -564,6 +565,9 @@ def _public_probe(
     payload.update(extra)
     if payload["role"] == "distinguish":
         payload["candidate_ids"] = candidate_ids_from_outcomes(payload.get("expected_outcomes") or [])
+    style_options = complete_style_options(payload.get("choice_kind"), payload.get("style_options"))
+    if style_options:
+        payload["style_options"] = style_options
     return payload
 
 
@@ -717,6 +721,7 @@ def _evaluate_contexts(
     version = set_version or candidate_set_version([yes_times, no_times])
     outcomes = [
         {"answer_class": "yes", "supports": yes_times, "conflicts": no_times},
+        {"answer_class": "weak_yes", "supports": yes_times, "conflicts": no_times},
         {"answer_class": "no", "supports": no_times, "conflicts": yes_times},
         {"answer_class": "unsure", "supports": [], "conflicts": []},
     ]
@@ -752,7 +757,7 @@ def _evaluate_contexts(
             "calculationResultId": None,
         }],
     )
-    if distinguish_contract_errors(probe):
+    if not probe or distinguish_contract_errors(probe):
         return None
     return probe
 

scripts/rectification/probe_question_contract.py

@@ -0,0 +1,115 @@
+"""Shared probe → choice-card contract for Python event probes.
+
+Must stay aligned with frontend/src/lib/rectification-agentic/v9/probe-question-contract.ts.
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Any, Sequence
+
+QUESTION_CONTRACT_VERSION = "probe-question-v1"
+ANSWER_CLASSES = ("yes", "weak_yes", "no", "unsure")
+EXISTENCE_STYLE_OPTIONS: tuple[dict[str, str], ...] = (
+    {"label": "明确发生且时间吻合", "answer_class": "yes"},
+    {"label": "发生过但程度较弱", "answer_class": "weak_yes"},
+    {"label": "明确没有发生", "answer_class": "no"},
+    {"label": "这段记不清楚", "answer_class": "unsure"},
+)
+QUALITY_STYLE_OPTIONS: tuple[dict[str, str], ...] = (
+    {"label": "发挥明显失常或压力很大", "answer_class": "yes"},
+    {"label": "有压力但不算明显失常", "answer_class": "weak_yes"},
+    {"label": "发挥正常、没有明显失常", "answer_class": "no"},
+    {"label": "这段记不清楚", "answer_class": "unsure"},
+)
+VARGA_NONE_STYLE_OPTION = {"label": "都不是这些特质", "answer_class": "no"}
+UNSURE_STYLE_OPTION = {"label": "这段记不清楚", "answer_class": "unsure"}
+_FORBIDDEN = ("外貌", "体质", "胎记", "疤痕", "伤疤", "身高", "体型")
+_CLOCK = re.compile(r"(?:[01]?\d|2[0-3]):[0-5]\d")
+
+
+def probe_question_kind(value: Any) -> str:
+    if value in {"varga_style", "event_quality"}:
+        return str(value)
+    return "existence"
+
+
+def clipped_probe_label(value: Any, minimum: int = 4, maximum: int = 80) -> str | None:
+    if not isinstance(value, str):
+        return None
+    text = " ".join(value.split())
+    if len(text) < minimum or len(text) > maximum:
+        return None
+    if any(token in text for token in _FORBIDDEN) or _CLOCK.search(text):
+        return None
+    return text
+
+
+def _incoming_option(row: Any) -> dict[str, str] | None:
+    if not isinstance(row, dict):
+        return None
+    answer = row.get("answer_class") or row.get("answerClass")
+    label = clipped_probe_label(row.get("label"))
+    if not label or answer not in ANSWER_CLASSES:
+        return None
+    payload = {"label": label, "answer_class": str(answer)}
+    sign = row.get("sign")
+    if isinstance(sign, str) and sign.strip():
+        payload["sign"] = sign.strip()
+    return payload
+
+
+def complete_style_options(
+    choice_kind: Any,
+    style_options: Sequence[Any] | None = None,
+) -> list[dict[str, str]] | None:
+    kind = probe_question_kind(choice_kind)
+    incoming = [item for item in (_incoming_option(row) for row in (style_options or [])) if item]
+    by_class: dict[str, dict[str, str]] = {}
+    if kind == "varga_style":
+        for option in incoming:
+            by_class[option["answer_class"]] = option
+        by_class.setdefault("unsure", dict(UNSURE_STYLE_OPTION))
+        scoring = [item for item in ANSWER_CLASSES if item != "unsure" and item in by_class]
+        if len(scoring) < 2:
+            return None
+        by_class.setdefault("no", dict(VARGA_NONE_STYLE_OPTION))
+        if "yes" not in by_class or "weak_yes" not in by_class:
+            return None
+    else:
+        catalog = QUALITY_STYLE_OPTIONS if kind == "event_quality" else EXISTENCE_STYLE_OPTIONS
+        for option in catalog:
+            by_class[option["answer_class"]] = dict(option)
+        for option in incoming:
+            by_class[option["answer_class"]] = option
+    ordered: list[dict[str, str]] = []
+    seen: set[str] = set()
+    for answer_class in ANSWER_CLASSES:
+        option = by_class.get(answer_class)
+        if not option:
+            return None
+        label = option["label"]
+        if label in seen and option.get("sign"):
+            label = f"{label}（{option['sign']}）"
+        if label in seen:
+            label = f"{label}·{answer_class}"
+        seen.add(label)
+        ordered.append({**option, "label": label})
+    labels = {item["label"] for item in ordered}
+    classes = {item["answer_class"] for item in ordered}
+    if len(ordered) != 4 or labels != {item["label"] for item in ordered} or classes != set(ANSWER_CLASSES):
+        return None
+    if len(labels) != 4:
+        return None
+    return ordered
+
+
+def is_renderable_probe(probe: dict[str, Any]) -> bool:
+    gain = probe.get("information_gain")
+    if not isinstance(gain, (int, float)) or gain <= 0:
+        return False
+    candidate_ids = probe.get("candidate_ids") or []
+    outcomes = probe.get("expected_outcomes") or []
+    if len(candidate_ids) < 2 or len(outcomes) < 2:
+        return False
+    return complete_style_options(probe.get("choice_kind"), probe.get("style_options")) is not None