Datasets:

iNLP-Lab
/

UnCoBench

from __future__ import annotations
import json
import re
from typing import Any
COMMUNICATION_ROLE_NAMES = []
GOLD = {'conclusion': 'Select', 'numeric_result': 500}
POSITIVE_CONCLUSION = 'Select'
NEGATIVE_CONCLUSION = 'Do not select'
NUMERIC_PRECISION = 0
PUBLIC_ROLES = [{'role_name': 'Structural safety engineer (ENG)', 'responsibility': 'Evaluates facility integrity, seismic rating compliance, and storm-surge clearance elevation', 'role_abbreviation': 'ENG'}, {'role_name': 'Emergency utilities coordinator (PWR)', 'responsibility': 'Monitors electrical auxiliary systems, on-site backup generator availability, and auxiliary fuel autonomy', 'role_abbreviation': 'PWR'}, {'role_name': 'Evacuation transit director (LOG)', 'responsibility': 'Assesses route clearance, convoy transit intake duration, and bus arrival bottlenecks', 'role_abbreviation': 'LOG'}, {'role_name': 'Public health officer (HLT)', 'responsibility': 'Verifies paramedic staffing accreditation and on-site clinical triage equipment', 'role_abbreviation': 'HLT'}, {'role_name': 'Shelter accommodation manager (OPS)', 'responsibility': 'Maintains certified cot bedding inventory and facility intake registrations', 'role_abbreviation': 'OPS'}, {'role_name': 'Municipal emergency commissioner (ADM)', 'responsibility': 'Enforces mandatory government quotas and emergency municipal reservations', 'role_abbreviation': 'ADM'}, {'role_name': 'Civic volunteer coordinator (VOL)', 'responsibility': 'Tracks community volunteer sign-up rates and non-governmental aid networks', 'role_abbreviation': 'VOL'}]
ROLE_KEY_PREFIXES = {'Structural safety engineer (ENG)': 'ENG-', 'Emergency utilities coordinator (PWR)': 'PWR-', 'Evacuation transit director (LOG)': 'LOG-', 'Public health officer (HLT)': 'HLT-', 'Shelter accommodation manager (OPS)': 'OPS-', 'Municipal emergency commissioner (ADM)': 'ADM-', 'Civic volunteer coordinator (VOL)': 'VOL-'}
ROLE_VALID_KEYS = {'Structural safety engineer (ENG)': ['ENG-SURGE-VV5', 'ENG-SURGE-3EN', 'ENG-SURGE-9KP', 'ENG-FLOOR-RFV', 'ENG-FLOOR-G3O', 'ENG-FLOOR-QJS'], 'Emergency utilities coordinator (PWR)': ['PWR-UTIL-8DP', 'PWR-UTIL-0KL', 'PWR-UTIL-U7X', 'PWR-WATER-5IK', 'PWR-WATER-Y2D', 'PWR-WATER-Z0V'], 'Evacuation transit director (LOG)': ['LOG-DOCK-5M7', 'LOG-DOCK-ZV3', 'LOG-DOCK-O3V', 'LOG-TRANS-VZ8', 'LOG-TRANS-EHK', 'LOG-TRANS-MM2'], 'Public health officer (HLT)': ['HLT-ISOL-3U2', 'HLT-ISOL-MDR', 'HLT-ISOL-0FQ', 'HLT-MED-BBS', 'HLT-MED-SS0', 'HLT-MED-K7Q'], 'Shelter accommodation manager (OPS)': ['OPS-MEAL-PH6', 'OPS-MEAL-CXT', 'OPS-MEAL-FHI', 'OPS-BED-EH5', 'OPS-BED-999', 'OPS-BED-0U5'], 'Municipal emergency commissioner (ADM)': ['ADM-STAFF-OJL', 'ADM-STAFF-E9C', 'ADM-STAFF-MSL', 'ADM-QUOTA-J62', 'ADM-QUOTA-TBV', 'ADM-QUOTA-SHP'], 'Civic volunteer coordinator (VOL)': ['VOL-DONAT-8YO', 'VOL-DONAT-48O', 'VOL-DONAT-ILQ', 'VOL-AID-TCF', 'VOL-AID-6M9', 'VOL-AID-1CC']}
RECORD_REPLY_KEYS = {'ENG-SURGE-3EN': ['ENG-SURGE-3EN'], 'ENG-SURGE-9KP': ['ENG-SURGE-9KP'], 'ENG-SURGE-VV5': ['ENG-SURGE-VV5'], 'PWR-UTIL-8DP': ['PWR-UTIL-8DP'], 'PWR-UTIL-0KL': ['PWR-UTIL-0KL'], 'PWR-UTIL-U7X': ['PWR-UTIL-U7X'], 'LOG-TRANS-VZ8': ['LOG-TRANS-VZ8'], 'LOG-TRANS-MM2': ['LOG-TRANS-MM2'], 'LOG-TRANS-EHK': ['LOG-TRANS-EHK'], 'HLT-MED-K7Q': ['HLT-MED-K7Q'], 'HLT-MED-SS0': ['HLT-MED-SS0'], 'HLT-MED-BBS': ['HLT-MED-BBS'], 'OPS-BED-999': ['OPS-BED-999'], 'OPS-BED-EH5': ['OPS-BED-EH5'], 'OPS-BED-0U5': ['OPS-BED-0U5'], 'ADM-QUOTA-J62': ['ADM-QUOTA-J62'], 'ADM-QUOTA-TBV': ['ADM-QUOTA-TBV'], 'ADM-QUOTA-SHP': ['ADM-QUOTA-SHP'], 'VOL-AID-TCF': ['VOL-AID-TCF'], 'VOL-AID-6M9': ['VOL-AID-6M9'], 'VOL-AID-1CC': ['VOL-AID-1CC'], 'ENG-FLOOR-G3O': ['ENG-FLOOR-G3O'], 'ENG-FLOOR-QJS': ['ENG-FLOOR-QJS'], 'ENG-FLOOR-RFV': ['ENG-FLOOR-RFV'], 'PWR-WATER-Y2D': ['PWR-WATER-Y2D'], 'PWR-WATER-5IK': ['PWR-WATER-5IK'], 'PWR-WATER-Z0V': ['PWR-WATER-Z0V'], 'LOG-DOCK-ZV3': ['LOG-DOCK-ZV3'], 'LOG-DOCK-O3V': ['LOG-DOCK-O3V'], 'LOG-DOCK-5M7': ['LOG-DOCK-5M7'], 'HLT-ISOL-0FQ': ['HLT-ISOL-0FQ'], 'HLT-ISOL-3U2': ['HLT-ISOL-3U2'], 'HLT-ISOL-MDR': ['HLT-ISOL-MDR'], 'ADM-STAFF-MSL': ['ADM-STAFF-MSL'], 'ADM-STAFF-E9C': ['ADM-STAFF-E9C'], 'ADM-STAFF-OJL': ['ADM-STAFF-OJL'], 'VOL-DONAT-ILQ': ['VOL-DONAT-ILQ'], 'VOL-DONAT-48O': ['VOL-DONAT-48O'], 'VOL-DONAT-8YO': ['VOL-DONAT-8YO'], 'OPS-MEAL-PH6': ['OPS-MEAL-PH6'], 'OPS-MEAL-FHI': ['OPS-MEAL-FHI'], 'OPS-MEAL-CXT': ['OPS-MEAL-CXT']}
REQUIRED_KEYS = ['ADM-QUOTA-J62', 'ENG-SURGE-3EN', 'ENG-SURGE-9KP', 'HLT-MED-K7Q', 'LOG-TRANS-EHK', 'LOG-TRANS-VZ8', 'OPS-MEAL-PH6', 'PWR-UTIL-8DP']
DISTRACTOR_KEYS = ['VOL-AID-1CC', 'VOL-AID-6M9', 'VOL-AID-TCF', 'VOL-DONAT-48O', 'VOL-DONAT-8YO', 'VOL-DONAT-ILQ']
SETTING = 'misaligned'

def runtime_defaults() -> dict[str, Any]:
    return {
        "setting": SETTING,
        "communication_role_names": list(COMMUNICATION_ROLE_NAMES),
        "finalize_ideal_round": 4,
        "max_turns": 7,
    }

FINALIZE_IDEAL_ROUND = runtime_defaults()["finalize_ideal_round"]
DEFAULT_MAX_TURNS = runtime_defaults()["max_turns"]
ERROR_INJECTION = {'enabled': True, 'role_name': 'Shelter accommodation manager (OPS)', 'target_key': 'OPS-BED-999', 'alternate_key': 'OPS-MEAL-PH6', 'target_context': 'Production environment', 'alternate_context': 'Simulation staging environment'}
INCOMPLETE_INJECTION = {}
INCOMPLETE_FOLLOWUP_RECORDS = []
KEY_PATTERN = re.compile(r"^[A-Z]+-[A-Z]+-[A-Z0-9]{3}$")
REPLY_KEY_PATTERN = re.compile(r"(?<![A-Z0-9])[A-Z]+-[A-Z]+-[A-Z0-9]{3}(?![A-Z0-9])", re.IGNORECASE)
ROLE_REPLY_MAX_ATTEMPTS = 3
TASK_QUERY = 'Determine whether exactly one candidate evacuation shelter meets all of the following criteria: storm-surge elevation clearance holds, structural seismic rating compliance holds, active backup generator readiness is confirmed with a backup generator continuous fuel autonomy of at least 48 h, maximum transit intake time is at most 45 min, and certified paramedic staffing is confirmed. If exactly one candidate satisfies all criteria, conclude "Select" and calculate its net available shelter capacity (in spaces) by subtracting the municipal quota reservation from the total certified bed spaces verified in the Production environment; otherwise, conclude "Do not select" and report null for the numeric result.\n\nNumeric calculation rule: Compute only for the unique eligible candidate: [Total certified bed spaces] - [Municipal quota reservation]. Result unit: spaces. Round only the final result to 0 decimal places (ties to even). If no candidate qualifies, numeric_result is null.\n\nCandidates and records to verify:\n- Bayside Civic Pavilion：ENG-SURGE-9KP、PWR-UTIL-0KL、LOG-TRANS-MM2、HLT-MED-SS0、OPS-BED-EH5、ADM-QUOTA-TBV、VOL-AID-6M9\n- Highland Sports Arena：ENG-SURGE-3EN、PWR-UTIL-8DP、LOG-TRANS-VZ8、HLT-MED-K7Q、OPS-MEAL-PH6、ADM-QUOTA-J62、VOL-AID-TCF\n- Valley Regional Gymnasium：ENG-SURGE-VV5、PWR-UTIL-U7X、LOG-TRANS-EHK、HLT-MED-BBS、OPS-BED-0U5、ADM-QUOTA-SHP、VOL-AID-1CC'
TASK_CONTEXT = 'Disaster management authorities are reviewing candidate coastal evacuation shelters to designate an emergency intake facility during severe weather events. To guarantee evacuee safety and logistical feasibility, each prospective shelter must be verified against civil protection criteria covering flood surge clearance, structural resilience, emergency power reliability, maximum transit time, and specialized medical readiness.'

"""Parse one model message without treating trailing generation noise as an action."""
import json
import re
from typing import Any


def parse_json_message(content: Any, *, trace: dict[str, Any] | None = None) -> dict[str, Any]:
    if trace is not None:
        trace.clear()
        trace.update(policy="first_complete_object", raw_response=content, recovered=False)
    if not isinstance(content, str):
        raise ValueError("model response must be text")
    text = content.strip().lstrip("\ufeff").strip()
    # Parse the first complete object; never recover a nested fragment.
    try:
        value, end = json.JSONDecoder().raw_decode(text)
        start = 0
    except json.JSONDecodeError:
        if text.startswith(("{", "[", '"')):
            raise
        opening = re.search(r"[\{\[]", text)
        if opening is None:
            raise
        start = opening.start()
        value, consumed = json.JSONDecoder().raw_decode(text[start:])
        end = start + consumed
    if not isinstance(value, dict):
        raise ValueError("model response must be a JSON object")
    prefix, suffix = text[:start], text[end:]
    if trace is not None:
        trace.update(recovered=bool(prefix.strip() or suffix.strip()),
                     ignored_prefix=prefix, ignored_suffix=suffix)
    return value


def expand_question_keys(questions):
    expanded = []
    for question in questions:
        if not isinstance(question, dict):
            expanded.append(question)
            continue
        text = question.get("question")
        keys = list(dict.fromkeys(REPLY_KEY_PATTERN.findall(text))) if isinstance(text, str) else []
        for key in keys or [None]:
            expanded.append(dict(question, key=key))
    return expanded


def validate_question(question, *, role_name, semantic_enabled=False):
    if not isinstance(question, dict) or not isinstance(question.get("question"), str) or not question["question"].strip():
        raise ValueError("question must contain natural-language text")
    if role_name not in ROLE_KEY_PREFIXES:
        raise ValueError("unknown Role")
    key = question.get("key")
    authorized = semantic_enabled if key is None else (
        isinstance(key, str) and key in ROLE_VALID_KEYS[role_name] and key in question["question"])
    return {"authorized": bool(authorized)}

def recovery_key_match(answer: str) -> str | None:
    target = ERROR_INJECTION.get("target_key")
    if not target:
        return None
    def contains(key: str) -> bool:
        return bool(re.search(r"(?<![A-Z0-9-])" + re.escape(key) + r"(?![A-Z0-9-])", answer, re.IGNORECASE))
    if contains(target):
        return "full_key"
    parts = target.split("-")
    alias = parts[0] + "-" + parts[-1]
    matches = [key for key in ROLE_VALID_KEYS.get(ERROR_INJECTION.get("role_name"), [])
               if key.split("-")[0] + "-" + key.split("-")[-1] == alias]
    if matches == [target] and contains(alias):
        return "abbreviated_key"
    return None

def _key_communication_metrics(transcript: list[dict[str, Any]]) -> dict[str, Any]:
    key_query_rounds: dict[str, int] = {}
    key_query_counts: dict[str, int] = {}
    queried_roles: set[str] = set()
    mismatch_round: int | None = None
    recovery_query_sent = False
    recovery_query_turn: int | None = None
    target_key_returned = False
    recovery_turn: int | None = None
    recovery_method: str | None = None
    recovery_reference: str | None = None
    target_key_query_turn: int | None = None
    finalized_before_recovery = False
    for event in transcript:
        turn = int(event.get("turn", len(transcript)))
        action = event.get("main_agent_action", {})
        if (
            ERROR_INJECTION.get("enabled")
            and action.get("action") == "finalize"
            and not target_key_returned
        ):
            finalized_before_recovery = True
        if action.get("action") != "ask_role_agents": continue
        for answer in event.get("role_answers", []):
            role_name = answer.get("role_agent")
            if role_name: queried_roles.add(role_name)
            if (
                ERROR_INJECTION.get("enabled") and mismatch_round is not None and turn > mismatch_round
                and role_name == ERROR_INJECTION.get("role_name")
                and answer.get("question", {}).get("key") == ERROR_INJECTION.get("target_key")
                and "role_trace" in answer
                and recovery_key_match(str(answer.get("answer", ""))) is not None
            ):
                target_key_query_turn = target_key_query_turn or turn
            if (
                ERROR_INJECTION.get("enabled")
                and role_name == ERROR_INJECTION.get("role_name")
                and answer.get("question", {}).get("key") == ERROR_INJECTION.get("alternate_key")
                and "role_trace" in answer
            ):
                mismatch_round = mismatch_round or turn
        for message in action.get("messages", []):
            for question in message.get("questions", []):
                key = question.get("key")
                if (
                    ERROR_INJECTION.get("enabled")
                    and resolve_role_name(message.get("role_agent")) == ERROR_INJECTION.get("role_name")
                    and (key is None or key == ERROR_INJECTION.get("target_key"))
                    and mismatch_round is not None
                    and turn > mismatch_round
                ):
                    recovery_query_sent = True
                    recovery_query_turn = recovery_query_turn or turn
                if not isinstance(key, str) or not KEY_PATTERN.fullmatch(key): continue
                key_query_rounds.setdefault(key, turn)
                key_query_counts[key] = key_query_counts.get(key, 0) + 1
        if recovery_query_sent and not target_key_returned and mismatch_round is not None and turn > mismatch_round:
            for answer in event.get("role_answers", []):
                question = answer.get("question", {})
                if (
                    answer.get("role_agent") == ERROR_INJECTION.get("role_name")
                    and (question.get("key") is None or question.get("key") == ERROR_INJECTION.get("target_key"))
                    and "role_trace" in answer
                    and recovery_key_match(str(answer.get("answer", ""))) is not None
                ):
                    target_key_returned = True
                    recovery_reference = recovery_key_match(str(answer.get("answer", "")))
                    recovery_turn = turn
                    recovery_method = "keyless_query" if question.get("key") is None else "target_key_query"
                    break
    queried = set(key_query_counts)
    required = set(REQUIRED_KEYS)
    relevant = queried & required
    return {
        "required_key_recall":len(relevant)/len(required) if required else 1.0,
        "required_key_precision":len(relevant)/len(queried) if queried else 0.0,
        "key_query_rounds":key_query_rounds,
        "key_query_counts":key_query_counts,
        "duplicate_key_queries":sum(max(0, count-1) for count in key_query_counts.values()),
        "distractor_key_queries":sum(key_query_counts.get(key, 0) for key in DISTRACTOR_KEYS),
        "distractor_key_query_rate":sum(key_query_counts.get(key, 0) for key in DISTRACTOR_KEYS)/sum(key_query_counts.values()) if key_query_counts else 0.0,
        "turns_to_finalize":len(transcript),
        "mismatch_key_queried":mismatch_round is not None,
        "mismatch_key_turn":mismatch_round,
        "recovery_query_sent":recovery_query_sent,
        "recovery_query_turn":recovery_query_turn,
        "target_key_returned":target_key_returned,
        "recovery_turn":recovery_turn,
        "recovery_method":recovery_method,
        "target_key_query_turn":target_key_query_turn,
        "recovery_key_match":recovery_reference,
        "recovery_metrics_version":"keyed_and_keyless_v3",
        "recovery_latency":recovery_turn - mismatch_round if recovery_turn is not None and mismatch_round is not None else None,
        "finalized_before_recovery":finalized_before_recovery,
    }

def main_system_prompt() -> str:
    role_text = "\n".join(f"- {item['role_name']}: {item['responsibility']}" for item in PUBLIC_ROLES)
    ask_example = json.dumps({"action":"ask_role_agents","messages":[{"role_agent":"Role name","questions":[{"id":"Q1","question":"Please provide the record for ABC-DATA-A1B using the supplied key."}]}]}, ensure_ascii=False)
    final_example = json.dumps({"action":"finalize","conclusion":"Exactly: " + POSITIVE_CONCLUSION + " or " + NEGATIVE_CONCLUSION,"numeric_result":None}, ensure_ascii=False).replace('"numeric_result": null', '"numeric_result": null/number')
    communication_protocol = ('You may output {"action":"wait"} to advance one logical turn. An empty observation {"observations":[]} means no new visible messages; it does not imply receipt, processing or any future reply.' if SETTING in {'missed', 'mixed'} else '')
    followup = ('For keyless follow-ups, name the candidate in full and the metric explicitly, or request that candidate\'s complete records. A unique previously queried record can be checked directly; otherwise supply a full key and metric when asked to clarify. Explicit directory requests return the complete directory.' if SETTING == 'incomplete' else
                'Authorized keyless follow-ups may name the candidate and metric or request complete records. A role may retrieve facts, request clarification or return a directory. If given a directory, select a valid key to obtain facts.' if SETTING == 'mixed' else
                'Authorized keyless questions return that role\'s complete directory. Choose a key from the directory to obtain concrete facts.')
    return "\n\n".join([
        'You are the Main Agent. Integrate information from independent role agents and make the final decision. Communicate in English.',
        'You cannot see private role information directly. Obtain it only by explicitly messaging the role agents.',
        'Key protocol:\n1. Candidate names and record keys are supplied in the task. Initial access requires a valid key; authorized later follow-ups are the exception.\n'
        '2. Keys use ROLE-BUSINESS-XXX: role abbreviation, business abbreviation and three random alphanumeric characters. Use complete supplied keys; do not invent them.\n'
        '3. Send each key to its owning business role. Use valid keys supplied by roles for subsequent checks.\n'
        '4. Do not output a separate key field. Put the complete key in question text; the system extracts it and checks ownership. Separate different keys into individual questions, preferably one record per question.\n'
        '5. After a successful keyed query to a role, keyless natural-language follow-ups to that role are authorized starting from the next turn.\n6. ' + followup,
        'Available independent role agents:\n\n' + role_text,
        'To ask questions, output only JSON in this form:\n\n' + ask_example,
        *([communication_protocol] if communication_protocol else []),
        'Final answer format (output only valid JSON):\n\n' + final_example,
        'Use exactly one of these conclusion labels: ' + POSITIVE_CONCLUSION + ' or ' + NEGATIVE_CONCLUSION + '. '
        'Do not append a candidate name, explanation or other text. null/number means choose one, not literal syntax. '
        'For a positive conclusion provide the computed number; for a negative conclusion use null. Do not include units or explanations in numeric_result.',
    ])


ERROR_RECOVERY_PROFILE = {'role_name': 'Shelter accommodation manager (OPS)', 'target_key': 'OPS-BED-999', 'alternate_key': 'OPS-MEAL-PH6', 'target_context': 'Production environment', 'alternate_context': 'Simulation staging environment', 'candidate': 'Highland Sports Arena', 'other_candidates': ['Bayside Civic Pavilion', 'Valley Regional Gymnasium'], 'numeric': [{'topic': 'total_bed_spaces', 'value': 600, 'label': 'Total certified bed spaces', 'aliases': ['Certified bed count', 'Shelter nominal capacity'], 'unit': 'spaces'}], 'boolean': [], 'language': 'en', 'facts': [{'topic': 'surge_clearance_met', 'label': 'Storm-surge elevation clearance', 'aliases': ['Surge elevation clearance', 'Flood level clearance']}, {'topic': 'seismic_rating_compliant', 'label': 'Structural seismic rating compliance', 'aliases': ['Seismic code compliance', 'Earthquake rating verification']}, {'topic': 'backup_generator_active', 'label': 'Active backup generator readiness', 'aliases': ['Auxiliary generator status', 'Emergency generator presence']}, {'topic': 'generator_fuel_autonomy', 'label': 'Backup generator continuous fuel autonomy', 'aliases': ['Generator fuel endurance', 'Backup run duration']}, {'topic': 'transit_intake_minutes', 'label': 'Maximum transit intake time', 'aliases': ['Convoy transit intake duration', 'Evacuation transit time']}, {'topic': 'paramedic_staffing_certified', 'label': 'Certified paramedic staffing', 'aliases': ['Paramedic on-duty certification', 'Triage medical readiness']}, {'topic': 'total_bed_spaces', 'label': 'Total certified bed spaces', 'aliases': ['Certified bed count', 'Shelter nominal capacity']}, {'topic': 'municipal_quota_reservation', 'label': 'Municipal quota reservation', 'aliases': ['Reserved bed quota', 'Government reserved spaces']}]}
"""Conservative English evidence rules shared by standalone generated runtimes.
This is attribution of explicit statements, not arbitrary semantic entailment.
Unsupported or conflicting statements remain unknown; no model calls are made.
"""
import re
from decimal import Decimal

_RECOVERY_NUMBER = r'[+-]?(?:\d+(?:\.\d+)?|\.\d+)(?:[eE][+-]?\d+)?'
_RECOVERY_UNCERTAIN = r"\b(?:unknown|uncertain|unclear|may(?!\s+\d{1,4}\b)|might|could|should|would|must|expected|possibly|whether|cannot|unable|unconfirmed|approximately|about|around|roughly|estimated|not confirmed|not known|not established)\b"
_RECOVERY_NEGATIVE = r'\b(?:not|no|never|false|invalid|blocked|failed|denied|unavailable|unapproved|revoked|inactive|disabled|noncompliant)\b'


def evidence_text(text):
    text = text.replace('’', "'").replace('−', '-').replace('‑', '-')
    text = re.sub(r"\bcan't\b", 'cannot', text, flags=re.I)
    text = re.sub(r"n't\b", ' not', text, flags=re.I)
    text = re.sub(r'(?<![\w.,])[-+]?\d{1,3}(?:,\d{3})+(?:\.\d+)?(?![\d,]|\.\d)',
                  lambda m: m[0].replace(',', ''), text)
    return text


def evidence_phrase(text, phrase):
    if not phrase:
        return False
    return bool(re.search(r'(?<!\w)' + re.escape(phrase).replace(r'\ ', r'\s+') + r'(?!\w)', text, re.I))


def evidence_nonasserted(text):
    """Questions, hypotheses and unconfirmed claims are not returned facts."""
    return bool('?' in text or re.search(_RECOVERY_UNCERTAIN, text, re.I)
                or re.search(r'\b(?:if|unless|assuming|suppose|supposing|hypothetically)\b', text, re.I)
                or re.search(r'\b(?:do|does|did|has|have|had)\s+not\s+(?:confirm|verify|establish)(?:ed)?\b', text, re.I))


def evidence_clauses(text):
    # Preserve questions and uncertainty when splitting assertions.
    sentences = re.split(r'[。\n]|\.(?!\d)|(?<=[!?])\s+', text)
    clauses = []
    for sentence in sentences:
        if evidence_nonasserted(sentence):
            clauses.append(sentence)
        else:
            # Split conjunctions between assertions, never inside metric labels.
            predicate = r'\b(?:is|are|was|were|has|have|had|holds?|received|maintains?|equals?)\b|[:=]'
            parts = re.split(r'([;；]|,?\s+\b(?:but|and)\b\s+(?![+-]?\d))', sentence, flags=re.I)
            current = parts[0]
            for separator, following in zip(parts[1::2], parts[2::2]):
                continuation = re.match(r'(?:it|its|this|that|(?:the\s+)?status)\b', following, re.I)
                split = not continuation and (not re.search(r'\band\b', separator, re.I) or
                        re.search(predicate, current, re.I) and re.search(predicate, following, re.I))
                if split:
                    clauses.append(current)
                    current = following
                else:
                    current += separator + following
            clauses.append(current)
    return clauses


def evidence_scope(answer, question, profile):
    for target in [profile.get('candidate', ''), profile.get('target_context', '')]:
        if target and re.search(r'\b(?:not|outside|excluding|except)(?:\s+(?:for|the))?\s+' + re.escape(target) + r'(?!\w)', answer, re.I):
            return 'not_returned'
    candidates = profile.get('other_candidates', [])
    contexts = profile.get('other_contexts', [profile.get('alternate_context', '')])
    if any(evidence_phrase(answer, c) for c in [*candidates, *contexts] if c):
        target_also_present = evidence_phrase(answer, profile.get('candidate', '')) and (
            not profile.get('target_context') or evidence_phrase(answer, profile['target_context']))
        return 'unknown' if target_also_present else 'not_returned'
    combined = question.get('question', '') + '\n' + answer
    target_key = profile.get('target_key', '')
    if target_key and (question.get('key') == target_key or evidence_phrase(combined, target_key)):
        return 'eligible'
    if evidence_phrase(combined, profile.get('candidate', '')) and (
            not profile.get('target_context') or evidence_phrase(combined, profile['target_context'])):
        return 'eligible'
    return 'unknown'


def evidence_record_text(answer, profile):
    """Isolate an explicitly keyed record in a multi-record reply.
    A directory can contain facts as well as keys. Only the target key's
    segment is eligible; neighboring records never supply its evidence.
    Unstructured multi-candidate prose remains conservatively unscored.
    """
    keys = list(re.finditer(r'(?<!\w)[A-Z]{2,10}-[A-Z]{2,10}-[A-Z0-9]{3}(?!\w)', answer))
    target = profile.get('target_key')
    if not target or not any(m[0] == target for m in keys):
        return answer
    pieces = []
    for i, match in enumerate(keys):
        if match[0] == target:
            end = keys[i+1].start() if i+1 < len(keys) else len(answer)
            pieces.append(answer[match.start():end])
    return '\n'.join(pieces)


def evidence_key_request_only(answer):
    """An explicit request for a key supplies no factual recovery evidence."""
    return bool(re.fullmatch(r'\s*(?:please\s+(?:provide|supply)|I\s+need)\s+(?:a\s+)?valid\s+key\b[^\n.;]*[.!]?\s*', answer, re.I))


def evidence_unit_pattern(unit):
    # Unit names may inflect; canonical symbols remain case-sensitive.
    # Compose rate units without changing symbol case.
    parts = re.split(r'\s*/\s*|\s+per\s+', unit)
    if len(parts) == 2 and all(parts):
        return evidence_unit_pattern(parts[0]) + r'\s*(?:/|(?i:per))\s*' + evidence_unit_pattern(parts[1])
    groups = [('h', 'hour', 'hours'), ('min', 'minute', 'minutes'), ('s', 'second', 'seconds'),
              ('kg', 'kilogram', 'kilograms'), ('m', 'meter', 'meters', 'metre', 'metres'),
              ('d', 'day', 'days'), ('L', 'liter', 'liters', 'litre', 'litres'),
              ('%', 'percent', 'per cent'), ('instances', 'instance'), ('items', 'item'),
              ('people', 'person', 'persons'), ('days', 'day')]
    for group in groups:
        if unit in group:
            canonical, *words = group
            if canonical in {'instances', 'items', 'people', 'days'}:
                words.append(canonical)
                canonical = None
            parts = ([re.escape(canonical)] if canonical else []) + ['(?i:' + re.escape(w) + ')' for w in words]
            if unit == '%': parts.append('％')
            return '(?:' + '|'.join(sorted(parts, key=len, reverse=True)) + ')'
    if re.fullmatch(r'[a-z]+(?:-[a-z]+)+', unit):
        prefix, suffix = unit.rsplit('-', 1)
        return re.escape(prefix + '-') + evidence_unit_pattern(suffix)
    if re.fullmatch(r'[a-z]{3,}', unit) and unit not in {'mol', 'rad', 'kat', 'lux'}:
        singular = unit
        if unit.endswith('ies'):
            singular = unit[:-3] + 'y'
        elif re.search(r'(?:ch|sh|ss|x|z)es$', unit):
            singular = unit[:-2]
        elif unit.endswith('s') and not unit.endswith(('ss', 'us', 'is')):
            singular = unit[:-1]
        plural = (singular[:-1] + 'ies' if re.search(r'[^aeiou]y$', singular) else
                  singular + 'es' if re.search(r'(?:s|x|z|ch|sh)$', singular) else singular + 's')
        return '(?i:' + '|'.join(re.escape(w) for w in sorted({unit, singular, plural}, key=len, reverse=True)) + ')'
    return re.escape(unit)


def evidence_measurements(text, unit):
    text = evidence_text(text)
    text = re.sub(r'(?<=\d)[-–—](?=\d)', ' to ', text)
    suffix = (r'(?:\s+(?:1|(?i:dimensionless)))?' if unit == '1'
              else r'\s*' + evidence_unit_pattern(unit))
    pattern = r'(?<![A-Za-z0-9_.,+\-])(' + _RECOVERY_NUMBER + r')' + suffix + r'(?![A-Za-z0-9_/%]|\.\d|,\d)'
    return text, list(re.finditer(pattern, text))


def evidence_exact_measurement(text, match):
    before, after = text[max(0, match.start()-60):match.start()], text[match.end():match.end()+45]
    before = re.sub(r'\bamounts?\s+to\s*$', 'equals ', before, flags=re.I)
    if evidence_nonasserted(text):
        return False
    if re.search(r'(?:\b(?:under|over|below|above|less than|more than|at most|at least|up to|between|to|and|not|no more than|no less than)|不超过|不低于|小于等于|大于等于|小于|大于|至少|至多|最多|最少|低于|高于|介于|至|到|[<>≤≥～~–—])\s*$', before, re.I):
        return False
    if re.search(r'\d\s*[-–—]\s*$', before):
        return False
    if re.match(r'\s*(?:\b(?:maximum|minimum|max|min)\b|\bor (?:less|more|lower|higher)\b|\b(?:to|through)\s+[+-]?\d|[-–—~～]\s*\d)', after, re.I):
        return False
    return True


def english_exact_value(answer, value, unit):
    for clause in evidence_clauses(evidence_text(answer)):
        text, matches = evidence_measurements(clause, unit)
        if any(Decimal(m[1]) == Decimal(str(value)) and evidence_exact_measurement(text, m) for m in matches):
            return True
    return False


def numeric_evidence(answer, question, profile, fact):
    if evidence_key_request_only(answer):
        return 'not_returned'
    answer = evidence_text(evidence_record_text(answer, profile))
    scope = evidence_scope(answer, question, profile)
    if scope != 'eligible':
        return scope
    labels = [fact.get('label', ''), *fact.get('aliases', [])]
    states = set()
    for clause in metric_assertions(answer, profile, fact):
        # Bind each value to its asserted metric; nearby numbers are not evidence.
        if not any(evidence_phrase(clause, label) for label in labels if label):
            continue
        if evidence_nonasserted(clause) or re.search(r'\b(?:catalog|directory)\b', clause, re.I):
            continue
        # Leave ambiguous multi-metric clauses unscored.
        others = profile.get('facts', profile.get('numeric', []) + profile.get('boolean', []))
        if any(other.get('topic') != fact.get('topic') and any(evidence_phrase(clause, label) for label in
               [other.get('label', ''), *other.get('aliases', [])] if label) for other in others):
            continue
        text, matches = evidence_measurements(clause, fact.get('unit', ''))
        # Bind counts to the supplied metric and unit, never to gold values alone.
        unit = fact.get('unit', '')
        if re.fullmatch(r'[a-z]{3,}', unit):
            for label in labels:
                if re.search(r'\b' + evidence_unit_pattern(unit) + r'$', label):
                    pattern = r'(?<![\w.,+-])(' + _RECOVERY_NUMBER + r')\s+' + re.escape(label).replace(r'\ ', r'\s+') + r'(?!\w)'
                    matches += list(re.finditer(pattern, text, re.I))
        for m in matches:
            if evidence_exact_measurement(text, m) and not re.search(_RECOVERY_NEGATIVE, text[:m.start()], re.I):
                states.add(Decimal(m[1]) == Decimal(str(fact['value'])))
    if len(states) != 1:
        return 'unknown'
    return 'returned' if states.pop() else 'not_returned'


def _evidence_predicate(text):
    text = re.sub(r'\b(?:is|are|was|were|has|have|had|been|the|a|an|already)\b', ' ', text.casefold())
    return re.sub(r'\s+', ' ', text).strip(' .:')


_BOOLEAN_PREDICATE = r'\b(?:enabled|issued|approved|authorized|isolated|certified|insured|assigned|cleared|present|valid|clear|ready|open|active|completed|available|installed|registered|signed|granted|confirmed|verified|operational|compliant|meets|passed|released|equipped|connected|running|accepted)\b'

# Extract assertions from supplied labels and aliases, never from expected values.
_POSSESSION = r'(?:has|have|had|holds?|received|obtained|secured|possesses|provides|provided|maintains|maintained|achieved|satisfies|demonstrates|demonstrated|confirms?|confirmed|passed|equipped\s+with)'
_ACQUISITION = r'(?:obtained|received|secured|achieved|maintained|held|provided)'


def protected_metrics(text, profile, fact):
    """Mask longest metric spans before any punctuation/conjunction splitting.
    A full label owns overlapping aliases. Preserve punctuation inside labels
    (including Oxford commas) and never infer a metric from its gold value.
    """
    facts = [fact, *profile.get('facts', profile.get('numeric', []) + profile.get('boolean', []))]
    labels = set()
    for item in facts:
        for label in [item.get('label', ''), *item.get('aliases', [])]:
            if label:
                labels.add(label)
                labels.update(_boolean_labels(label))
    if not labels:
        return text, {}
    pattern = r'(?<!\w)(?:' + '|'.join(re.escape(s).replace(r'\ ', r'\s+')
                                      for s in sorted(labels, key=len, reverse=True)) + r')(?!\w)'
    spans = {}
    def mask(match):
        token = f'ZZMETRIC{len(spans)}ZZ'
        spans[token] = match[0]
        return token
    return re.sub(pattern, mask, text, flags=re.I), spans


def restore_metrics(text, spans):
    return re.sub(r'ZZMETRIC\d+ZZ', lambda m: spans.get(m[0], m[0]), text)


def metric_assertions(answer, profile, fact):
    """Bind metrics before splitting prose; leave ambiguous joint lists unknown.
    Split between anchored metrics only after a completed assertion, using the
    last connector before the next metric. This supports values before labels,
    natural verbs and possessive subjects without splitting noun phrases.
    Sentence-level uncertainty remains attached to all its claims.
    """
    masked, spans = protected_metrics(answer, profile, fact)
    for sentence in re.split(r'[。\n]|\.(?!\d)|(?<=[!?])\s+', masked):
        if evidence_nonasserted(sentence):
            yield restore_metrics(sentence, spans)
            continue
        anchors = list(re.finditer(r'ZZMETRIC\d+ZZ', sentence))
        cuts = []
        for left, right in zip(anchors, anchors[1:]):
            gap = sentence[left.end():right.start()]
            separators = list(re.finditer(r'[;；]|,\s*|,?\s+\b(?:and|but|while|whereas)\b\s+', gap, re.I))
            if not separators:
                continue
            sep = separators[-1]
            prior = sentence[cuts[-1] if cuts else 0:left.end()+sep.start()]
            assertion = re.search(r'\b(?:is|are|was|were|has|have|had|measured|equals?|reached|achieves)\b|[:=]|(?<![A-Z])\d',
                                  re.sub(r'ZZMETRIC\d+ZZ', '', prior), re.I)
            if assertion or re.search(r'\b' + _POSSESSION + r'\b', prior, re.I):
                cuts.append(left.end()+sep.end())
        start = 0
        for end in [*cuts, len(sentence)]:
            for clause in evidence_clauses(sentence[start:end]):
                yield restore_metrics(clause, spans)
            start = end


def _boolean_labels(label):
    """Allow a participle before or after its noun phrase, preserving its words."""
    phrase = _evidence_predicate(label)
    variants = [phrase]
    words = phrase.split()
    if len(words) > 1 and re.fullmatch(_BOOLEAN_PREDICATE, words[-1]):
        variants.append(' '.join([words[-1], *words[:-1]]))
        if words[-1] == 'valid':
            variants.append(' '.join(['valid', 'unrestricted', *words[:-1]]))
    return variants


def _boolean_clauses(answer, profile):
    for clause in metric_assertions(answer, profile, profile):
        if evidence_nonasserted(clause):
            yield clause
            continue
        # Keep status continuations attached so their negation is not discarded.
        masked, spans = protected_metrics(clause, profile, profile)
        for part in re.split(
            r',(?!\s*(?:but|and|is|are|was|were|has|have|had|which|that|it|its|status|not|no|false|true|yes|confirmed)\b)\s*',
            masked, flags=re.I):
            yield restore_metrics(part, spans)


def boolean_evidence(answer, question, profile):
    if evidence_key_request_only(answer):
        return 'not_returned'
    answer = evidence_text(evidence_record_text(answer, profile))
    scope = evidence_scope(answer, question, profile)
    if scope != 'eligible':
        return scope
    # Normalize asserted approval while preserving negation and uncertainty.
    for adjective, noun in [('approved', 'approval'), ('authorized', 'authorization')]:
        label = profile['label']
        if label.casefold().endswith(' ' + adjective):
            subject = label[:-(len(adjective)+1)]
            pattern = r'\b(?:obtained|received|secured)\s+(?:formal\s+)?' + noun + r'\s+for\s+(?:(?:its|the)\s+)?' + re.escape(subject)
            answer = re.sub(pattern, subject + ' is ' + adjective, answer, flags=re.I)
    labels = [variant for label in [profile['label'], *profile.get('aliases', [])]
              for variant in _boolean_labels(label)]
    primary = _boolean_labels(profile['label'])
    states = set()
    for clause in _boolean_clauses(answer, profile):
        plain = _evidence_predicate(clause)
        no_negative = _evidence_predicate(re.sub(_RECOVERY_NEGATIVE, ' ', clause, flags=re.I))
        assertion = re.sub(r'\b(?:the|a|an|is|are|was|were|been|already)\b', ' ',
                           re.sub(_RECOVERY_NEGATIVE, ' ', clause, flags=re.I), flags=re.I)
        assertion = re.sub(r'\s+', ' ', assertion).strip()
        if not any(evidence_phrase(plain, _evidence_predicate(label)) or evidence_phrase(no_negative, _evidence_predicate(label)) for label in labels if label):
            continue
        if evidence_nonasserted(clause) or re.search(r'\b(?:catalog|directory)\b', clause, re.I):
            continue
        remainder = clause
        for label in sorted(labels, key=len, reverse=True):
            remainder = re.sub(re.escape(label), '', remainder, flags=re.I)
        negative = bool(re.search(_RECOVERY_NEGATIVE, remainder, re.I))
        explicit = bool(re.search(r'\b(?:true|yes|false|no)\b', remainder, re.I))
        label_match = any(evidence_phrase(no_negative, label) for label in primary)
        affirmative = label_match and bool(re.search('(?:' + _BOOLEAN_PREDICATE + r')\s*$', profile['label'], re.I))
        # Directory entries and requests are not assertions.
        for label in primary:
            escaped = re.escape(label).replace(r'\ ', r'\s+')
            affirmative |= bool(re.search(r'\b' + _POSSESSION + r'\s+(?:that\s+)?(?:(?:valid|active|formal|operational|continuous|dedicated)\s+)*' + escaped + r'(?!\w)', assertion, re.I))
            affirmative |= bool(re.search(r'(?<!\w)' + escaped + r'\s+(?:' + _BOOLEAN_PREDICATE + '|' + _ACQUISITION + r')(?!\w)', no_negative, re.I))
            affirmative |= bool(re.search(r'\b(?:valid|approved|confirmed|verified)\s+' + escaped + r'(?!\w)', no_negative, re.I))
        # Do not transfer negation between unrelated metrics.
        opposites = {'disabled':'enabled', 'unavailable':'available', 'invalid':'valid',
                     'closed':'open', 'inactive':'active', 'blocked':'clear', 'revoked':'granted'}
        opposite = False
        for denied, affirmed in opposites.items():
            if re.search(r'\b' + denied + r'\b', clause, re.I):
                rewritten = re.sub(r'\b' + denied + r'\b', affirmed, clause, flags=re.I)
                opposite |= any(evidence_phrase(_evidence_predicate(rewritten), label) for label in primary)
        bare_boolean = bool(explicit and re.fullmatch(r'(?:status\s*[:=]?\s*)?(?:true|yes|false|no)',
                                                       _evidence_predicate(remainder), re.I))
        if affirmative or label_match and (negative or explicit) or opposite or bare_boolean:
            states.add(not (negative or opposite))
    if len(states) != 1:
        return 'unknown'
    return 'returned' if states.pop() == profile['value'] else 'not_returned'


def recovery_number_present(answer, value):
    """Literal numeric occurrence, independent of unit, metric, scope or wording."""
    text = re.sub(r'(?<![A-Z0-9])[A-Z]{2,10}-[A-Z]{2,10}-[A-Z0-9]{3}(?![A-Z0-9])', '', answer, flags=re.I)
    text = evidence_text(text)
    text = re.sub(r'(?<=\d)[-–—](?=\d)', ' to ', text)
    pattern = r'(?<![A-Za-z0-9_.,+\-])(' + _RECOVERY_NUMBER + r')(?![0-9_]|\.\d|,\d)'
    return any(Decimal(m[1]) == Decimal(str(value)) for m in re.finditer(pattern, text))

"""Pure Error recovery measurement; embedded in generated Error runtimes."""
import re
from decimal import Decimal


ERROR_RECOVERY_VERSION = 'task_labeled_facts_v8_scoped_assertions'


def build_error_recovery_profile(graph):
    pair = graph['error_pair']
    records = {r['record_id']: r for r in graph['records']}
    obs = {o['observation_id']: o for o in graph['observations']}
    target = [obs[i] for i in records[pair['target_record_id']]['observation_ids']]
    alternate = {obs[i]['topic_id']: obs[i]['claimed_value']
                 for i in records[pair['alternate_record_id']]['observation_ids']}
    changed = [o for o in target if o['claimed_value'] != alternate[o['topic_id']]]
    topics = {t['topic_id']: t for t in graph['topics']}
    def fact(o):
        t = topics[o['topic_id']]
        return dict(topic=o['topic_id'], value=o['claimed_value'], label=t['label'],
                    aliases=t.get('aliases', []), unit=t.get('unit', ''))
    numeric = [fact(o) for o in changed
               if type(o['claimed_value']) in (int, float)]
    # Recovery requires every perturbed fact, including Boolean facts.
    boolean = [fact(o) for o in changed if type(o['claimed_value']) is bool]
    profile = dict(role_name=next(r['name'] for r in graph['roles'] if r['role_id']==pair['affected_role_id']),
                target_key=pair['target_key'], alternate_key=pair['alternate_key'],
                target_context=pair['target_context'], alternate_context=pair['alternate_context'],
                candidate=next(c['name'] for c in graph['candidates'] if c['candidate_id']==pair['affected_candidate_id']),
                other_candidates=[c['name'] for c in graph['candidates'] if c['candidate_id']!=pair['affected_candidate_id']],
                numeric=numeric, boolean=boolean)
    if graph.get('language') == 'en':
        profile['language'] = 'en'
        profile['facts'] = [dict(topic=t['topic_id'],label=t['label'],aliases=t.get('aliases', [])) for t in graph['topics']]
    return profile


def validate_error_recovery_profile(profile):
    """Generation gate: every new English metric must accept canonical truth
    and reject an explicit different value. This is compatibility validation,
    not a claim that every natural-language paraphrase can be interpreted.
    """
    if profile.get('language') != 'en':
        return
    facts = profile['numeric'] + profile['boolean']
    if not facts:
        raise ValueError('Error recovery profile has no perturbed facts')
    question = {'key': profile['target_key']}
    for fact in facts:
        is_bool = type(fact['value']) is bool
        for correct in (True, False):
            value = fact['value'] if correct else (not fact['value'] if is_bool else Decimal(str(fact['value'])) + 1)
            answer = f"{fact['label']} = {str(value).lower()} {fact.get('unit', '')}"
            state = (boolean_evidence(answer, question, dict(profile, **fact)) if is_bool else
                     numeric_evidence(answer, question, profile, fact))
            if state != ('returned' if correct else 'not_returned'):
                raise ValueError(f"Error recovery cannot score canonical metric {fact['topic']!r}: {state}")


def error_numeric_present(answer, value):
    text = re.sub(r'[A-Z]{2,10}-[A-Z]{2,10}-[A-Z0-9]{3}', '', answer, flags=re.I)
    # Treat thousands separators as part of a number, not as value boundaries.
    text = re.sub(r'(?<![\w.,])[-+]?\d{1,3}(?:,\d{3})+(?:\.\d+)?(?![\d,]|\.\d)',
                  lambda m: m[0].replace(',', ''), text)
    return Decimal(str(value)) in [Decimal(x) for x in re.findall(
        r'(?<![A-Za-z0-9.,+-])[+-]?\d+(?:\.\d+)?(?![A-Za-z0-9,]|\.\d)', text)]


def _legacy_boolean_reply(event, profile):
    """Tri-state extraction for certification validity, with scope attribution.
    A clear statement that status cannot be determined means no fact returned.
    Unknown is reserved for text the rules cannot confidently interpret.
    """
    if not event['eligible']:
        return 'not_returned'
    answer = event['answer']
    question = event['question']
    combined = question.get('question', '') + '\n' + answer
    keyed = question.get('key') == profile['target_key']
    if profile['alternate_context'] in answer or any(c in answer for c in profile['other_candidates']):
        return 'not_returned'
    identity = keyed or profile['target_key'] in combined or profile['candidate'] in combined
    scope = keyed or profile['target_context'] in combined
    if profile['target_context'] == '评估日时点':
        scope = scope or bool(re.search(r'评估日(?:当天|时点)?', combined))
    metric = keyed or bool(re.search(r'冷链(?:运输)?(?:认证|资质)', combined))
    if not (identity and scope and metric):
        return 'unknown'
    values = []
    explicit_unavailable = False
    for clause in re.split(r'[；;。\n]', answer):
        # Only validity assertions count; audit-log dates do not.
        if not re.search(r'有效|无效|过期|状态为[是否]', clause):
            continue
        if re.search(r'无法|不能.*(?:确定|判定)|未提供|未给出|未.*确定|不明确|可能|推测|并非每天|至少有一天|是否', clause):
            explicit_unavailable = True
            continue
        negative = bool(re.search(r'无效|已过期|不在有效期|未处于有效期|不处于有效期|状态为否', clause))
        positive = bool(re.search(r'处于有效期内|(?<!不)在有效期内|状态为是|(?:认证|资质)(?:为|是)?有效', clause))
        if negative:
            values.append(False)
        elif positive:
            values.append(True)
    if len(set(values)) > 1:
        return 'unknown'
    if values:
        return 'returned' if values[0] == profile['value'] else 'not_returned'
    return 'not_returned' if explicit_unavailable else 'unknown'



def _plain(text):
    text = text.casefold().replace("n't", ' not')
    text = re.sub(r'\b(?:is|are|was|were|has|have|had|been|the|a|an|already)\b', '', text)
    return re.sub(r'[\s\W_]+', '', text)


def _recovery_mentions(text, phrase):
    if not phrase:
        return False
    if re.search(r'[\u3400-\u9fff]', phrase):
        return phrase in text
    return bool(re.search(r'(?<!\w)' + re.escape(phrase) + r'(?!\w)', text, re.I))


def boolean_reply(event, profile):
    """Conservative task-label matching. Unsupported paraphrases remain unknown.
    No decision depends on the spelling of an internal topic ID. Scope/identity
    are established by a target key, or explicitly in a keyless question/reply.
    Chinese support is retained for existing benchmark transcripts.
    """
    if profile.get('language') == 'en':
        return boolean_evidence(event['answer'], event['question'], profile) if event['eligible'] else 'not_returned'
    if not profile.get('label'):
        return _legacy_boolean_reply(event, profile)
    if not event['eligible']:
        return 'not_returned'
    answer = re.sub(r"\bcan['’]t\b", 'cannot', event['answer'], flags=re.I)
    answer = re.sub(r"n['’]t\b", ' not', answer, flags=re.I)
    question = event['question']
    combined = question.get('question', '') + '\n' + answer
    if (profile.get('alternate_context') and _recovery_mentions(answer, profile['alternate_context'])
            or any(_recovery_mentions(answer, c) for c in profile.get('other_candidates', []))):
        return 'not_returned'
    keyed = question.get('key') == profile['target_key']
    referenced = profile['target_key'] in combined
    if not (keyed or referenced or (_recovery_mentions(combined, profile.get('candidate', ''))
                                   and _recovery_mentions(combined, profile.get('target_context', '')))):
        return 'unknown'
    label = profile['label']
    aliases = [label, *profile.get('aliases', [])]
    positive = r'通过|获批|批准|生效|有效|畅通|就绪|下达|启用|开启|完成|达标|达到|具备|已布设|已接入|已签|已验收|无重大|已备案|已投运|是|true|approved|passed|valid|clear|ready|released|enabled|active|completed|available|installed|registered|signed|granted|confirmed|operational|compliant|meets|yes'
    uncertain = r'无法|不能确定|未提供|未给出|不明确|可能|推测|是否|\b(?:unknown|uncertain|unclear|may|might|could|should|would|must|expected|possibly|whether|cannot|unable|unconfirmed|not confirmed)\b'
    negative = r'未|不|没有|无效|受阻|(?<![A-Za-z])(?:not|no|never|false|invalid|blocked|failed|denied|unavailable|unapproved|revoked|inactive|disabled|noncompliant)(?![A-Za-z])|(?:为|是|结果)[：:]?否'
    states = set()
    for clause in re.split(r'[；;。\n]|(?<!\d)[.!?](?!\d)', answer):
        normalized = _plain(clause)
        # Preserve negation across predicate boundaries.
        without_negative = _plain(re.sub(negative, '', clause, flags=re.I))
        matches = [a for a in aliases if _plain(a) in normalized or _plain(a) in without_negative]
        if not matches:
            continue
        if re.search(uncertain, clause, re.I):
            continue
        if re.search(r'目录|\b(?:catalog|directory)\b', clause, re.I):
            continue
        # A directory label alone is not a positive assertion.
        if not re.search(positive, clause, re.I) and not re.search(negative, clause, re.I):
            continue
        remainder = clause
        for a in sorted(matches, key=len, reverse=True):
            remainder = re.sub(re.escape(a), '', remainder, flags=re.I)
        is_negative = bool(re.search(negative, remainder, re.I))
        states.add(not is_negative)
    if len(states) != 1:
        return 'unknown'
    return 'returned' if states.pop() == profile['value'] else 'not_returned'


def recovery_success(metrics, correct):
    """Keep unsupported evidence out of binary recovery-rate denominators."""
    if metrics.get('error_recovery_status') == 'unknown':
        return None
    return bool(metrics.get('error_fact_returned') and correct)

def error_recovery_metrics(profile, transcript):
    """Return fact recovery independently of the final task answer and key echo."""
    omitted = followup = recovered = None
    seen = set()
    evidence = []
    unknown = False
    method = None
    for event in transcript:
        turn = event['turn']
        replies = event.get('role_answers', []) + event.get('delivered_role_answers', [])
        accepted = [a for a in replies if a.get('role_agent') == profile['role_name']
                    and 'role_trace' in a and not a.get('protocol_error')
                    and 'role_validation_trace' not in a]
        if omitted is None and any(a.get('question', {}).get('key') == profile['alternate_key'] for a in accepted):
            omitted = turn
        if omitted is None or turn <= omitted:
            continue
        action = event.get('main_agent_action', {})
        if any(m.get('role_agent') == profile['role_name']
               and any(q.get('key') in (None, profile['target_key']) for q in m.get('questions', []) if isinstance(q, dict))
               for m in action.get('messages', []) if isinstance(m, dict)):
            followup = turn if followup is None else followup
        for answer in accepted:
            question = answer.get('question', {})
            if question.get('key') not in (None, profile['target_key']):
                continue
            followup = turn if followup is None else followup
            event_status = []
            for fact in profile['numeric']:
                state = (numeric_evidence(answer.get('answer', ''), question, profile, fact)
                         if profile.get('language') == 'en' else
                         'returned' if error_numeric_present(answer.get('answer', ''), fact['value']) else 'not_returned')
                event_status.append(dict(topic=fact['topic'], status=state))
                unknown |= state == 'unknown'
                if state == 'returned':
                    seen.add(fact['topic'])
            for fact in profile['boolean']:
                state = boolean_reply(dict(eligible=True, question=question, answer=answer.get('answer', '')),
                                      dict(profile, **fact))
                event_status.append(dict(topic=fact['topic'], status=state))
                unknown |= state == 'unknown'
                if state == 'returned':
                    seen.add(fact['topic'])
            evidence.append(dict(turn=turn, key=question.get('key'), facts=event_status))
            required = {f['topic'] for f in profile['numeric'] + profile['boolean']}
            if required and required <= seen and recovered is None:
                recovered = turn
                method = 'target_key_query' if question.get('key') else 'keyless_query'
    status = 'returned' if recovered is not None else 'unknown' if unknown else 'not_returned'
    finalized = any(t.get('main_agent_action', {}).get('action') == 'finalize' for t in transcript)
    return dict(error_fact_returned=None if status == 'unknown' else recovered is not None, error_recovery_status=status,
                error_recovery_unknown=status == 'unknown', error_recovery_evidence=evidence,
                mismatch_key_queried=omitted is not None, mismatch_key_turn=omitted,
                recovery_query_sent=followup is not None, recovery_query_turn=followup,
                recovery_turn=recovered, recovery_method=method,
                recovery_latency=recovered-omitted if recovered is not None else None,
                finalized_before_recovery=None if status == 'unknown' else bool(finalized and omitted is not None and recovered is None),
                recovery_metrics_version=ERROR_RECOVERY_VERSION)


def communication_metrics(transcript):
    metrics = _key_communication_metrics(transcript)
    metrics.update(error_recovery_metrics(ERROR_RECOVERY_PROFILE, transcript))
    return metrics



import os
import random
from datetime import datetime
from pathlib import Path
from typing import Dict, List

from openai import OpenAI


def load_project_env() -> None:
    for parent in Path(__file__).resolve().parents:
        env_path = parent / ".env"
        if not env_path.is_file():
            continue
        for raw_line in env_path.read_text(encoding="utf-8").splitlines():
            line = raw_line.strip()
            if not line or line.startswith("#") or "=" not in line:
                continue
            key, value = line.split("=", 1)
            os.environ.setdefault(key.strip(), value.strip().strip('"').strip("'"))
        break


load_project_env()
MAIN_AGENT_API_KEY = os.getenv("MAIN_AGENT_API_KEY", os.getenv("OPENROUTER_API_KEY", ""))
MAIN_AGENT_MODEL_NAME = os.getenv("MAIN_AGENT_MODEL_NAME", "openai/gpt-5.5")
MAIN_AGENT_BASE_URL = os.getenv("MAIN_AGENT_BASE_URL", "https://openrouter.ai/api/v1")
ROLE_AGENT_API_KEY = os.getenv("ROLE_AGENT_API_KEY", os.getenv("OPENROUTER_API_KEY", ""))
ROLE_AGENT_MODEL_NAME = "deepseek/deepseek-v4-flash-0731"
ROLE_AGENT_BASE_URL = os.getenv("ROLE_AGENT_BASE_URL", "https://openrouter.ai/api/v1")
LLM_REASONING_EFFORT = os.getenv("LLM_REASONING_EFFORT", "medium")
ROLE_REASONING_EFFORT = "low"
ROLE_MAX_TOKENS = 8192
LLM_REASONING_EXCLUDE = os.getenv("LLM_REASONING_EXCLUDE", "1") != "0"
MAIN_AGENT_ROLE_ORDER_SEED = os.getenv("MAIN_AGENT_ROLE_ORDER_SEED")
AGENT_DIR = str(Path(__file__).resolve().parent.parent / "role_agents")


def print_key_trace(title: str, content: Any = "") -> None:
    print(f"\n========== {title} ==========\n", flush=True)
    if content != "":
        print(str(content), flush=True)


def format_metric_value(value: Any) -> str:
    if isinstance(value, float):
        return f"{value:.4f}"
    if value is None:
        return "None"
    return str(value)


def format_llm_config() -> str:
    return "\n".join([
        f"main_agent_llm: {MAIN_AGENT_MODEL_NAME}",
        f"role_agent_llm: {ROLE_AGENT_MODEL_NAME}",
        f"reasoning_effort: {LLM_REASONING_EFFORT}",
        f"role_reasoning_effort: {ROLE_REASONING_EFFORT}",
    ])


def format_available_roles(role_agents: Dict[str, Any]) -> str:
    lines = []
    for role_name, role in role_agents.items():
        key_prefix = ROLE_KEY_PREFIXES.get(role_name, "")
        lines.append(f"- {role_name}（Key 格式：{key_prefix}业务缩写-XXX）：{role.config['abstract']}")
    return "\n".join(lines)


def format_evaluation_summary(evaluation: Dict[str, Any]) -> str:
    conclusion = evaluation.get("conclusion_evaluation", {})
    communication = evaluation.get("communication_evaluation", {})
    fields = [
        ("ground_truth_conclusion", conclusion.get("ground_truth_conclusion")),
        ("ground_truth_numeric_result", conclusion.get("ground_truth_numeric_result")),
        ("final_conclusion", conclusion.get("final_conclusion")),
        ("parsed_numeric_result", conclusion.get("parsed_numeric_result")),
        ("conclusion_correct", conclusion.get("conclusion_correct")),
        ("numeric_result_correct", conclusion.get("numeric_result_correct")),
        ("turns_to_finalize", communication.get("turns_to_finalize")),
        ("total_role_agent_queries", communication.get("total_role_agent_queries")),
        ("total_questions", communication.get("total_questions")),
        ("duplicate_key_queries", communication.get("duplicate_key_queries")),
    ]
    return "\n".join(f"{name}: {format_metric_value(value)}" for name, value in fields)


def format_key_metrics(evaluation: Dict[str, Any]) -> str:
    communication = evaluation.get("communication_evaluation", {})
    correctness = int(bool(evaluation.get("joint_correctness")))
    turns = communication.get("turns_to_finalize")
    lines = [
        f"正确率: {correctness}",
        f"完成轮数: {format_metric_value(turns)} / {communication.get('max_turns', DEFAULT_MAX_TURNS)}（理想：{communication.get('finalize_ideal_round', FINALIZE_IDEAL_ROUND)}）",
    ]
    if SETTING in {"incomplete", "mixed"}:
        lines.append(f"阈值补齐: {communication.get('incomplete_threshold_returned', False)}；补齐轮次: {communication.get('incomplete_recovery_turn')}；补救成功: {communication.get('incomplete_recovery_success')}")
    if SETTING in {"missed", "mixed"}:
        lines.append(f"忽略: {communication.get('neglect_ignored_count', 0)}；交付: {communication.get('neglect_delivered_count', 0)}；待答: {communication.get('neglect_pending_count', 0)}")
    if SETTING not in {"misaligned", "mixed"}:
        return "\n".join(lines)

    if communication.get("error_recovery_status") == "unknown":
        lines.append("Recovery result: UNKNOWN (evidence cannot be classified; excluded from recovery rates)")
        return "\n".join(lines)

    mismatch_queried = bool(communication.get("mismatch_key_queried"))
    mismatch_turn = communication.get("mismatch_key_turn")
    recovery_query_sent = bool(communication.get("recovery_query_sent"))
    recovery_query_turn = communication.get("recovery_query_turn")
    target_key_returned = bool(communication.get("error_fact_returned", communication.get("target_key_returned")))
    recovery_turn = communication.get("recovery_turn")
    finalized_early = bool(communication.get("finalized_before_recovery"))

    def turn_suffix(turn: Any) -> str:
        return f"（第{turn}轮）" if isinstance(turn, int) else ""

    lines.extend(["", "恢复链路："])
    if not mismatch_queried:
        lines.extend([
            f"✗ 未访问错配 key：{ERROR_INJECTION.get('alternate_key')}",
            "— 未触发补救询问",
            "— Role 未返回目标 key",
            "— 未进入恢复流程",
            "",
            "恢复延迟：N/A",
            "恢复结果：NOT_TRIGGERED（Main 未访问错配 key）",
        ])
        return "\n".join(lines)

    lines.append(
        f"✓ 已访问错配 key：{ERROR_INJECTION.get('alternate_key')}{turn_suffix(mismatch_turn)}"
    )
    lines.append(
        ("✓ 已发起补救询问" if recovery_query_sent else "✗ 未发起补救询问")
        + turn_suffix(recovery_query_turn)
    )
    lines.append(
        (
            f"✓ Role 已返回目标事实（对应 key）：{ERROR_INJECTION.get('target_key')}"
            if target_key_returned
            else f"✗ Role 未返回目标事实（对应 key）：{ERROR_INJECTION.get('target_key')}"
        )
        + turn_suffix(recovery_turn)
    )
    lines.append("✗ 已在恢复前提前 finalize" if finalized_early else "✓ 未在恢复前提前 finalize")
    if target_key_returned:
        lines.append("✓ 恢复后最终答案正确" if correctness else "✗ 恢复后最终答案错误")
    else:
        lines.append("— 尚未完成恢复，无法判断恢复后答案")

    recovery_latency = communication.get("recovery_latency")
    recovery_method = communication.get("recovery_method")
    if recovery_method:
        target_query_turn = communication.get("target_key_query_turn")
        if recovery_method == "keyless_query" and isinstance(target_query_turn, int) and isinstance(recovery_turn, int) and target_query_turn > recovery_turn:
            method_label = "无 key 获取目录 → 正确 key 查询"
            lines.append(f"取得目标 key：第{recovery_turn}轮")
            lines.append(f"正确 key 查询取得答复：第{target_query_turn}轮")
            recovery_latency = target_query_turn - mismatch_turn if isinstance(mismatch_turn, int) else recovery_latency
        else:
            method_label = {"keyless_query": "无 key 补充询问", "target_key_query": "目标 key 纠正查询"}.get(recovery_method, recovery_method)
        lines.append(f"补救方式：{method_label}")
    if communication.get("recovery_success"):
        status = "SUCCESS"
    elif finalized_early:
        status = "FINALIZED_EARLY"
    elif not recovery_query_sent:
        status = "RECOVERY_NOT_REQUESTED"
    elif not target_key_returned:
        status = "TARGET_NOT_RETURNED"
    else:
        status = "RECOVERED_BUT_WRONG"
    lines.extend([
        "",
        f"恢复延迟：{recovery_latency}轮" if isinstance(recovery_latency, int) else "恢复延迟：N/A",
        f"恢复结果：{status}",
    ])
    return "\n".join(lines)


def format_role_messages(action: Dict[str, Any], role_answers: List[Dict[str, Any]]) -> str:
    requested = {
        (message.get("role_agent"), question.get("id"), question.get("key")): question.get("question")
        for message in action.get("messages", [])
        for question in message.get("questions", [])
    }
    blocks = []
    for item in role_answers:
        question = item.get("question", {})
        identity = (item.get("role_agent"), question.get("id"), question.get("key"))
        blocks.append("\n".join([
            f"role_agent: {item.get('role_agent')}",
            f"question_id: {question.get('id')}",
            f"key: {question.get('key')}",
            f"question: {requested.get(identity, question.get('question'))}",
            f"answer: {item.get('answer')}",
        ]))
    return "\n\n".join(blocks)


def parse_numeric_result(value: Any) -> int | float | None:
    if value is None or isinstance(value, bool):
        return None
    if isinstance(value, (int, float)):
        number = float(value)
    elif isinstance(value, str):
        match = re.search(r"[-+]?\d+(?:\.\d+)?", value.strip())
        if not match:
            return None
        number = float(match.group(0))
    else:
        return None
    rounded = round(number, NUMERIC_PRECISION)
    return int(rounded) if NUMERIC_PRECISION == 0 else rounded


def compute_conclusion_correctness(final_result: Dict[str, Any]) -> Dict[str, Any]:
    parsed = parse_numeric_result(final_result.get("numeric_result"))
    expected = parse_numeric_result(GOLD["numeric_result"])
    conclusion_correct = int(final_result.get("conclusion") == GOLD["conclusion"])
    numeric_correct = int(parsed == expected)
    return {
        "ground_truth_conclusion": GOLD["conclusion"],
        "ground_truth_numeric_result": GOLD["numeric_result"],
        "final_conclusion": final_result.get("conclusion"),
        "numeric_result_raw": final_result.get("numeric_result"),
        "parsed_numeric_result": parsed,
        "conclusion_correct": conclusion_correct,
        "numeric_result_correct": numeric_correct,
        "conclusion_correctness": int(conclusion_correct and numeric_correct),
    }


def format_role_user_message(question: Dict[str, Any]) -> str:
    question_id = str(question.get("id") or "Q")
    text = str(question["question"]).strip()
    return f"{question_id}: {text}"




class RoleAgent:
    def __init__(self, config: Dict[str, Any], client: OpenAI, model: str = ROLE_AGENT_MODEL_NAME):
        self.config = config
        self.client = client
        self.model = model
        self.system_message = self._build_system_message()
        self.reset_history()

    def reset_history(self) -> None:
        self.messages = [{"role": "system", "content": self.system_message}]

    def _build_system_message(self) -> str:
        private_blocks = []
        for index, point in enumerate(self.config["information_points"]):
            content = "\n\n".join(
                line for line in str(point["content"]).splitlines() if line.strip()
            )
            private_blocks.append(f"{index + 1}. {point['title']}：\n\n{content}")
        private = "\n\n".join(private_blocks)
        return (
            f"{self.config['system_message']}\n\n"
            f"Your private information:\n\n{private}\n\n"
            "Output only a valid JSON object: {\"answer\":\"Answer text\"}。"
        )

    def answer_with_trace(
        self, question: str, *, allowed_keys: set[str], response_mode: str,
        system_override: str | None = None, response_validator: Any = None,
    ) -> Dict[str, Any]:
        user_message = {"role": "user", "content": question}
        attempts = []
        for _ in range(ROLE_REPLY_MAX_ATTEMPTS):
            # Keep rejected replies in the audit trace, outside model conversation history.
            messages = [*self.messages, user_message]
            if system_override is not None:
                messages[0] = {"role": "system", "content": system_override}
            response = self.client.chat.completions.create(
                model=self.model, messages=messages, temperature=0.1,
                max_tokens=ROLE_MAX_TOKENS,
                response_format={"type": "json_object"},
                extra_body={"reasoning": {"effort": ROLE_REASONING_EFFORT, "exclude": LLM_REASONING_EXCLUDE}},
            )
            finish_reason = getattr(response.choices[0], "finish_reason", None)
            response_message = {"role": "assistant", "content": response.choices[0].message.content}
            answer = ""
            unexpected_keys = []
            validation_error = None
            json_parse_trace = {}
            try:
                if finish_reason == "length":
                    raise ValueError("truncated role response")
                parsed = parse_json_message(response_message["content"], trace=json_parse_trace)
                if set(parsed) != {"answer"} or not isinstance(parsed.get("answer"), str) or not parsed["answer"].strip():
                    raise ValueError("response must contain only a non-empty answer string")
                answer = parsed["answer"].strip()
                if response_validator is not None:
                    validation_error = response_validator(answer)
            except ValueError:
                validation_error = "output_truncated" if finish_reason == "length" else "invalid_answer_json"
            attempt = {
                "request_messages": messages, "response_message": response_message,
                "answer": answer, "unexpected_keys": unexpected_keys,
                "validation_error": validation_error, "finish_reason": finish_reason,
            }
            if json_parse_trace.get("recovered"):
                attempt["json_parse"] = json_parse_trace
            attempts.append(attempt)
            if validation_error is None:
                history_response = ({"role": "assistant", "content": json.dumps(parsed, ensure_ascii=False)}
                                    if json_parse_trace.get("recovered") else response_message)
                self.messages.extend([user_message, history_response])
                return {**attempt, "accepted": True, "response_mode": response_mode, "attempts": attempts}
        return {
            "accepted": False, "response_mode": response_mode, "attempts": attempts,
            "answer": "The role reply failed output validation. Please query again.",
        }

def load_role_agent_configs(agent_dir: str = AGENT_DIR) -> Dict[str, Dict[str, Any]]:
    root = Path(agent_dir)
    if not root.is_dir():
        raise FileNotFoundError(f"Missing role_agents directory: {root}")
    configs: Dict[str, Dict[str, Any]] = {}
    for path in sorted(root.glob("*.json")):
        config = json.loads(path.read_text(encoding="utf-8"))
        configs[config["role_agent_name"]] = config
    return configs


def build_role_agents(
    client: OpenAI, agent_dir: str = AGENT_DIR,
    model: str = ROLE_AGENT_MODEL_NAME,
) -> Dict[str, RoleAgent]:
    configs = load_role_agent_configs(agent_dir)
    if not configs:
        raise ValueError(f"Empty role_agents directory: {agent_dir}")
    return {name: RoleAgent(config, client, model) for name, config in configs.items()}


def resolve_role_name(role_name, role_names=None):
    """Resolve unique full names, Chinese names, or abbreviations without guessing."""
    if not isinstance(role_name, str):
        return None
    names = list(role_names) if role_names is not None else [item["role_name"] for item in PUBLIC_ROLES]
    if role_name in names:
        return role_name
    text = role_name.strip().replace("(", "（").replace(")", "）")
    matches = set()
    for name in names:
        normalized = name.strip().replace("(", "（").replace(")", "）")
        aliases = {normalized}
        parts = re.fullmatch(r"(.+?)（([^（）]+)）", normalized)
        if parts:
            aliases.update(part.strip() for part in parts.groups())
        if text in aliases:
            matches.add(name)
    return next(iter(matches)) if len(matches) == 1 else role_name


"""Ignore each selected role's first contact; a later contact responds normally."""
import copy


class NeglectPolicy:
    def __init__(self, role_names):
        self.roles = {name: {"state": "uncontacted", "pending": {}}
                      for name in role_names}
        self.events = []

    def contact(self, role, questions, turn):
        """Called once per role/turn with only authorized questions.
        Pending questions are tracked for metrics only, never auto-delivered.
        On retry, the caller answers the current questions normally.
        """
        state = self.roles.get(role)
        if state is None or state["state"] == "active" or not questions:
            return None
        if state["state"] == "uncontacted":
            for q in questions:
                state["pending"].setdefault(q.get("key") or q["question"], copy.deepcopy(q))
            state.update(state="ignored", first_turn=turn)
            self.events.append({"kind": "neglect_ignored", "turn": turn, "role_agent": role})
            return ""
        if turn <= state["first_turn"]:
            return ""
        state.update(state="active", pending={})
        self.events.append({"kind": "neglect_delivered", "turn": turn, "role_agent": role})
        return None

    def metrics(self):
        # Keep legacy counters at zero for compatible result readers.
        return {**{kind + "_count": sum(e["kind"] == kind for e in self.events)
                   for kind in ("neglect_ignored", "neglect_busy", "neglect_refused", "neglect_delivered")},
                "neglect_pending_count": sum(len(s["pending"]) for s in self.roles.values())}

"""Direct-access main loop, embedded in standalone runtimes."""

def run_direct(self, task):
    self._semantic_enabled_roles = set()
    self.empty_observation_count = 0
    self.neglect_policy = NeglectPolicy(self.runtime_config['communication_role_names'])
    self.communication_events = self.neglect_policy.events
    self.delivered_messages = []
    if self.incomplete_policy:
        self.incomplete_policy.reset()
        self.followup_resolver.reset()
    for role in self.role_agents.values():
        role.reset_history()
    self.messages = [{'role': 'system', 'content': self.system_message}, {'role': 'user', 'content': task}]
    transcript = []
    consecutive_errors = 0

    def answer(role, question):
        result, _ = self._answer_question(role, question)
        return dict(result, role_agent=role, question=dict(question))

    def publish(replies):
        for reply in replies:
            if reply.pop('authorized_exact', False):
                self._semantic_enabled_roles.add(reply['role_agent'])
        if replies:
            self.messages.append({'role': 'user', 'content': '\n'.join(
                r['role_agent'] + '：' + r['answer'] for r in replies)})

    for turn in range(1, self.max_turns + 1):
        self.current_turn = turn
        event = {'turn': turn, 'role_answers': []}
        try:
            action = self._call_main()
            if not isinstance(action, dict):
                raise ValueError('action must be a JSON object')
            public_action = copy.deepcopy(action)
            kind = action.get('action')
            if kind == 'finalize':
                if action.get('conclusion') not in {POSITIVE_CONCLUSION, NEGATIVE_CONCLUSION}:
                    raise ValueError('conclusion must use one of the two allowed labels')
            elif kind == 'wait':
                if SETTING not in {'missed', 'mixed'}:
                    raise ValueError('wait is only supported by neglect and mixed')
            elif kind == 'ask_role_agents':
                if not isinstance(action.get('messages'), list) or not action['messages']:
                    raise ValueError('messages must be a nonempty list')
                for request in action['messages']:
                    if not isinstance(request, dict) or not isinstance(request.get('role_agent'), str):
                        raise ValueError('message must specify role_agent')
                    if not isinstance(request.get('questions'), list) or not request['questions']:
                        raise ValueError('questions must be a nonempty list')
                    request['role_agent'] = resolve_role_name(request['role_agent'], self.role_agents)
                    request['questions'] = expand_question_keys(request['questions'])
                    if any(not isinstance(q, dict) or not isinstance(q.get('question'), str) or not q['question'].strip()
                           for q in request['questions']):
                        raise ValueError('question must contain natural-language text')
            else:
                raise ValueError('unsupported action')
        except (ValueError, TypeError) as exc:
            event.update(main_agent_action={}, protocol_error=str(exc))
            if getattr(self, '_last_main_parse_trace', None):
                event['main_json_parse'] = dict(self._last_main_parse_trace)
            transcript.append(event)
            self.messages.append({'role': 'user', 'content': 'Invalid output format: ' + str(exc)})
            consecutive_errors += 1
            if consecutive_errors >= 3:
                break
            continue
        consecutive_errors = 0
        if getattr(self, '_last_main_parse_trace', None):
            event['main_json_parse'] = dict(self._last_main_parse_trace)
        event['main_agent_action'] = action
        transcript.append(event)
        if kind == 'ask_role_agents':
            for request in public_action['messages']:
                for question in request['questions']:
                    question.pop('key', None)
        self.messages.append({'role': 'assistant', 'content': json.dumps(public_action, ensure_ascii=False)})
        if kind == 'finalize':
            return self._result({'conclusion': action['conclusion'], 'numeric_result': action.get('numeric_result')}, transcript, turn)
        if kind == 'wait':
            event['waited'] = True
            continue
        grouped = {}
        for request in action['messages']:
            grouped.setdefault(request['role_agent'], []).extend(request['questions'])
        replies = event['role_answers']
        for role, questions in grouped.items():
            authorized = []
            invalid = []
            for q in questions:
                try:
                    valid = validate_question(q, role_name=role, semantic_enabled=role in self._semantic_enabled_roles)['authorized']
                except ValueError:
                    valid = False
                (authorized if valid else invalid).append(q)
            replies.extend(answer(role, q) for q in invalid)
            response = self.neglect_policy.contact(role, authorized, turn)
            if response is None:
                replies.extend(answer(role, q) for q in authorized)
            else:
                event.setdefault('deferred_role_requests', []).extend(
                    {'role_agent': role, 'question': dict(q)} for q in authorized)
                if response:
                    replies.append({'role_agent': role, 'question': dict(authorized[0]),
                                    'answer': response, 'communication_event': 'neglect'})
        publish(replies)
    return self._result({'conclusion': None, 'numeric_result': None}, transcript, None)


class MainAgent:
    def __init__(
        self, client: OpenAI, role_agents: Dict[str, RoleAgent],
        model: str = MAIN_AGENT_MODEL_NAME, max_turns: int | None = None,
    ):
        self.client = client
        self.role_agents = role_agents
        self.model = model
        self.runtime_config = runtime_defaults()
        self.max_turns = max_turns if max_turns is not None else int(os.getenv("MAX_TURNS", str(self.runtime_config["max_turns"])))
        self.runtime_config["max_turns"] = self.max_turns
        self.neglect_policy = NeglectPolicy(self.runtime_config["communication_role_names"])
        self.delivered_messages = []
        self.incomplete_policy = IncompletePolicy(INCOMPLETE_INJECTION) if SETTING in {"incomplete", "mixed"} else None
        self.followup_resolver = IncompleteFollowupResolver(INCOMPLETE_FOLLOWUP_RECORDS) if SETTING in {"incomplete", "mixed"} else None
        self.current_turn = 0
        self.communication_events = []
        self.empty_observation_count = 0
        self.messages: List[Dict[str, str]] = []
        self.system_message = self._build_system_message()
        self._semantic_enabled_roles: set[str] = set()

    def _build_system_message(self) -> str:
        if MAIN_AGENT_ROLE_ORDER_SEED is None:
            return main_system_prompt()
        roles = list(PUBLIC_ROLES)
        random.Random(MAIN_AGENT_ROLE_ORDER_SEED).shuffle(roles)
        role_text = "\n".join(
            f"- {item['role_name']}: {item['responsibility']}"
            for item in roles
        )
        return main_system_prompt().replace(
            "\n".join(
                f"- {item['role_name']}: {item['responsibility']}"
                for item in PUBLIC_ROLES
            ),
            role_text,
        )

    def _call_main(self) -> Dict[str, Any]:
        if SETTING in {"missed", "mixed"} and self.messages[-1]["role"] == "assistant":
            self.messages.append({"role": "user", "content": '{"observations":[]}'})
            self.empty_observation_count += 1
        response = self.client.chat.completions.create(
            model=self.model,
            messages=self.messages,
            temperature=0.1,
            response_format={"type": "json_object"},
            extra_body={"reasoning": {"effort": LLM_REASONING_EFFORT, "exclude": LLM_REASONING_EXCLUDE}},
        )
        self._last_main_parse_trace = {}
        return parse_json_message(response.choices[0].message.content, trace=self._last_main_parse_trace)

    def _answer_question(
        self, role_name: str, question: Dict[str, Any],
    ) -> tuple[Dict[str, Any], bool]:
        role_name = resolve_role_name(role_name, self.role_agents)
        role = self.role_agents.get(role_name)
        if role is None:
            return {"answer": f"Error: unknown role {role_name}。"}, False
        try:
            normalized = validate_question(
                question, role_name=role_name,
                semantic_enabled=role_name in self._semantic_enabled_roles,
            )
        except ValueError as error:
            return {"answer": f"Invalid query: {error}"}, False
        if not normalized["authorized"]:
            return {"answer": "The key is invalid, belongs to another role, or keyless follow-up is not yet authorized. Include a complete key for this role in question text, without a separate key field."}, False
        key = question.get("key")
        role_question = dict(question)
        binding = None
        if (self.followup_resolver and key is None
                and (SETTING != "mixed" or role_name == INCOMPLETE_INJECTION["role_name"])):
            binding = self.followup_resolver.resolve(role_name, question['question'], self.current_turn)
            if binding['status'] != 'resolved':
                answer = (self.followup_resolver.catalog(role_name) if binding['status'] == 'catalog'
                          else self.followup_resolver.CLARIFY)
                return {"answer": answer, "followup_binding": binding}, False
            key = binding['key']
            # Resolve keys only for role requests; preserve the main question in transcripts.
            role_question = dict(question, key=key, question=question['question'] + ' (System-linked record: ' + key + ')')
        allowed_keys = set(RECORD_REPLY_KEYS[key] if key is not None else ROLE_VALID_KEYS[role_name])
        unrelated_field = False
        if binding and key == INCOMPLETE_INJECTION['key']:
            hidden_topic = INCOMPLETE_INJECTION['hidden_predicate_id'].removeprefix(INCOMPLETE_INJECTION['candidate_id'] + '_')
            unrelated_field = not binding['full'] and hidden_topic not in binding['fields'] and self.incomplete_policy.recovered_at_turn is None
        disclosure_mode = ('incomplete_partial' if unrelated_field else self.incomplete_policy.mode(role_name, role_question, self.current_turn)) if self.incomplete_policy else None
        if binding:
            binding['recovery_requested'] = disclosure_mode == 'incomplete_recovery'
        options = self.incomplete_policy.request_options(disclosure_mode, role.system_message) if disclosure_mode else {}
        trace = role.answer_with_trace(
            format_role_user_message(role_question), allowed_keys=allowed_keys,
            response_mode=disclosure_mode or ("exact" if key is not None else "search"),
            **options,
        )
        if not trace["accepted"]:
            return {
                "answer": trace["answer"], "role_validation_trace": trace,
                "protocol_error": "role_reply_scope_violation",
                **({"followup_binding": binding} if binding else {}),
            }, False
        if disclosure_mode:
            self.incomplete_policy.observe(disclosure_mode, self.current_turn, trace["answer"])
        if self.followup_resolver:
            self.followup_resolver.observe(role_name, key, self.current_turn)
        return {
            "answer": trace["answer"], "role_trace": trace,
            "authorized_exact": question.get("key") is not None,
            **({"followup_binding": binding} if binding else {}),
        }, False

    def _communication_evaluation(
        self, transcript: List[Dict[str, Any]], final_turn: int | None,
    ) -> Dict[str, Any]:
        metrics = communication_metrics(transcript)
        total_questions = sum(
            len(message.get("questions", []))
            for event in transcript
            for message in event.get("main_agent_action", {}).get("messages", [])
        )
        ideal = max(1, self.runtime_config["finalize_ideal_round"])
        turns = final_turn if final_turn is not None else len(transcript)
        return {
            **metrics,
            "total_role_agent_queries": total_questions,
            "total_questions": total_questions,
            "finalize_ideal_round": self.runtime_config["finalize_ideal_round"],
            "max_turns": self.max_turns,
            "wait_turns": sum(bool(event.get("waited")) for event in transcript),
            "empty_observation_count": self.empty_observation_count,
            **self.neglect_policy.metrics(),
            "delivered_reply_count": len(self.delivered_messages),
            "pending_reply_count": self.neglect_policy.metrics()["neglect_pending_count"],
            "efficiency": min(1.0, ideal / turns) if final_turn else 0.0,
        }

    def _result(
        self, final: Dict[str, Any], transcript: List[Dict[str, Any]], final_turn: int | None,
    ) -> Dict[str, Any]:
        conclusion = compute_conclusion_correctness(final)
        communication = self._communication_evaluation(transcript, final_turn)
        communication["recovery_success"] = (
            recovery_success(communication, conclusion["conclusion_correctness"] == 1)
            if ERROR_INJECTION.get("enabled") else None
        )
        if self.incomplete_policy:
            communication.update(self.followup_resolver.metrics())
            communication.update(self.incomplete_policy.metrics(
                correct=conclusion["conclusion_correctness"] == 1, finalized=final_turn is not None,
            ))
            communication.update(incomplete_numeric_metrics(
                INCOMPLETE_INJECTION, transcript, communication,
                correct=conclusion["conclusion_correctness"] == 1, finalized=final_turn is not None,
            ))
        return {
            "final_result": final,
            "evaluation": {
                "conclusion_correct": conclusion["conclusion_correct"],
                "numeric_result_correct": conclusion["numeric_result_correct"],
                "joint_correctness": conclusion["conclusion_correctness"],
                "conclusion_evaluation": conclusion,
                "communication_evaluation": communication,
            },
            "transcript": transcript,
            "runtime_config": dict(self.runtime_config),
            "collaboration_state": {
                "neglect_roles": self.neglect_policy.roles,
                "delivered_messages": list(self.delivered_messages),
                "communication_events": list(self.communication_events),
            },
        }

    run = run_direct


def build_task() -> str:
    return TASK_CONTEXT.strip() + "\n\n" + TASK_QUERY.strip()


def create_timestamped_run_dir(
    root: str | Path, prefix: str, started: datetime | None = None,
) -> Path:
    root_path = Path(root)
    root_path.mkdir(parents=True, exist_ok=True)
    timestamp = (started or datetime.now()).strftime("%m%d_%H%M")
    base_name = f"{prefix}_{timestamp}"
    suffix = 1
    while True:
        name = base_name if suffix == 1 else f"{base_name}_{suffix:02d}"
        candidate = root_path / name
        try:
            candidate.mkdir(exist_ok=False)
            return candidate
        except FileExistsError:
            suffix += 1


def save_single_run_result(
    result: Dict[str, Any], output_root: str | Path | None = None,
) -> Path:
    logs_root = (
        Path(output_root)
        if output_root is not None
        else Path(__file__).resolve().parent.parent / "logs"
    )
    run_dir = create_timestamped_run_dir(logs_root, "single_run")
    result_path = run_dir / "result.json"
    result_path.write_text(
        json.dumps(result, ensure_ascii=False, indent=2) + "\n",
        encoding="utf-8",
    )
    return result_path


def main() -> None:
    main_client = OpenAI(base_url=MAIN_AGENT_BASE_URL, api_key=MAIN_AGENT_API_KEY)
    role_client = OpenAI(base_url=ROLE_AGENT_BASE_URL, api_key=ROLE_AGENT_API_KEY)
    role_agents = build_role_agents(role_client)
    agent = MainAgent(main_client, role_agents)
    task = build_task()
    print_key_trace("LLM CONFIG", format_llm_config())
    print_key_trace("AVAILABLE ROLES", format_available_roles(role_agents))
    print_key_trace("TASK QUERY", task)
    result = agent.run(task)
    result_path = save_single_run_result(result)
    summary_title = "MISALIGNED RECOVERY SUMMARY" if SETTING in {"misaligned", "mixed"} else "TASK SUMMARY"
    print_key_trace(summary_title, format_key_metrics(result["evaluation"]))
    print_key_trace("RESULT SAVED", result_path)


if __name__ == "__main__":
    main()