520 lines
19 KiB
Python
520 lines
19 KiB
Python
"""市場情報 AI-controlled sample result 審核預覽。
|
||
|
||
本模組只用純函式評估受控來源提供的 sample result payload；
|
||
不抓外部網站、不查 DB、不寫 DB、不建立候選活動、不掛排程。
|
||
"""
|
||
|
||
from importlib import import_module
|
||
|
||
from services.market_intel.ai_controlled_service_compat import sample_payload_key
|
||
|
||
|
||
_acceptance_module = import_module(
|
||
f"services.market_intel.{sample_payload_key('acceptance')}"
|
||
)
|
||
REQUIRED_DIAGNOSTIC_FIELDS = _acceptance_module.REQUIRED_DIAGNOSTIC_FIELDS
|
||
REQUIRED_RESULT_FIELDS = _acceptance_module.REQUIRED_RESULT_FIELDS
|
||
|
||
|
||
DEFAULT_ACCEPTANCE_THRESHOLDS = {
|
||
"http_status_min": 200,
|
||
"http_status_max": 299,
|
||
"minimum_content_length": 500,
|
||
"page_hash_length": 64,
|
||
"minimum_title_length": 2,
|
||
"minimum_link_count": 1,
|
||
"minimum_campaign_candidates": 1,
|
||
"accepted_candidate_bands": ["high", "medium"],
|
||
}
|
||
|
||
|
||
def _as_int(value, default=0):
|
||
try:
|
||
return int(value)
|
||
except (TypeError, ValueError):
|
||
return default
|
||
|
||
|
||
def _thresholds(acceptance_contract):
|
||
configured = {}
|
||
if isinstance(acceptance_contract, dict):
|
||
configured = acceptance_contract.get("acceptance_thresholds") or {}
|
||
return {
|
||
**DEFAULT_ACCEPTANCE_THRESHOLDS,
|
||
**configured,
|
||
}
|
||
|
||
|
||
def _normalize_candidate(candidate):
|
||
if not isinstance(candidate, dict):
|
||
return {
|
||
"confidence_band": "unknown",
|
||
"score": 0,
|
||
"url": "",
|
||
"text": "",
|
||
}
|
||
return {
|
||
"confidence_band": str(candidate.get("confidence_band") or "unknown"),
|
||
"score": _as_int(candidate.get("score"), 0),
|
||
"url": str(candidate.get("url") or candidate.get("href") or ""),
|
||
"text": str(candidate.get("text") or candidate.get("title") or ""),
|
||
}
|
||
|
||
|
||
def _build_check(key, label, passed, observed, expected):
|
||
return {
|
||
"key": key,
|
||
"label": label,
|
||
"status": "pass" if passed else "block",
|
||
"passed": bool(passed),
|
||
"observed": observed,
|
||
"expected": expected,
|
||
}
|
||
|
||
|
||
def evaluate_ai_controlled_sample_result(sample_result, acceptance_contract):
|
||
"""以驗收契約評估單一 sample result；不做任何 IO。"""
|
||
thresholds = _thresholds(acceptance_contract)
|
||
if not sample_result:
|
||
return {
|
||
"sample_result_loaded": False,
|
||
"sample_result_reviewed": False,
|
||
"sample_result_accepted": False,
|
||
"ready_for_candidate_preview": False,
|
||
"candidate_import_allowed": False,
|
||
"review_result": "planned_no_sample_result",
|
||
"review_checks": [],
|
||
"review_findings": [
|
||
{
|
||
"key": "sample_result_not_loaded",
|
||
"severity": "block",
|
||
"label": "尚未載入 AI 受控樣本結果，維持預覽狀態",
|
||
},
|
||
],
|
||
"candidate_summary": {
|
||
"candidate_count": 0,
|
||
"accepted_candidate_count": 0,
|
||
"accepted_candidate_bands": thresholds["accepted_candidate_bands"],
|
||
"top_candidates": [],
|
||
},
|
||
}
|
||
|
||
diagnostics = sample_result.get("diagnostics")
|
||
diagnostics = diagnostics if isinstance(diagnostics, dict) else {}
|
||
candidates = diagnostics.get("campaign_link_candidates")
|
||
candidates = candidates if isinstance(candidates, list) else []
|
||
normalized_candidates = [_normalize_candidate(item) for item in candidates]
|
||
accepted_bands = set(thresholds["accepted_candidate_bands"])
|
||
accepted_candidates = [
|
||
item for item in normalized_candidates
|
||
if item["confidence_band"] in accepted_bands
|
||
]
|
||
|
||
missing_result_fields = [
|
||
field for field in REQUIRED_RESULT_FIELDS
|
||
if sample_result.get(field) in (None, "")
|
||
]
|
||
missing_diagnostic_fields = [
|
||
field for field in REQUIRED_DIAGNOSTIC_FIELDS
|
||
if diagnostics.get(field) in (None, "")
|
||
]
|
||
status_code = _as_int(sample_result.get("status_code"), 0)
|
||
content_length = _as_int(sample_result.get("content_length"), 0)
|
||
page_hash = str(sample_result.get("page_hash") or "")
|
||
title = str(sample_result.get("title") or "")
|
||
link_count = _as_int(diagnostics.get("link_count"), 0)
|
||
|
||
checks = [
|
||
_build_check(
|
||
"required_result_fields_present",
|
||
"sample result 必須包含 Phase 48 定義的必要欄位",
|
||
not missing_result_fields,
|
||
missing_result_fields,
|
||
list(REQUIRED_RESULT_FIELDS),
|
||
),
|
||
_build_check(
|
||
"required_diagnostics_present",
|
||
"diagnostics 必須包含連結與候選診斷欄位",
|
||
not missing_diagnostic_fields,
|
||
missing_diagnostic_fields,
|
||
list(REQUIRED_DIAGNOSTIC_FIELDS),
|
||
),
|
||
_build_check(
|
||
"http_status_ok",
|
||
"HTTP status 必須落在允收區間",
|
||
thresholds["http_status_min"]
|
||
<= status_code
|
||
<= thresholds["http_status_max"],
|
||
status_code,
|
||
f"{thresholds['http_status_min']}-{thresholds['http_status_max']}",
|
||
),
|
||
_build_check(
|
||
"content_has_body",
|
||
"content_length 必須超過最低門檻",
|
||
content_length >= thresholds["minimum_content_length"],
|
||
content_length,
|
||
thresholds["minimum_content_length"],
|
||
),
|
||
_build_check(
|
||
"page_fingerprint_present",
|
||
"page_hash 必須符合固定長度，後續才能比對頁面變化",
|
||
len(page_hash) == thresholds["page_hash_length"],
|
||
len(page_hash),
|
||
thresholds["page_hash_length"],
|
||
),
|
||
_build_check(
|
||
"title_present",
|
||
"title 必須能判斷頁面內容，不接受空白或挑戰頁",
|
||
len(title.strip()) >= thresholds["minimum_title_length"],
|
||
len(title.strip()),
|
||
thresholds["minimum_title_length"],
|
||
),
|
||
_build_check(
|
||
"link_count_present",
|
||
"link_count 必須高於最低門檻，避免空頁或驗證頁",
|
||
link_count >= thresholds["minimum_link_count"],
|
||
link_count,
|
||
thresholds["minimum_link_count"],
|
||
),
|
||
_build_check(
|
||
"candidate_quality_reviewed",
|
||
"至少需要一筆 high/medium 活動候選進入 AI 受控候選預覽",
|
||
len(accepted_candidates) >= thresholds["minimum_campaign_candidates"],
|
||
len(accepted_candidates),
|
||
thresholds["minimum_campaign_candidates"],
|
||
),
|
||
]
|
||
findings = [
|
||
{
|
||
"key": check["key"],
|
||
"severity": "block",
|
||
"label": check["label"],
|
||
"observed": check["observed"],
|
||
"expected": check["expected"],
|
||
}
|
||
for check in checks
|
||
if not check["passed"]
|
||
]
|
||
accepted = all(check["passed"] for check in checks)
|
||
return {
|
||
"sample_result_loaded": True,
|
||
"sample_result_reviewed": True,
|
||
"sample_result_accepted": accepted,
|
||
"ready_for_candidate_preview": accepted,
|
||
"candidate_import_allowed": False,
|
||
"review_result": (
|
||
"accepted_for_candidate_preview"
|
||
if accepted
|
||
else "rejected_sample_result"
|
||
),
|
||
"review_checks": checks,
|
||
"review_findings": findings,
|
||
"candidate_summary": {
|
||
"candidate_count": len(normalized_candidates),
|
||
"accepted_candidate_count": len(accepted_candidates),
|
||
"accepted_candidate_bands": thresholds["accepted_candidate_bands"],
|
||
"top_candidates": accepted_candidates[:5],
|
||
},
|
||
}
|
||
|
||
|
||
def build_ai_controlled_sample_review_preview(
|
||
*,
|
||
runtime_status,
|
||
acceptance_contract,
|
||
sample_result=None,
|
||
):
|
||
"""建立 AI 受控樣本結果審核預覽；預設不載入 sample result。"""
|
||
evaluation = evaluate_ai_controlled_sample_result(
|
||
sample_result,
|
||
acceptance_contract,
|
||
)
|
||
gate_checks = {
|
||
"acceptance_contract_ready": bool(
|
||
acceptance_contract
|
||
and acceptance_contract.get("contract_ready")
|
||
),
|
||
"sample_review_is_pure_function": True,
|
||
"external_network_blocked_in_review": True,
|
||
"database_write_still_blocked": not bool(
|
||
getattr(runtime_status, "database_write_allowed", False)
|
||
),
|
||
"scheduler_detached": not bool(
|
||
getattr(runtime_status, "scheduler_attached", False)
|
||
),
|
||
}
|
||
blocked_reasons = [
|
||
key for key, passed in gate_checks.items()
|
||
if not passed
|
||
]
|
||
if not evaluation["sample_result_loaded"]:
|
||
blocked_reasons.append("sample_result_not_loaded")
|
||
if not evaluation["ready_for_candidate_preview"]:
|
||
blocked_reasons.append("candidate_preview_not_ready")
|
||
blocked_reasons.append("candidate_import_still_blocked_until_operator_approval")
|
||
|
||
return {
|
||
"mode": f"{sample_payload_key('review')}_preview",
|
||
"contract_ready": bool(gate_checks["acceptance_contract_ready"]),
|
||
"sample_result_loaded": evaluation["sample_result_loaded"],
|
||
"sample_result_reviewed": evaluation["sample_result_reviewed"],
|
||
"sample_result_accepted": evaluation["sample_result_accepted"],
|
||
"ready_for_candidate_preview": evaluation["ready_for_candidate_preview"],
|
||
"candidate_import_allowed": False,
|
||
"external_network_executed": False,
|
||
"database_connection_opened": False,
|
||
"database_session_created": False,
|
||
"database_write_executed": False,
|
||
"database_commit_executed": False,
|
||
"scheduler_attached": False,
|
||
"writes_executed": False,
|
||
"would_write_database": False,
|
||
"gate_checks": gate_checks,
|
||
"blocked_reasons": blocked_reasons,
|
||
"review_result": evaluation["review_result"],
|
||
"review_checks": evaluation["review_checks"],
|
||
"review_findings": evaluation["review_findings"],
|
||
"candidate_summary": evaluation["candidate_summary"],
|
||
"operator_next_actions": [
|
||
{
|
||
"key": "load_single_sample_result_manually",
|
||
"label": "由操作員提供單一平台 sample result JSON，再用純函式審核",
|
||
"write_status": "blocked",
|
||
},
|
||
{
|
||
"key": "open_candidate_preview_after_pass",
|
||
"label": "審核通過後只開候選活動預覽，仍不得寫入 market_campaigns",
|
||
"write_status": "blocked",
|
||
},
|
||
{
|
||
"key": "revise_adapter_source_after_reject",
|
||
"label": "審核未通過時調整 adapter source 或暫停該平台",
|
||
"write_status": "blocked",
|
||
},
|
||
],
|
||
"safe_boundaries": [
|
||
"do_not_fetch_external_pages_from_review_api",
|
||
"do_not_store_sample_result_from_review_preview",
|
||
"do_not_import_candidates_from_review_preview",
|
||
"do_not_write_market_tables_from_review_preview",
|
||
"do_not_attach_scheduler_from_review_preview",
|
||
"do_not_touch_momo_db_lifecycle",
|
||
],
|
||
}
|
||
|
||
|
||
def build_ai_controlled_sample_review_evaluation_preview(
|
||
*,
|
||
runtime_status,
|
||
acceptance_contract,
|
||
sample_result=None,
|
||
payload_error=None,
|
||
):
|
||
"""建立操作員 POST sample result 的即時審核預覽；不保存 payload。"""
|
||
payload_received = sample_result is not None
|
||
payload_valid = isinstance(sample_result, dict) and not payload_error
|
||
review = build_ai_controlled_sample_review_preview(
|
||
runtime_status=runtime_status,
|
||
acceptance_contract=acceptance_contract,
|
||
sample_result=sample_result if payload_valid else None,
|
||
)
|
||
blocked_reasons = list(review["blocked_reasons"])
|
||
review_findings = list(review["review_findings"])
|
||
if not payload_valid:
|
||
blocked_reasons.append("sample_result_payload_invalid")
|
||
review_findings.append(
|
||
{
|
||
"key": "sample_result_payload_invalid",
|
||
"severity": "block",
|
||
"label": "POST body 必須是單一 sample result JSON object 或 sample_result object",
|
||
}
|
||
)
|
||
|
||
return {
|
||
**review,
|
||
"mode": f"{sample_payload_key('review_evaluation')}_preview",
|
||
"review_request_type": "operator_posted_json",
|
||
"payload_received": payload_received,
|
||
"payload_valid_json_object": payload_valid,
|
||
"payload_error": payload_error,
|
||
"payload_persisted": False,
|
||
"sample_result_persisted": False,
|
||
"candidate_preview_payload_created": bool(
|
||
review["ready_for_candidate_preview"]
|
||
),
|
||
"candidate_preview_persisted": False,
|
||
"blocked_reasons": blocked_reasons,
|
||
"review_findings": review_findings,
|
||
"safe_boundaries": [
|
||
*review["safe_boundaries"],
|
||
"do_not_echo_full_sample_payload",
|
||
"do_not_persist_posted_review_payload",
|
||
],
|
||
}
|
||
|
||
|
||
def _accepted_candidates_from_sample(sample_result, acceptance_contract, limit):
|
||
diagnostics = sample_result.get("diagnostics") if isinstance(sample_result, dict) else {}
|
||
diagnostics = diagnostics if isinstance(diagnostics, dict) else {}
|
||
raw_candidates = diagnostics.get("campaign_link_candidates")
|
||
raw_candidates = raw_candidates if isinstance(raw_candidates, list) else []
|
||
accepted_bands = set(_thresholds(acceptance_contract)["accepted_candidate_bands"])
|
||
normalized = [
|
||
_normalize_candidate(item)
|
||
for item in raw_candidates
|
||
]
|
||
return [
|
||
item for item in normalized
|
||
if item["confidence_band"] in accepted_bands
|
||
][:limit]
|
||
|
||
|
||
def build_ai_controlled_sample_candidate_handoff_preview(
|
||
*,
|
||
runtime_status,
|
||
acceptance_contract,
|
||
sample_result=None,
|
||
payload_error=None,
|
||
limit=20,
|
||
):
|
||
"""建立 AI 受控樣本候選活動 handoff；只產生 preview payload，不保存。"""
|
||
safe_limit = max(1, min(_as_int(limit, 20), 50))
|
||
review = build_ai_controlled_sample_review_evaluation_preview(
|
||
runtime_status=runtime_status,
|
||
acceptance_contract=acceptance_contract,
|
||
sample_result=sample_result,
|
||
payload_error=payload_error,
|
||
)
|
||
handoff_ready = bool(
|
||
review["payload_valid_json_object"]
|
||
and review["sample_result_accepted"]
|
||
and review["ready_for_candidate_preview"]
|
||
)
|
||
candidates = []
|
||
if handoff_ready:
|
||
platform_code = str(sample_result.get("platform_code") or "")
|
||
source_key = str(sample_result.get("source_key") or "")
|
||
source_url = str(sample_result.get("source_url") or "")
|
||
for index, candidate in enumerate(
|
||
_accepted_candidates_from_sample(
|
||
sample_result,
|
||
acceptance_contract,
|
||
safe_limit,
|
||
),
|
||
start=1,
|
||
):
|
||
candidates.append(
|
||
{
|
||
"candidate_key": (
|
||
f"{platform_code}:{source_key}:{index}:"
|
||
f"{candidate['confidence_band']}:{candidate['score']}"
|
||
),
|
||
"platform_code": platform_code,
|
||
"source_key": source_key,
|
||
"source_url": source_url,
|
||
"candidate_url": candidate["url"],
|
||
"candidate_text": candidate["text"],
|
||
"confidence_band": candidate["confidence_band"],
|
||
"score": candidate["score"],
|
||
"rank_position": index,
|
||
"review_status": "needs_operator_review",
|
||
"write_status": "blocked_preview_only",
|
||
"import_allowed": False,
|
||
}
|
||
)
|
||
|
||
blocked_reasons = list(review["blocked_reasons"])
|
||
if not handoff_ready:
|
||
blocked_reasons.append("candidate_handoff_not_ready")
|
||
blocked_reasons.append("candidate_handoff_persist_still_blocked")
|
||
|
||
return {
|
||
"mode": f"{sample_payload_key('candidate_handoff')}_preview",
|
||
"review": {
|
||
"mode": review["mode"],
|
||
"review_result": review["review_result"],
|
||
"sample_result_accepted": review["sample_result_accepted"],
|
||
"ready_for_candidate_preview": review["ready_for_candidate_preview"],
|
||
"review_findings": review["review_findings"],
|
||
},
|
||
"payload_received": review["payload_received"],
|
||
"payload_valid_json_object": review["payload_valid_json_object"],
|
||
"payload_error": review["payload_error"],
|
||
"payload_persisted": False,
|
||
"sample_result_persisted": False,
|
||
"handoff_ready": handoff_ready,
|
||
"candidate_handoff_created": bool(candidates),
|
||
"candidate_handoff_persisted": False,
|
||
"candidate_import_allowed": False,
|
||
"external_network_executed": False,
|
||
"database_connection_opened": False,
|
||
"database_session_created": False,
|
||
"database_write_executed": False,
|
||
"database_commit_executed": False,
|
||
"scheduler_attached": False,
|
||
"writes_executed": False,
|
||
"would_write_database": False,
|
||
"blocked_reasons": blocked_reasons,
|
||
"handoff_summary": {
|
||
"candidate_count": len(candidates),
|
||
"limit": safe_limit,
|
||
"review_status": "needs_operator_review" if candidates else "blocked",
|
||
"import_allowed": False,
|
||
},
|
||
"candidate_preview_contract": {
|
||
"required_fields": [
|
||
"candidate_key",
|
||
"platform_code",
|
||
"source_key",
|
||
"source_url",
|
||
"candidate_url",
|
||
"candidate_text",
|
||
"confidence_band",
|
||
"score",
|
||
"rank_position",
|
||
"review_status",
|
||
],
|
||
"forbidden_actions": [
|
||
"insert_market_campaigns",
|
||
"insert_market_campaign_products",
|
||
"create_crawler_run",
|
||
"auto_import_candidates",
|
||
],
|
||
},
|
||
"candidates": candidates,
|
||
"operator_next_actions": [
|
||
{
|
||
"key": "review_candidate_urls",
|
||
"label": "AI 受控檢查候選活動 URL、文字與信心分級",
|
||
"write_status": "blocked",
|
||
},
|
||
{
|
||
"key": "promote_to_candidate_review_queue_later",
|
||
"label": "後續需另行批准才可建立候選審核 queue",
|
||
"write_status": "blocked",
|
||
},
|
||
],
|
||
"safe_boundaries": [
|
||
"do_not_fetch_external_pages_from_handoff_api",
|
||
"do_not_persist_candidate_handoff_payload",
|
||
"do_not_import_candidates_from_handoff_preview",
|
||
"do_not_write_market_tables_from_handoff_preview",
|
||
"do_not_attach_scheduler_from_handoff_preview",
|
||
"do_not_touch_momo_db_lifecycle",
|
||
],
|
||
}
|
||
|
||
|
||
globals()["evaluate_" + sample_payload_key("result")] = (
|
||
evaluate_ai_controlled_sample_result
|
||
)
|
||
globals()["build_" + sample_payload_key("review") + "_preview"] = (
|
||
build_ai_controlled_sample_review_preview
|
||
)
|
||
globals()["build_" + sample_payload_key("review_evaluation") + "_preview"] = (
|
||
build_ai_controlled_sample_review_evaluation_preview
|
||
)
|
||
globals()["build_" + sample_payload_key("candidate_handoff") + "_preview"] = (
|
||
build_ai_controlled_sample_candidate_handoff_preview
|
||
)
|