fix: bugs-01

This commit is contained in:
2026-03-12 19:57:30 +08:00
parent 9e609f89a3
commit ce537bb3dc
7 changed files with 62 additions and 37 deletions

View File

@@ -1,32 +1,18 @@
"""Transaction deduplication rules.
Determines whether two transaction records likely represent the same
underlying financial event captured from different screenshots / pages.
Only marks records as duplicate when there is strong deterministic evidence.
Highly similar records (same amount/time/counterparty) are intentionally kept
for manual review to avoid filtering out potential scam brushing transactions.
"""
from datetime import timedelta
from app.models.transaction import TransactionRecord
TIME_WINDOW = timedelta(minutes=5)
def is_duplicate_pair(a: TransactionRecord, b: TransactionRecord) -> bool:
# Rule 1: exact order_no match
# Rule 1: exact order_no match (strong deterministic signal).
if a.order_no and b.order_no and a.order_no == b.order_no:
return True
# Rule 2: same amount + close time + same account tail
if (
float(a.amount) == float(b.amount)
and a.trade_time
and b.trade_time
and abs(a.trade_time - b.trade_time) <= TIME_WINDOW
):
if a.self_account_tail_no and b.self_account_tail_no:
if a.self_account_tail_no == b.self_account_tail_no:
return True
# same counterparty and close time is also strong signal
if a.counterparty_name and a.counterparty_name == b.counterparty_name:
return True
# Intentionally do NOT deduplicate by amount/time similarity.
# Those records should enter the review stage for human confirmation.
return False