ExpressionDFilter result not as expected

env

python 3.12.9
pylib 0.9.7

data

https://github.com/chenditc/investment_data/releases/tag/2026-01-11
随机验证了一下，$close / $factor = 未复权价格，数据应该没问题

problem

问题：按$close>2000，选出来的票，实际后复权价格不符合$close>2000

from qlib.data import D
from qlib.data.filter import ExpressionDFilter

expressionDFilter = ExpressionDFilter(rule_expression='$close > 2000')
instruments_cfg = D.instruments(market='csi300', filter_pipe=[expressionDFilter])
instruments = D.list_instruments(instruments=instruments_cfg,
                                 start_time='2025-01-01',
                                 end_time='2026-01-09',
                                 as_list=True)
print(instruments)

output

['SH600837', 'SH601989', 'SZ002049']

instruments = ['SZ002049']
fields = ['$close', '$volume', 'Ref($close, 1)', 'Mean($close, 3)', '$high-$low', '$factor']
df = D.features(instruments,
                fields,
                start_time='2025-01-01',
                end_time='2026-01-09',
                freq='day')
print(df)

output

                          $close        $volume  Ref($close, 1)  Mean($close, 3)  $high-$low   $factor
instrument datetime                                                                                   
SZ002049   2025-01-02  54.424286  272530.312500       57.611427        57.426189    3.606804  0.894989
           2025-01-03  53.189999  173090.734375       54.424286        55.075237    1.682606  0.895003
           2025-01-06  52.715714  117793.007812       53.189999        53.443333    1.217205  0.895004
           2025-01-07  54.022858  137582.578125       52.715714        53.309525    1.440968  0.895011
           2025-01-08  53.162857  181623.218750       54.022858        53.300476    2.452293  0.894998
...                          ...            ...             ...              ...         ...       ...
           2025-12-23  68.699997  163727.125000       69.284286        68.697617    1.185257  0.897922
           2025-12-24  69.967140  225979.796875       68.699997        69.317139    1.508530  0.897936
           2025-12-25  71.088570  274110.406250       69.967140        69.918571    1.894615  0.897923
           2025-12-26  72.031425  330256.906250       71.088570        71.029045    2.496223  0.897924
           2025-12-29  70.765717  262366.625000       72.031425        71.295235    1.948502  0.897928

Reproduce the supplied ExpressionDFilter example, then trace the ExpressionDFilter and D.instruments entry points to determine how the $close threshold is evaluated. Compare that result with the D.features output and the stated adjustment expectations; the issue is done when the discrepancy is fixed or its intended behavior is covered by a clear regression test.