blast culling_limit option behavior

I am using culling_limit 1 as a parameter

From the manual :
Delete a hit that is enveloped by at least this many higher-scoring hits

My understanding :
The culling limit can be used to remove redundant hits. In practice it sets the number of hits returned per subject sequence

The command line $blastn -query reads.fa -subject locus.fa -strand plus -culling_limit 1 -dust no -out result.csv -outfmt 6

One unexpected result :

qseqid  sseqid  pident  length  mismatch    gapopen qstart  qend    sstart  send    evalue  bitscore
QJLFG:08700:06611   gi|372099098:113208001-113426000    98.131  107 1   1   1   106 51978   52084   5.11E-49    185
QJLFG:08700:06611   gi|372099098:113208001-113426000    79.167  120 14  8   103 215 217412  217527  4.15E-15    73.1
QJLFG:08700:06611   gi|372099098:113208001-113426000    97.561  41  1   0   103 143 217437  217477  1.49E-14    71.3

The “3rd hit” as far as i understand is redundant regarding the “2nd hit” : same subject region, same part of the read involved, but it’s a shorter alignement with a higher e-value

Why is it not discarded with culling_limit 1 ?