Tom Lane wrote:
> Jesper Krogh <jesper@krogh.cc> writes:
>> It seems to me that the row estimates on a ts_vector search is a bit on
>> the low side for terms that is not in th MCV-list in pg_stats:
>
> tsvector has its own selectivity estimator that's not like plain scalar
> equality. Look into src/backend/tsearch/ts_selfuncs.c if you want to
> see the rules.
Thanks.
least_common_frequence / 2
Which also gives 33 in my situation.
--
Jesper