[antonsoo]

Database.read() applies the write converter a second time, returning integers for boolean columns and floats for decimal columns. For example:

import json
from csvw import TableGroup
from csvw.db import Database

tg = TableGroup.fromvalue({'tables': [{'url': 'data', 'tableSchema': {
    'columns': [{'name': 'flag', 'datatype': 'boolean'}]}}]})
db = Database(tg)
db.write(data=[{'flag': True}, {'flag': False}])
print(json.dumps(db.read()['data']))

Before: [{"flag": 1}, {"flag": 0}]. After: [{"flag": true}, {"flag": false}].

Use the existing DBType.read converter for scalar reads. Add type-sensitive boolean/decimal regressions, including nulls; equality alone misses this because True == 1. Decimal storage keeps SQLite's existing REAL precision.

Validation: full suites pass on Python 3.12 (841 passed, 16 expected failures) and 3.9 (844 passed, 13 expected failures), with 100% configured coverage. Changed source/test Flake8 passes. Shipped Frictionless CSV data and exactly representable decimal values exercise CSV/direct writes, renamed columns and memory/file SQLite: all 12 typed pipelines are repaired, while eight complete non-boolean/table controls and raw SQLite row hashes remain exact. A fresh installed wheel reproduces the results. The Frictionless fixture's unrelated foreign-key metadata is excluded explicitly; its CSV bytes are unchanged.