# Final Django query control

This optional variant changes only the list/search/tag queries. It uses
`select_related`, a field-limited author/post projection, and a field-limited tag
prefetch. Original handlers implement the other five endpoints. It does not patch
or overwrite `blog/` or `core/`.

From the repository root, run `bin/build`, `bin/reference setup`, then
`bin/optimized`. This requires Python 3.14 and uv in addition to the Go and
PostgreSQL prerequisites. The default database is the
dedicated `assignment_harness_django`, and the default port is 18000. Do not run
the reference server on the same port at the same time.

The same 66-case suite and six full-data response comparisons pass. To reproduce:

```sh
harness/harness verify --db 'dbname=assignment_harness_django' \
  --url http://127.0.0.1:18000 --golden fixtures/contract
harness/harness seed --db 'dbname=assignment_harness_django'
harness/harness bench --db 'dbname=assignment_harness_django' \
  --url http://127.0.0.1:18000 --pid MASTER_PID --full \
  --workload posts --requests 3 --repeats 5 --concurrency 1 \
  --output reports/optimized-new-posts.json
```

Repeat for `search` and `tag`. Compare with separately run Go measurements using
the same fixture, preparation, request counts and budgets. These short full-response
runs do not support a reliable p95 estimate or a production capacity claim.

The query counts are comparable, not every implementation detail. Django still
constructs ORM instances and runs its schema/JSON pipeline. Go uses pgx and Go
structs. SQL shapes, drivers and serialization differ. This experiment measures an
optimized implementation comparison, not an isolated programming-language effect.