Value shapes

PostgreSQL answers a value of the wrong type with an error rather than “no match”, so filter[project]=garbage on a uuid column is a request-triggerable 500. (MySQL and SQLite coerce instead — a non-numeric string becomes 0 and quietly matches row zero.) Declare a FilterValueShape on any typed column and values the column cannot hold never reach the driver:

use RoundlyConsulting\QueryBuilder\Enums\FilterValueShape;
use RoundlyConsulting\QueryBuilder\Enums\RequestedOperator;

->allowedFilters(
    AllowedFilter::operators('author', [RequestedOperator::Not], 'author_id', shape: FilterValueShape::Id),
    AllowedFilter::nullable('project', 'project_id', FilterValueShape::Uuid),
    AllowedFilter::relation('label', 'labels', 'labels.id', FilterValueShape::Uuid),
)

// ?filter[author]=garbage     → empty result, no driver error
// ?filter[author]=not:garbage → excludes nothing
// ?filter[project]=12345      → empty result

Boolean also hands the query a real boolean — the text 'false' is not false to any engine. Combined with operators() it gives a flag column a negation:

->allowedFilters(
    AllowedFilter::operators('active', [RequestedOperator::Not], shape: FilterValueShape::Boolean), // not:true
)

How it behaves

• Accepted by operators(), nullable() and relation(); AllowedFilter::boolean() is operators() with the Boolean shape. The default Text guards nothing, so no existing filter changes.
• Values of the wrong shape are dropped before the query is built. If none survive, a match returns an empty result — never the unfiltered list — and a negation excludes nothing, since no row could have matched.
• Id rejects out-of-range numbers too: PostgreSQL fails on an out-of-range bigint exactly as on a malformed uuid.
• A typed shape cannot be combined with a LIKE-family operator (contains, ncontains, starts) or partialByDefault — the declaration throws UnsupportedOperator instead of failing on the first request.