[matt-edmondson]

Fixes #355

What changed

• SchemaDataValidator: Float and Double values now go through ExpectNumber, which holds a number to its storage type's finite range. Float must satisfy |x| ≤ float.MaxValue and Double must parse to a finite double. Vector, colour, quantity and semantic-type components with Float/Double storage use the same path, so they get the same check. The message states the range, for example Expected a number within the range of Float (±3.4028235E+38) for type 'Float' but found 1e39.
• SchemaDataValues (the optional part of the issue): GetFloat, GetDouble and the vector/colour component readers treat a non-finite result as malformed and throw InvalidOperationException, the same way they already handle a value that doesn't fit. They no longer hand back ±Infinity.

The migrator's Conforms goes through the validator, so it no longer accepts such values either.

Tests

• SchemaDataValidatorTypeTests: ±1e39 in a Float is an error, while float.MaxValue, 1e38 and 0 are accepted. 1e309 in a Double is an error and 1e39 is not. Also covered: a Float vector component (error at data.Value[1]), a colour component, a Float-stored quantity, and a semantic type over Float.
• GeneratedDataEditorTests.TestANumberPastItsTypesRangeThrowsRatherThanReadingAsInfinity: GetFloat, GetDouble and GetVector3 throw. A Double read of 1e39 still succeeds.

With the fix reverted, the 7 new tests fail. With it, the full Schema.Test suite passes on net10.0 (728/728).

This touches SchemaDataValidator.cs like #359 (for #349), but in different methods. It's branched from main on its own, not stacked on that PR.

Main currently fails CA1506 on Schema/Models/Schema.cs (#353 fixes that). I ran the tests locally with that one rule suppressed, and the suppression isn't part of this PR. CI here will stay red on CA1506 until #353 lands.

---

Generated by Claude Code

[matt-edmondson]

CI is red on ci / .NET / Test, but the cause isn't this PR. The only error is CA1506: 'Schema' is coupled with '97' different types at Schema/Models/Schema.cs(18,22), which this diff doesn't touch. main fails the same way at 253db1a (run).

#353 has the fix and is mergeable. I haven't copied that refactor in here, since it would duplicate #353 across three open PRs and they would then conflict with it. Once #353 merges I'll bring main into this branch so CI re-runs. Locally, with only CA1506 suppressed, Schema.Test passes 728/728.

---

Generated by Claude Code