Context

PR #911 documented the allow_inf_nan=False convention in CLAUDE.md and applied it to Task and AcceptanceCriterion. However, many ConfigDict declarations across the codebase still lack this setting.

Problem

~150+ ConfigDict(frozen=True) declarations do not include allow_inf_nan=False. Models with numeric float fields (costs, rates, percentages, timestamps) could accept inf/NaN values that break downstream arithmetic.

Proposed Solution

Add allow_inf_nan=False to every ConfigDict declaration in src/synthorg/. Priority modules (have float fields):

• providers/models.py (8 declarations)
• settings/models.py (3 declarations)
• templates/schema.py (5 declarations)
• tools/base.py (1 declaration)
• budget/spending_summary.py (2 declarations)
• budget/config.py (2 declarations)
• budget/coordination_metrics.py (6 declarations)
• api/dto.py (30+ declarations)
• engine/ models (many declarations)
• security/ models (many declarations)
• communication/ models (many declarations)
• core/ models (many declarations)

Models without numeric fields get it defensively (guards against future float additions).

Source

Found during PR #911 review (CodeRabbit inline comment on CLAUDE.md).