C.32.ACE:9 - Consequences

Consequence	Benefit	Cost
Evals are typed evaluations over declared criteria.	Variant comparison can proceed without collapsing criteria, readings, and decisions.	The team must write the parity frame before using the result in a pattern for the next question.
Expectation-failure tests remain one eval operation when their expectation is declared.	Error prevention remains available without replacing optimization.	Some pass-fail dashboards can no longer drive decisions by themselves.
Losing variants remain useful.	Architecture exploration keeps stepping stones and source-space learning.	The variant archive needs deliberate upkeep.
Proxy and counter-characteristic risks are explicit.	Goodhart pressure is visible before eval results drive work.	More rows may remain monitored as guardrails rather than optimized.