Skip to main content

parse_context_measure_request

Function parse_context_measure_request 

Source
pub fn parse_context_measure_request(
    params: &Value,
) -> Result<Option<HarnessMeasureRequest>, String>
Expand description

Read the context_measure param of evolution.run. Ok(None) = not requested; a malformed object is an error, never a silently defaulted request.

Deliberately parsed into the SAME HarnessMeasureRequest type harness_measure uses, rather than a context-specific twin. The two arms replay the same task suite on the same deterministic split under the same seed and the same turn cap — only the config being varied differs — so a separate request type would buy nothing but the freedom for the two splits to drift apart, and a context grade taken over a different task set than a harness grade is not comparable with it or with a car-bench-harness CLI run.

Opt-in for the same reason harness_measure is: a benchmark replay is a paid side effect (real model calls, real money), and a daemon that starts spending them because a cycle happened to diagnose something is a daemon nobody can leave running.