What RadialD does
When concurrent jobs have the same authority and the same deterministic work identity, RadialD can execute that physical work once and let the other callers join the verified result. In graph mode, identical prefixes can be shared until the first divergent stage.
job A: A → B → X
job B: A → B → Y
physical work:
┌→ X
A → B ──┤
└→ YWhat it does not do
RadialD does not guess that arbitrary programs are equivalent, does not cross authority domains, and does not turn a compute optimization into an assurance conclusion. If the deterministic sharing contract is incomplete, the safe path is isolated execution.
How it improves DDCAL
DDCAL can use RadialD underneath selected deterministic assurance stages to avoid repeating identical computation while retaining the same logical outputs and evidence boundaries. DDC remains responsible for the assurance decision; RadialD improves execution efficiency.
