Evaluate blocking policies and correlated evidence sources

[marianoviola]

Outcome

Measure how candidate-generation policy and dependent evidence sources affect IdentityREIM calibration and graph resolution.

Acceptance criteria

• At least two blocking policies are compared under matched entity splits.
• Candidate priors are recalibrated for each blocking population.
• Correlated-source failure modes are measured against the conditional-independence model.
• Calibration, pair, partition, and false-component metrics are reported separately.
• Any model change is versioned independently from graph policy changes.