Candidate Lab

Candidate Lab explains whether Penrose has enough eligible evidence to synthesize testable candidates and where candidate generation loses yield.

The abstraction ladder

decision-backed observations
        ↓
supported family principles
        ↓
cross-family mechanisms
        ↓
generated candidate population
        ↓
capability admission
        ↓
testing and independent confirmation

Each arrow is a real filter. Penrose does not fabricate an abstraction merely to avoid an empty screen.

Funnel terms

• Budget — candidates preregistered before generation.
• Emitted — immutable candidate objects actually produced.
• Admitted — candidates compatible with the current executor and data capabilities.
• Repaired — candidates changed only through bounded mechanical transformations.
• Tested — candidates that reached the referee.
• Confirmed — distinct survivors that cleared a separate confirmation cohort.

Emission is not admission. Admission is not a survivor verdict.

Why yield can be zero

• empty active corpus;
• insufficient independent support for a family;
• synthetic or unfaithful evidence excluded;
• unsupported candidate semantics;
• unavailable required series;
• generator timeout or transient provider failure;
• all candidates rejected by capability preflight.

Candidate Lab should report these reasons rather than presenting zero yield as a mystery.

Safety state

The page must show whether PENROSE_GENERATIVE_LAYER is enabled. Enabling generation does not permit confirmation access or P9 approval.