Why does Lazy<T> keep throwing after the first initialization failure?

Some Lazy initialization modes cache the factory exception. Re-reading Value can therefore repeat the same failure rather than retrying initialization. Select a mode and lifetime that match the required policy. For transient failures requiring retry, replace the failed lazy through a synchronized owner or use an explicit retryable state machine; do not assume every Value access starts over. Keep retries bounded and avoid repeating non-idempotent partial side effects. Use a factory that fails once then succeeds and verify the chosen behavior under concurrency. Changing the mode can permit multiple factory executions, so it changes more than exception handling. A cached failure may be correct for immutable invalid configuration.