Reproducibility

Installation diagnostics

federated-survival doctor --output doctor.json

The report records the package version, Python environment, dependency versions, and availability of all model and protocol adapters.

Audited quickstart

federated-survival quickstart \
  --model DeepSurv \
  --protocol FedAvg \
  --seed 42 \
  --output results/quickstart

The output contains the resolved settings, round-wise metrics, survival predictions, a summary, and a run manifest with artifact hashes.

Seven-model software check

federated-survival run \
  --config federated_survival/examples/seven_model_smoke.json \
  --output results/seven_model_smoke

Model-protocol coverage check

federated-survival run \
  --config federated_survival/examples/four_protocol_smoke.json \
  --output results/four_protocol_smoke

Every configuration-driven invocation writes a resolved configuration, environment metadata, metrics, an explicit failure table, artifact hashes, timestamps, and final run status. Retain these files with any reported result.

The smoke configurations verify software execution. Their metric values should not be interpreted as a statistical comparison of models or protocols.

Function-based comparisons

The composable Python API runs a selected model and federated protocol on paired train/test and client partitions. For repeatable comparisons, record the input data version, split seeds, training budgets, package and dependency versions, and save result.metrics alongside generated plots. The function API does not write a manifest automatically; use the YAML workflow when a complete audit trail is required. See Composable Python experiments.