Output Files

Fed-GWAS writes outputs to the paths declared in each experiment's configs/server/config.yaml and configs/center_x/config.yaml.

Result tree

The shipped tiny correctness config writes to experiments/correctness/tiny_even/results_2/:

results_2/
  server/
    logs/
      server_log.txt
      stage_metrics.csv
    intermediate/
      king_session_*/
      lr_session_*/
  center_1/
    logs/
      client_0_log.txt
      stage_metrics.csv
      client_metrics.csv
    intermediate/
      run_<run_id>/
  center_2/
    logs/
    intermediate/
  stage_metrics.csv
  client_metrics.csv
  retention_manifest.json

Exact filenames can vary by stage and retention tier.

Important files

File	Producer	Purpose
server/logs/server_log.txt	Server strategy	Stage progress, relay behavior, KING/LR execution, completion status.
center_x/logs/client_*_log.txt	Client app	Per-client stage progress and PLINK command details.
stage_metrics.csv	Monitoring runtime	Stage duration and status metrics.
client_metrics.csv	Client monitoring runtime	Client-side resource and network samples when enabled.
king_results_<client_id>.txt	Client KING stage	Local accumulated kinship output after de-anonymization where possible.
lr_results_latest.assoc.logistic	Server LR aggregator	Latest server-side PLINK logistic output copy.
retention_manifest.json	Retention utility	Files kept or pruned at completion.
qc_report.md, lr_report.md, evaluation_report.md	Evaluation tools	Post-run agreement reports against centralized baseline.

Retention

Experiment config.yaml can set:

retention:
  tier: standard
  auto_apply_on_complete: true

Retention is applied when the server reaches done. Use experiments/tools/apply_run_retention.py --dry-run to preview manual pruning.