party2d/docs/FULL_RUN_OPERATIONS.md

Full model run operations

This note describes how to launch, monitor, and summarize the long full model run that generates the posterior estimates.

Do not start a full run until raw-data and Stan-data preflight checks pass.

Hardware note for runtime reporting

The production run should record elapsed time for this workstation as:

Hardware: 4 cores of an AMD Ryzen 9 7945HX

After the run, record the wall-time reported by src/sh/show_run_progress.sh or outputs/model_outputs/latest/run_*/diagnostics/run_metrics.json.

Launch with durable logging

From the repository root:

mkdir -p outputs/logs
ts="$(date +%Y%m%d_%H%M%S)"
STAN_REFRESH=100 \
  nohup bash run_estimation.sh full \
  > "outputs/logs/full_run_${ts}.log" 2>&1 &
echo $! > "outputs/logs/full_run_${ts}.pid"

The wrapper configures local raw data, local R libraries, project-local temp space, and the writable project-local CmdStan copy by default.

Monitor during the run

Live log tail:

tail -f outputs/logs/full_run_<timestamp>.log

Convenience progress command:

bash src/sh/show_run_progress.sh

Stan progress is printed every STAN_REFRESH iterations (default 100) and is captured in the durable log.

Inspect after completion

After a successful run, the chain CSVs are under:

outputs/model_outputs/latest/run_<timestamp>/chains/

Run diagnostics and Stan logs are archived under:

outputs/model_outputs/latest/run_<timestamp>/diagnostics/

Use:

bash src/sh/show_run_progress.sh

to print the latest run status, wall time, sampler configuration, divergences, tree-depth hits, and command-configuration verification status.