A run log turns "this saves us time" into a figure a finance director can check. That is the difference between a tool a team enjoys and a system a company keeps funding.
It has to be written by the pipeline. Nobody fills in a spreadsheet at eleven at night after a render queue finishes, and a log that depends on discipline is a log that stops after week three.
The useful fields are few: batch identifier, shot type, renders consumed, model and version, spend, gate outcomes, operator time, final disposition. From those you can derive cost per accepted asset, acceptance rate per shot type, and where a pipeline change actually paid.
What should a run log record?
Batch, shot type, renders consumed, model and version, spend, gate outcomes, operator time and final disposition. Everything else can be derived from those.
Why must it be automatic?
Because a manual log depends on somebody doing admin after a long day, and it stops around week three. An automatic one still exists in month six, which is when you need it.