Batch processing
Also known as: bulk processing, queued processing
Running the same operation over many files in one unattended pass, instead of opening and handling each clip individually.
Updated
What it means
Batch processing is what makes a large card tractable. One instruction is applied to every file in a set, in order, without a person waiting between items or deciding anything in the middle.
Converting two hundred clips, sampling frames from all of them or scoring each one are all batch jobs, and none of them is interesting individually. The value is entirely in not having to be present.
The appeal is not speed on any single file. It is that the work survives inattention: a queue that takes an hour needs no supervision, while two hundred manual operations need all of it and still go wrong.
How it works in practice
A sound batch job is restartable. Each item records whether it finished, so an interrupted run resumes instead of starting over, which is the same reasoning behind a Resumable upload on the transfer side.
Items that fail are set aside rather than stopping the queue, because one corrupt file should not strand the other hundred and ninety nine behind it.
Ordering is worth thinking about. Running the cheap passes over everything first, such as Frame sampling, reveals which clips deserve an expensive look, so the slow stage runs over a much shorter list.
What to watch out for
Uniform treatment is the hidden assumption. One setting across a mixed card means something is wrong for part of it, and a batch tuned on a bright afternoon session quietly mangles the clips shot at dusk.
Failures also hide in bulk. A job reporting two hundred items complete while eleven produced empty output looks like a success, so counting what came out matters as much as counting what went in.
Resource limits bite differently at scale too. A conversion that is comfortable on one file can exhaust memory on forty in parallel, which usually shows up as a Transcoding failure rather than as anything about the batch.
How RawClip handles it
An order is handled as one queued job rather than file by file, so an upload of several hundred clips is analysed, scored and assembled in a single pass that nobody has to sit and watch, and that resumes rather than restarts.