arkennemasis Run Report (pass rate + flagged)

A run of the arkennemasis variation pipeline ends with a folder full of delivered images, but "it looks fine" isn't a number. ArkRunReport turns the durable job records into the number that actually matters: first-time pass rate - how many cells passed on the first generation attempt. It's the system's primary health metric, and if you're billing a client for this pipeline, it's also the number you quote.

Why first-time rather than total? Because retries cost money. A pipeline that passes 40% first-time burns a paid regeneration on six out of ten cells; one that passes 90% first-time is nearly twice as cheap to run. When the QC critic is newly added, or the recipe changed, or a client switches to a different material, first-time pass rate is the number that tells you whether you made it better or worse before you spend another batch.

How it works

The node reads every job record in jobs_dir (the JSON files the pipeline writes per cell), tallies statuses, and counts attempts per cell: a cell that passed with one attempt is first-time; passed with more is "passed after retry". It also reports flagged cells - the ones that need an operator, not a retry - and the total generation calls, plus calls per delivered image.

Inputs and outputs

• jobs_dir - where the job records live.
• product - filter to one product; blank counts everything in the folder.
• write_to - optional folder to write report.txt and report.json, so the report outlives the session.
• report - the full human-readable breakdown: attempted, first-time, after retry, flagged, calls, calls per delivered, then the by-status table.
• first_time_pass_rate - the float, 0–1, for whatever wants a number.
• flagged_count, flagged_keys - how many, and which ones.

Install and gotchas

Part of the arkennemasis pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis
pip install -r ComfyUI/custom_nodes/comfyui-arkennemasis/requirements.txt

Restart, or install via ComfyUI Manager by repo URL.

Two things to remember. First, it reads the durable job records, so it must run after the writes - that's the ordering ArkRunCollect enforces, and if your report shows everything pending, the barrier is missing. Second, flagged cells are a category of their own: the report deliberately lists them because a flag means "a human needs to look", not "retry it". If your flagged count keeps climbing on the same product, that's a prompt problem, not a luck problem - and the QC critic is where you'd start looking.