Command reference
oko scorecard conduct
List the operator's rejections charged to a model, as the rules a later session of that model is given.
Invocation#
oko scorecard conduct (--model MODEL | --session SESSION) [--json]Inputs and options#
- The frustration register and the transcripts of the sessions its moments were said in, read on this machine. Neither the broker nor the fleet's card is read, so it answers for one model without measuring every live session. With --session the model is the one that session's newest model call named, so a session asks for the rules of the model it runs on.
Output and state effect#
- One line per rejection charged to MODEL, oldest first: `<saidAt> (<session> #<ordinal>): <his words>`, then `<N> rejection(s) charged to <model>`. With --json, {model, rules: [{said, saidAt, session, ordinal, model}]}. A model is matched with or without its provider prefix, or by the name the model catalogue declares for it.
Refusals and safe exits#
- `oko scorecard conduct takes exactly one of --model MODEL or --session SESSION` and `oko scorecard conduct does not take <flag>` (exit 2). --session naming a session with no measured transcript or no model call refuses as a verdict's attribution does (`session <id> has no measured transcript …`, `session <id> made no model call before …`). An unreadable frustration register or transcript root refuses with its cause. A verb other than standing or conduct: `oko scorecard takes standing or conduct; got <verb>`. A moment no transcript attributes is charged to no model and appears under no model's rules.