Command reference
oko scorecard standing
Read the measured standing and declared supervisor wording for each executor and model, rather than choosing a tone from the message text.
Invocation#
oko scorecard standing [--session SESSION | --model MODEL] [--json]Inputs and options#
- The same signed-in defect and claim registers, source-backed Stado compiler charges and live session inventory as sessions scorecard. A selector narrows the projection, and both selectors together are refused.
- The operator's verdicts from the rulings register: each ruling on a `verdict:worst:<session>#<ordinal>` subject (`oko rulings record --session S --ordinal N --subject verdict:worst:S#N`) judges the answer that session gave just before his message #N. The model of that answer is read from the session's own OMP transcript, the newest model call before his message, as compiler charges attribute a commit; a superseded verdict no longer stands.
Output and state effect#
- Standing is verified repair credit points less distinct charged broken commits. A negative standing has a declared concerned level after one charged commit and frustrated after repeated charged commits; nonnegative standing is neutral. The JSON projection retains charged source locations and credit factors, the declared standingLevel or its named failure, and the same pending evidence as sessions scorecard.
- A session or model whose answer the operator judged carries his verdicts in credits.operatorVerdicts and the lowest standing there is (-9223372036854775808), whatever its credits and charges sum to and even when the credit registers are unread; its wording is frustrated, and supervised sends quote his words beside the charged commits. The capacity mix ranks the judged model last. The top-level operatorVerdicts lists every standing verdict with ruling, session, ordinal, said, saidAt, model and answeredAt; the text output prints one `operator verdict` line per verdict.
- Supervisor sends, redirects, review answers and standing supervision dispatch render the declared opening before the instruction, naming the charged commits and compiler errors. The Oko Desktop Work → Sessions → Scorecard screen displays each session's and model's standing, wording level and close-review status, and for a judged card the operator's words, the message they were said in and the model of the answer he judged.
Refusals and safe exits#
- An unreadable charge or credit source, unresolved compiler provenance or absent session refuses supervised delivery with its failed source and leaves standingLevelFailure in the scorecard; Oko does not guess a neutral tone. An unknown selector, unsupported verb or conflicting selectors refuses by name.
- A verdict whose answer cannot be attributed stays listed with `unread` naming why: `session <id> has no measured transcript for the answer judged at <time>; <causes>`, `session <id> made no model call before the verdict at <time>` or `session <id> model call <call> at <time> names no model`. An unreadable rulings register is verdictsFailure. `oko rulings record --subject verdict:worst` without a message is refused: `verdict:worst names no message: a verdict is verdict:worst:<session>#<ordinal>, the message he gave it in`.