niceeval show for terminal troubleshooting, the fixed query JSON protocol for machines and CI, and view for browser review.
niceeval show --run <run-id> first to confirm the Run’s time, Members, Verdicts, scores, coverage, usage, and limits. The output’s Attempt locator is the precise target for the next step. For an Attempt, run niceeval show @<attempt-locator> and select a projection for the question at hand:
--source shows archived source and assertion facts; --execution shows a bounded execution outline; --timing shows activity, phases, and duration; --usage shows sealed token, request, and cost totals; and --diff shows the archived file-change window. To expand an execution-outline item, use its displayed stable ID: niceeval show @<attempt-locator> --execution --expand <stable-id>.
A result cannot be opened
Published facts have no manual-edit repair path. Make a new Run when you need different facts.
niceeval run recover only closes a Run whose former owner is proven to have ended, while keeping its published Attempts.
A result is incomplete or cannot be compared
Scripts, AI, and CI first runniceeval query discover to find the correct fixed operation, then use explain to inspect the request before running run. The result identifies missing facts, issues, Evidence, and correction. query outputs only a machine protocol; it is not for reading terminal formatting.
For a browser interface, run niceeval view --run <run-id> --no-open. Do not parse the View page or read or query private SQLite tables.
For comparisons, use runs.compare. Its exact and paired modes reject incompatible domains or pairs rather than silently narrowing a denominator.