Skip to main content
Run evaluations with niceeval exp. NiceEval provides three read paths: niceeval query for AI, scripts, and CI; niceeval show for people reading in a terminal; and niceeval view for browser review. All three read published facts within a PublicationCutoff through fixed Inspection operations. An Attempt is visible even when its origin Run is still active.

​
niceeval query

query is the versioned niceeval.query/v1 JSON operation protocol. A successful operation and a protocol-level domain failure each write exactly one canonical document to stdout. Argument errors and process failures that cannot form a document go to stderr. Start with discover, then construct a request from the returned schema. Do not guess fields or invent another query language with SQL, JSON paths, or formulas.
  • discover returns the static catalog, and the schema, permitted selectors, error union, and next minimal request for each operation. It accepts no --record and reads no Record.
  • explain validates a complete request and describes the operation to run and valid next requests.
  • run executes the request and returns that operation’s closed result.
Continuation tokens bind the operation, canonical request, source identity, and PublicationCutoff. A changed binding returns a restart correction.

​
niceeval show

show is the fixed terminal rendering entry point. With no selector, it shows the default overview. --run <run-id> shows one exact Run and can be repeated. --experiment <experiment-id> shows one exact Experiment and can also be repeated. @<locator> selects an Attempt. It cannot be used with --run or --experiment, which also cannot be used together. A command can provide at most one Attempt locator. For an Attempt, --source shows captured source facts and Assertion sites. --execution shows a bounded execution outline, and --timing shows timing activities. --usage shows usage totals and observations. --diff shows file-change windows. These five options are mutually exclusive, and each requires an Attempt locator. --expand <stable-id> can only be used with --execution. The stable ID must be an itemId, toolOccurrenceId, or commandId returned by the outline; it cannot be a positional number or legacy execution handle.

​
niceeval view

view is NiceEval’s fixed first-party local Insight. With no --run, it opens the default overview. Repeated --run selects several exact Runs. It accepts no positional arguments and does not load project pages, themes, components, or renderers. --no-open does not ask the OS to open a browser. --port selects the port for the 127.0.0.1 listener. After the SPA assets, local authorized session, and generation-bound Inspection endpoint are ready, the command writes one openable loopback URL to stdout. With --json, it instead writes niceeval.view/v1 NDJSON ready and closed lifecycle events. Startup and runtime failures use the exit code and stderr. View reads only the project’s canonical .niceeval/record.sqlite. When a new publication appears, the page prompts for refresh and changes to the new cutoff only after confirmation. It provides no --record, SQL, deployment, sharing, Report, Page, theme, renderer, route, or operation parameters.

​
Choose a read path

query, show, and view share closed Inspection operation results. They do not share formatters, routes, components, themes, or presentation schemas.