sl-status

name	sl-status
description	Read-only dashboard for a human-grounded subjective-labeling project. Shows lifecycle phase, open round, D_t/G_t, Session progress, audit and challenge evidence, seven-region coverage, stopping gates, sealed-test validity, executor scorecards, production provenance, implementation holds, and next valid action. Use for /sl-status or progress checks.

Show subjective-label project status

Read state without modifying artifacts, opening seals, rerunning models, or advancing a phase.

Read first

Read ../../ref/ref-output-style.md, ../../ref/ref-assets.md, and ../../ref/ref-stages.md.

Resolve the project directory and inspect, when present:

• REPORT.md, config.yaml, and .state.json;
• latest closed policy manifest and guideline components;
• cumulative human gold;
• open and recent round manifests, Session progress, checkpoints, policy diffs, audit and challenge metrics, coverage, and risk ledger;
• sealed-test status and access log metadata without reading protected ids or labels;
• evaluation registry, validity, and scorecards;
• production manifest, terminal dispositions, risk queue, and final audit;
• migration receipts and implementation holds.

Output

Lead with:

project {id} · state {state} · round {round/phase}
policy {G_t/G*} · human gold {n} · unresolved {n}
quality {gate} · stability {gate/streak} · coverage {gate} · risk {gate}
sealed test {reserved/frozen/released/valid/invalid}
next {single valid action}

Then show only relevant detail:

• H/L/N and seven-region coverage;
• audit and challenge metrics in separate rows;
• consensus-audit shared errors;
• last material guideline diff and backward-impact result;
• executor qualification and scorecard summary after evaluation;
• terminal/provenance shares and final-audit findings during completion;
• explicit HOLD entries with missing capability and preserved state.

Distinguish not started, pending human, HOLD, failed, invalid, and complete. Never infer success from missing files, model consensus, a legacy kappa field, or a rendered report that conflicts with canonical records.

Recommend exactly one next valid command or human decision based on the state machine.