By name: sl-status description: "Show current state of a subjective-label project: iteration count, gallery size, panel-internal κ, public-dataset κ trajectory, disagreement category breakdown, suggested next step. Read-only. Use when the researcher says /sl-status or wants to check progress."
Skill: sl-status
Read-only status dashboard.
Protocol
Step 1. Resolve project_dir (from arg or cwd).
Step 2. Read:
- {project_dir}/.state.json
- {project_dir}/gallery/gallery.json
- {project_dir}/gallery/guideline.md (first 30 lines for preview)
- {project_dir}/validation/trajectory.jsonl (if exists)
- {project_dir}/gallery/history/*.diff (last 3 for preview)
- {project_dir}/iterations/iter_*/pool_stats.json (residual trajectory)
- {project_dir}/iterations/iter_*/projection/separation.json (geometric signal)
Step 3. Print status block:
============================================================ Project: {project_dir} Topic: {config.topic} Labels: {config.label_values} Iteration: {state.iteration} Status: {state.status} (initialized / iterating / converged / scaled)
Gallery: {N} entries across {K} label values Guideline: {M} rules, last updated iter {state.last_guideline_update}
Panel κ (latest iter): {value} Public-κ trajectory (last 5 iters): iter 1 GoEmotions κ=0.31 iter 2 GoEmotions κ=0.42 iter 3 GoEmotions κ=0.48 ← ceiling 0.46, CONVERGED
Disagreement profile (latest iter): A boundary: 12 cases B ambiguity: 3 cases C novel: 1 case D noise: 8 cases
Geometric separation (latest iter, n=3 labels): overall silhouette: 0.41 per-label: high (n=18, sil=0.51) low (n=14, sil=0.38) none (n=31, sil=0.12, fragments=2) warnings: - label 'none' shows fragmentation (2 sub-clusters) — possible schema gap - labels {low, none} overlap heavily (sil=0.05) — add a tiebreaker plot: iterations/iter_3/projection/projection.png
Residual trajectory (sample pool the classifier cannot absorb): iter 1 full pool=12000 (no classifier yet) iter 2 residual pool= 4100 (-66%) iter 3 residual pool= 1280 (-69%) iter 4 residual pool= 310 (-76%) ← shrinking-net working
Next step: {contextual recommendation based on state} ============================================================