Evidence Score is a deterministic marker-ranking signal, not a calibrated probability. Review Level is a rule-based Critic category. Abstention (Unknown) represents conservative fail-closed safety, not a biological cell identity. Statistical claims and OOD novelty require standalone benchmark artifacts.
Observability-only mode: {{ load_error }}
Checkpoint / ETA / host metrics are available under 运行观测. This panel does not write predictions.
CPU%: {{ cpu.get('cpu_percent', 'n/a') }} · cores {{ cpu.get('cpu_count_logical', 'n/a') }}
Platform: {{ cpu.get('platform', 'n/a') }}
{% if gpu.get('available') %}GPU source: {{ gpu.get('source') }}
No GPU reported (nvidia-smi / torch.cuda unavailable).
{% endif %}Source: {{ stale.get('source', 'n/a') }}
Review state: {{ stale.get('review_state', 'n/a') }}
Reason: {{ stale.get('stale_reason', 'n/a') }}
Run root: {{ observability.run_root or '' }}
| Method | Fold | Status | Elapsed / Duration (s) | Failure reason | Prediction SHA-256 |
|---|---|---|---|---|---|
| {{ row.method }} | {{ row.fold_id }} |
{{ row.status }} | {% if row.elapsed_seconds is not none %}{{ '%.1f'|format(row.elapsed_seconds) }}{% else %}—{% endif %} / {% if row.duration_seconds is not none %}{{ '%.1f'|format(row.duration_seconds) }}{% else %}—{% endif %} | {{ row.failure_reason or row.error or '—' }} | {{ (row.prediction_sha256 or '—')[:16] }}{% if row.prediction_sha256 %}…{% endif %} |
| No checkpoint status files found under this run root. | |||||
| Name | Present | Bytes | SHA-256 |
|---|---|---|---|
{{ p.name }} |
{{ 'yes' if p.present else 'no' }} | {{ p.byte_size if p.byte_size is not none else '—' }} | {{ p.sha256 or p.hash_skipped or p.error or '—' }} |
| No product inventory. | |||
| Cluster | Baseline Identity | CL ID | Evidence Score | Review Level | Novelty / OOD | Critic Flags | Cells | Review Status | Actions |
|---|---|---|---|---|---|---|---|---|---|
| {{ row.cluster }} | {{ row.cell_type }} {% if row.cluster in diff_summary.diff_rows and diff_summary.diff_rows[loop.index0].status == 'pending_change' %} Override Pending {% endif %} | {{ row.cl_id }} |
{{ "%.3f"|format(row.score) }} | {{ row.confidence }} | {{ row.novelty_decision }} ({{ "%.3f"|format(row.novelty_score) }}) | {{ row.flags }} | {{ row.n_cells }} |
| Cluster | Original Identity | Proposed / Pending Identity | Affected Cells | Impact Status | Rationale / Reason |
|---|---|---|---|---|---|
| {{ row.cluster }} | {{ row.original_type }} | {% if row.status == 'pending_change' %} {{ row.pending_type }} {% else %} {{ row.pending_type }} {% endif %} | {{ row.n_cells }} | {% if row.status == 'pending_change' %} MODIFIED {% else %} UNCHANGED {% endif %} | {{ row.reason }} |
Manual overrides append to annotation_audit_log.jsonl. After Apply,
derived report/evidence/figures/manifest are stale until you
regenerate and re-sign. Packs remain data-only and still pass critic/abstention gates.
Reviewer: {{ signoff.reviewer_name }} ({{ signoff.reviewer_role }})
Signed At: {{ signoff.timestamp }}
Notes: {{ signoff.notes }}
No formal sign-off certificate recorded yet.
{% endif %}Download a self-contained, machine-readable JSON bundle containing dataset metadata, cluster state histories, checklist completion snapshot, override diffs, audit trails, and signed certificate.
📥 Export Review Packet (review_packet.json)