Caller calibration: {{ phase }}

Objective: lexicographic-safety-v1. Reject safety failures; minimize false calls and wrong tier-A identities; maximize exact recovery, then sensitivity; prefer fewer free parameters, then stable candidate identity.

Protocol: {{ protocol }}
Evidence: {{ evidence }}
Eligible roster: {{ roster }}

{% if selected_id %}

Selected/frozen candidate: {{ selected_id }}

{% else %}

No candidate selected.

{% endif %}

No-calls remain in sensitivity and specificity denominators and are listed separately from negative calls. Consequently specificity plus FPR can be less than one. Precision depends on the study prevalence. Exact recovery requires a known complete variant identity set. Counts use one frozen representative per independent group.

These are descriptive research results. Rendering a report does not authorize model promotion. Mandatory strata may overlap; their group counts are not added to the pooled count.

{% for entry in entries %}

{{ entry.label }}: {{ entry.id }}

{% if entry.reasons %}

Gate reasons

{% endif %} {% for population in entry.populations %}

{{ population.name }}

{% if population.rows %} {% for label, value in population.rows %}{% endfor %}
MetricObserved value and uncertainty
{{ label }}{{ value }}
{% else %}

insufficient-evidence: no eligible groups in this mandatory stratum.

{% endif %} {% endfor %}
{% endfor %}

Cutoff operating points

{% if plots %}

Only measured scalar cutoffs are connected. No extrapolated endpoints or AUC are inferred. No-calls use the same eligible denominator at every cutoff.

{% for plot in plots %}

{{ plot.kind }}

{% for point in plot.points %}{{ point.label }}{% endfor %} 011 {{ plot.x_label }} {{ plot.y_label }}
Fixed policy {{ plot.policy }}; no-calls {{ plot.no_calls }}/{{ plot.eligible }}.
{% endfor %}
{% else %}

No cutoff-search curve in this report. Fixed or multidimensional policies are compared in the tables above.

{% endif %}