Feature Usage & Effectiveness

{{ report.total_functions }} features · {{ '{:,}'.format(report.total_calls) }} calls in window · window: {{ report.window }}

By parent

{% for p in report.parents %}

{{ p.name }} · {{ p.function_count }} fn

{{ '{:,}'.format(p.total_calls) }} calls / {{ p.calls_per_day }} per day

success= 90 else 'warn' if p.success_rate_pct >= 70 else 'bad' }}'>{{ p.success_rate_pct }}% errors{{ p.errors }} p50{{ p.typical_p50_ms if p.typical_p50_ms is not none else '-' }}ms {% if p.avg_score is not none %} useful= 0.7 else 'warn' if p.avg_score >= 0.4 else 'bad' }}'>{{ '%.2f'|format(p.avg_score) }} {% endif %}

{% endfor %}

By function

{% for f in report.functions %} {% endfor %}

Parent	Function	Kind	Calls	Calls/day	Success %	Usefulness	Rated %	p50 ms	p95 ms	Avg result	Last used
{{ f.parent }}	{{ f.operation }}	{{ f.kind }}	{{ '{:,}'.format(f.total_calls) }}	{{ f.calls_per_day }}	= 90 else 'warn' if f.success_rate_pct >= 70 else 'bad' }}'>{{ f.success_rate_pct }}%	{% if f.avg_score is not none %} = 0.7 else 'warn' if f.avg_score >= 0.4 else 'bad' }}'>{{ '%.2f'\|format(f.avg_score) }} {% else %}—{% endif %}	{% if f.receipts %}{{ f.rated_pct }}% ({{ f.rated }}/{{ f.receipts }}){% else %}—{% endif %}	{{ f.p50_ms if f.p50_ms is not none else '-' }}	{{ f.p95_ms if f.p95_ms is not none else '-' }}	{{ '{:,}'.format(f.avg_result_size) if f.avg_result_size is not none else '-' }}	{{ f.last_seen[:10] if f.last_seen else '-' }}

Usage: Calls = invocations in window; Calls/day normalises by the active date span. Effectiveness: Success % = calls that did not error; p50/p95 = latency percentiles; Avg result = mean serialized result size (bytes) of successful calls. Usefulness (0-1) and Rated % are self-reported by the agent via rate_recall (useful=1.0 / partial=0.5 / noise=0.0; unused excluded from the mean). This is an agent_inference-grade operational signal only — it never affects memory heat, promotion, or ranking (R8). Click a parent card to filter, click any column header to sort.