{% extends "mirrorwall/base.html" %}
{% from "mirrorwall/components.html" import badge, empty_state, json_viewer, table %}
{% block title %}Reliability — LoadCoach{% endblock %}

{# Every statistic is a bounded value: present with its sample count, or an em dash carrying the
   reason it is absent (ADR-0016 rule 5). The dash comes from MirrorWall's `measurement` filter,
   so it is the same dash, with the same tooltip and aria-label, every other page uses. #}
{% macro pct(stat) -%}
{% if stat.value is none %}{{ none | measurement(stat.reason) }}{% else %}{{ "%.0f%%"|format(stat.value * 100) }}{% endif %}
{%- endmacro %}
{% macro ms(stat) -%}
{% if stat.value is none %}{{ none | measurement(stat.reason) }}{% else %}{{ stat.value | measurement(none, "ms") }}{% endif %}
{%- endmacro %}
{% macro num(stat, fmt="%.1f") -%}
{% if stat.value is none %}{{ none | measurement(stat.reason) }}{% else %}{{ fmt|format(stat.value) }}{% endif %}
{%- endmacro %}
{% macro trend(verdict) -%}
{% if verdict.status == "regressed" %}{{ badge("regression", tone="danger", title=verdict.reason) }}
{% elif verdict.status == "stable" %}{{ badge("stable", tone="success", title=verdict.reason) }}
{% else %}{{ badge("not evaluated", tone="neutral", title=verdict.reason) }}{% endif %}
{%- endmacro %}
{% macro breaker(entry) -%}
{% if entry.circuit_state == "open" %}{{ badge("open", tone="danger", title=entry.circuit_reason) }}
{% elif entry.circuit_state == "half_open" %}{{ badge("half-open", tone="warning", title=entry.circuit_reason) }}
{% else %}{{ badge("closed", tone="success", title=entry.circuit_reason) }}{% endif %}
{%- endmacro %}

{% block content %}
<h2>Reliability</h2>
<p class="muted">Production evidence per model and task profile: what actually happened when jobs
ran. Every rate needs {{ minimums.rate_samples }} samples, every percentile
{{ minimums.percentile_samples }}, and the routing factor {{ minimums.factor_attempts }} counted
attempts in the last 7 or 30 days — below that a value is shown as — with the reason, never as a
number. A regression is a drop of at least {{ "%.0f%%"|format(minimums.regression_drop * 100) }}
in validated success against the model's own history, at z ≥ {{ minimums.regression_z }}.</p>

{% if task or model %}
<p class="muted">Filtered to{% if task %} task <span class="mono">{{ task }}</span>{% endif %}{% if model %} model <span class="mono">{{ model }}</span>{% endif %}. <a href="/reliability">Show all</a></p>
{% endif %}

{% if not entries %}
{{ empty_state("No production evidence yet. Statistics appear once jobs have run; feedback from callers (POST /jobs/{id}/feedback) is folded in as it arrives.") }}
{% else %}
{% set columns = [
  {"label": "Model"}, {"label": "Task profile"},
  {"label": "Attempts 7d / 30d / all", "numeric": true},
  {"label": "Factor", "numeric": true},
  {"label": "Answered", "numeric": true}, {"label": "Validated", "numeric": true},
  {"label": "Accepted", "numeric": true}, {"label": "Quality", "numeric": true},
  {"label": "p50", "numeric": true}, {"label": "p95", "numeric": true},
  {"label": "Trend"}, {"label": "Breaker"}
] %}
{% set rows = [] %}
{% for entry in entries %}
{% set week = entry.windows["7d"] %}
{% set _ = rows.append([
  entry.canonical_id,
  entry.task_profile_id,
  week.counted ~ " / " ~ entry.windows["30d"].counted ~ " / " ~ entry.windows["all"].counted,
  ('<span title="' ~ (entry.factor.reason | e) ~ '">' ~ "%.3f"|format(entry.factor.value) ~ (" (neutral)" if entry.factor.neutral else "") ~ '</span>') | safe,
  pct(week.success_rate()), pct(week.validation_pass_rate()),
  pct(week.acceptance()), num(week.quality(), "%.2f"),
  ms(week.p50()), ms(week.p95()),
  trend(entry.regression), breaker(entry)
]) %}
{% endfor %}
{{ table(columns, rows, caption="The last seven days, per model and task profile. Hover a dash for why it is absent; hover a factor for how it was computed.",
         table_id="reliability", sortable=true, complete=true, row_count=entries | length) }}

<h3>Every window</h3>
{% set columns = [
  {"label": "Model"}, {"label": "Task profile"}, {"label": "Window"},
  {"label": "Attempts", "numeric": true}, {"label": "Answered", "numeric": true},
  {"label": "Validated", "numeric": true}, {"label": "Errors", "numeric": true},
  {"label": "Timeouts", "numeric": true}, {"label": "Cancelled", "numeric": true},
  {"label": "Accepted", "numeric": true}, {"label": "Quality", "numeric": true},
  {"label": "p50", "numeric": true}, {"label": "p95", "numeric": true},
  {"label": "Tokens/s", "numeric": true}, {"label": "Output tokens", "numeric": true}
] %}
{% set rows = [] %}
{% for entry in entries %}
{% for name in windows %}
{% set stats = entry.windows[name] %}
{% set _ = rows.append([
  entry.canonical_id, entry.task_profile_id, name,
  stats.counted, pct(stats.success_rate()), pct(stats.validation_pass_rate()),
  stats.errors, stats.timeouts, stats.cancellations,
  pct(stats.acceptance()), num(stats.quality(), "%.2f"),
  ms(stats.p50()), ms(stats.p95()),
  num(stats.tokens_per_second(), "%.1f"), num(stats.output_tokens(), "%.0f")
]) %}
{% endfor %}
{% endfor %}
{{ table(columns, rows, table_id="reliability-windows", complete=true, row_count=rows | length) }}

{% set regressions = entries | selectattr("regression.regressed") | list %}
{% if regressions %}
<h3>Regressions</h3>
<ul>
{% for entry in regressions %}
<li><span class="mono">{{ entry.canonical_id }}</span> · {{ entry.task_profile_id }} — {{ entry.regression.reason }}</li>
{% endfor %}
</ul>
{% endif %}
{% endif %}
{% endblock %}