Used by DashboardHygieneAnalyzer to carry the dashboard title.
omitempty keeps the wire form clean for non-dashboard findings.

Anchored regex against Finding.Dashboard. Empty field never matches.
Wires through config.IgnoreConfig.Dashboard.

PanelTargets walks rows recursively, returns (panel-title, expr)
pairs filtered to Prometheus targets. BaseURL is a defensive copy
used by the dashboard-hygiene analyzer to build absolute dashboard
URLs in Fix snippets.

New analyzer flags Grafana dashboards whose panel queries reference
missing Prometheus metrics. Skeleton + nil-Graf warning path; full
algorithm in subsequent commits.

Walks every Grafana dashboard, parses Prometheus targets via
promqlx, and groups (dashboard, missing-metric) pairs. Severity
Medium per the design. Recording-rule outputs and fix snippet
come in later commits.

…rename

Code-quality follow-ups to the happy-path commit:
- add Dashboard tiebreaker to the comparator so output is
  deterministic across map iterations
- extract the comparator into findingLess to keep Analyze
  under the gocyclo limit (was 15, the extra branch pushed
  it to 16)
- extract sample-cap 5 to a named const for grep-ability
- rename buildFinding parameter to mirror the call-site
  variable name

Same missing metric across multiple panels yields one finding;
distinct missing metrics in the same dashboard emit separate
findings.

A recording rule whose output is not yet in head series must
still be treated as a known metric. Mirrors the resolution flow
in unusedmetrics, including the VictoriaMetrics graceful-degrade
sentinel.

… test

Code-quality follow-ups to the recording-rule resolution commit:
- expand the two BuildInfo-adjacent comments to explain WHY each
  VM-flavor check exists (404 path vs 200-empty-groups path), so a
  future reader doesn't need to cross-reference unusedmetrics
- strengthen the RR test by querying AlertA from a second panel:
  proves the type filter (r.Type == "recording") actually held,
  not just that the recording-rule output was added to exists

Grafana template variables like ${metric}_total sanitise to
__remetric_var___total - a valid PromQL identifier that parses
cleanly and leaks into the extracted metric set. Change isSentinel
to a Contains check so any name containing the sentinel substring
is treated as a sanitiser artifact and filtered.

Without this, dashboardhygiene and unusedmetrics would treat
template-variable expressions as references to bogus metrics
named '__remetric_var__*'.

…and Loki targets

Grafana template-variable queries (${metric}_total) and non-Prometheus
datasources must not generate findings or warnings. The
template-variable path relies on promqlx filtering sentinel-derived
metric names; the Loki path relies on PanelTargets filtering by
datasource type.

Per-dashboard fetch errors degrade to warnings without aborting
the analyzer. Search() failure is fatal. VictoriaMetrics without
--vmalert emits the recording-rules-unavailable warning.

Renders a paste-ready instruction block: restore the metric or
remove the broken queries. Drops the URL line when no absolute
dashboard URL is available. Caps the panel list at 10 entries
with a '... and N more' tail.

… builtin

Code-quality follow-ups to the fix-snippet commit:
- pull the broken-panel docs URL from findings.DocURL(ClassBrokenPanel)
  instead of a hardcoded literal, so the single source of truth in
  internal/findings/ stays authoritative
- replace explicit limit-clamping with the Go 1.21+ min() builtin
- replace C-style index loop with idiomatic range over the slice

New top-level subject 'dashboards' with one action 'broken'.
Requires --prometheus and --grafana. Honors --output, --min-severity,
--ignore-dashboard, --ignore-metric, --fail-on, --limit, --timeout.

…pty.go

Code-quality follow-ups to the dashboards broken subcommand:
- drop the CLI's local re-sort; the analyzer already orders by
  (severity desc, sample-count desc, dashboard asc, metric asc),
  and the filter passes are stable. The local sort was discarding
  the sample-count tiebreaker - meaningful signal that broken-from-
  many-panels metrics rank higher within a severity tier
- move brokenPanelCopy to empty.go for parity with cardinalityCopy
  and labelPatternCopy
- extend TestEmptyCopy_Values to cover brokenPanelCopy and the
  previously-uncovered unusedMetricsCopy

Both flows include the new analyzer; without --grafana it emits
a warning and zero findings (consistent with unusedmetrics).

…l page

Real content for the broken-panel finding class. Updates the
catalog, the unused-metric cross-link, the mkdocs nav, and the
'What's still missing in v0.1' README section since the analyzer
now ships in v0.1.

Provisions a Grafana dashboard whose only panel queries a metric
Prometheus does not scrape; runs 'remetric dashboards broken' and
asserts the finding is emitted with class=broken-panel.

…text

Follow-up to the analyzer landing. Updates user-facing copy that
still listed the v0.1 analyzer set as four (now five with
broken-panel) and omitted --ignore-dashboard from the ignore-*
table.