monitoring: systemd services, domain expiry, DNS correctness, public endpoints

Four more check types, 42 endpoints, taking the estate from 39 to 81.

── systemd services (infra/401) ────────────────────────────────────────────
Every unit we deploy, checked every 5 minutes with a 16-minute heartbeat.

This closes the gap that let a real bug run unnoticed earlier today: a backup
script left forgejo, lnbits, headscale and memos stopped, and NOTHING caught
it. The dumps exited 0, the artefacts were correct, the deploy said failed=0,
and liveness only proves the HOST is up - not that anything on it serves.

One endpoint PER UNIT but only ONE timer per host: the check iterates that
host's units and pushes a result for each, the way check-backups.sh reports per
source. Four units on vipy would otherwise mean four scripts, services and
timers. A single host-level red light would also say "something on vipy is
down" without saying which, which is not the question anyone has.

Keys are host-qualified because unit names collide - caddy runs on four
machines. Which units a host runs lives in host_vars/<host>/monitored_services,
because "what runs here" is a property of the machine, the same reasoning as
the cross-host ports.

nut-driver-enumerator is deliberately excluded: it is a oneshot generator that
is `enabled` but always `inactive`, so it would report down forever. Checked
live before excluding it.

── domain, DNS and public endpoints (infra/402) ────────────────────────────
The first checks that PULL rather than push, and that is the right way round:
all three are about how the outside world sees us, so they must be measured
from outside. Nothing is installed anywhere - no script, no timer, no token.
They also have no heartbeat, because a heartbeat answers "did the reporter
report"; when Gatus does the checking itself, failure is immediate.

  domain   1 endpoint,  24h, [DOMAIN_EXPIRATION] > 336h (two weeks)
  dns      11 endpoints, 24h, [DNS_RCODE] == NOERROR and [BODY] == the IP
  public   14 endpoints, 5m,  11 HTTPS + 3 TCP

Expected A records are derived from inventory (hostvars[host].ansible_host),
not written down again. The estate's recurring bug is an address recorded in a
second place and left behind when the machine moved; asserting against
inventory means a renumbered box is one edit, not two.

Expected HTTP status was checked live per site rather than assumed. Two return
401 - the Gatus dashboard and the DATUM dashboard, both behind basic auth - and
that is what is asserted: expecting 200 there would go green precisely when the
auth broke. headscale asserts /health rather than /, which is a 404 by design.
Every HTTPS check also carries [CERTIFICATE_EXPIRATION] > 168h, which is free
on an endpoint already being polled and catches a renewal that silently stops.

Two things learned the hard way, both now in comments:

  * A domain-expiry endpoint needs a URL SCHEME. Gatus derives the endpoint type
    from the prefix (endpoint.Type()), so a bare "contrapeso.xyz" is UNKNOWN and
    the whole config is rejected. It is https:// plus a DOMAIN_EXPIRATION
    condition and no status assertion, so the registrar's parking page at the
    apex is irrelevant. Upstream also enforces a 5m minimum interval for that
    placeholder, because it uses a free whois service.

  * That rejection proved skip-invalid-config-update was worth adding. Gatus
    logged "the configuration file was updated, but it is not valid, the old
    configuration will continue being used" and kept running. Without it the
    reload path calls panic() and one malformed contributed file takes the
    monitor down.

Verified: 15/15 service endpoints UP; 26/27 public-facing UP. The single DOWN is
public_memos, correctly - memos-box is powered off, and the condition result
reads [STATUS] (502) == 200. memos-box and arbret-staging-box were shut down
deliberately from the Proxmox UI to test the liveness endpoints; their endpoints
stay registered and will go green when the VMs return.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

ansible/host_vars/forgejo_runner_local/main.yml

@@ -0,0 +1,12 @@
+---
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - forgejo-runner

ansible/host_vars/fulcrum_box_local/main.yml

@@ -4,3 +4,14 @@
 # which publishes the port. See host_vars/knots_box_local/main.yml for why this
 # lives in host_vars rather than in the role's defaults.
 fulcrum_ssl_port: 50002
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - fulcrum

ansible/host_vars/knots_box_local/main.yml

@@ -12,3 +12,15 @@
 bitcoin_p2p_port: 8333
 datum_gateway_api_port: 7152
 datum_gateway_stratum_port: 23334
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - bitcoind
+  - datum-gateway

ansible/host_vars/memos_box_local/main.yml

@@ -0,0 +1,12 @@
+---
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - memos

ansible/host_vars/monitoring/main.yml

@@ -0,0 +1,12 @@
+---
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - caddy

ansible/host_vars/nodito/main.yml

@@ -31,3 +31,15 @@ ups_port: auto
 ups_user: counterweight
 ups_offdelay: 120  # Seconds after shutdown before UPS cuts outlet power
 ups_ondelay: 30    # Seconds after mains returns before UPS restores outlet power
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - nut-server
+  - nut-monitor

ansible/host_vars/spacey/main.yml

@@ -0,0 +1,13 @@
+---
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - headscale
+  - caddy

ansible/host_vars/vipy/main.yml

@@ -0,0 +1,15 @@
+---
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - forgejo
+  - lnbits
+  - caddy
+  - phoenixd

ansible/host_vars/watchtower/main.yml

@@ -0,0 +1,13 @@
+---
+
+# Systemd services deployed on this host, monitored every 5 minutes.
+#
+# The fact lives with the machine rather than in a central map, for the same
+# reason the cross-host ports do: "what runs here" is a property of the host,
+# and a central list is one more thing to forget to update when a service moves.
+#
+# Only units WE deploy belong here. Distro units (ssh, cron) have their own
+# supervision and would be noise.
+monitored_services:
+  - caddy
+  - ntfy

ansible/infra/401_service_monitoring.yml

@@ -0,0 +1,83 @@
+---
+# Is each systemd-deployed service actually running?
+#
+# Every 5 minutes, with a 16-minute Gatus heartbeat - three missed runs before
+# it alarms, so a reboot or a slow check does not page anyone, but a host that
+# stops reporting does.
+#
+# This closes the gap that let a real bug run unnoticed: a backup script left
+# forgejo, lnbits, headscale and memos stopped, and NOTHING caught it. The dumps
+# exited 0, the artefacts were correct, the deploy said failed=0, and liveness
+# only proves the HOST is up - not that anything on it is serving.
+#
+# One endpoint PER UNIT, not per host. A host running four services needs four
+# endpoints, or a single red light says "something on vipy is down" without
+# saying which - and that is the question you actually have at 3am. But only ONE
+# timer per host: the check iterates that host's units and pushes a result for
+# each, the same way check-backups.sh reports per source. Four units on vipy
+# would otherwise mean four scripts, four services and four timers.
+#
+# Which units each host runs is in host_vars/<host>/main.yml as
+# monitored_services, because "what runs here" is a property of the machine.
+#
+# Keys are host-qualified because unit names collide - caddy runs on four
+# machines. Gatus computes sanitize(group)_sanitize(name), so group "services"
+# and name "vipy/caddy" give services_vipy-caddy.
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Register one endpoint per unit. Runs first: Gatus reloads within 30s, and the
+# host play above takes minutes, so every endpoint exists before its first push.
+# ─────────────────────────────────────────────────────────────────────────────
+- name: Register the service checks with Gatus
+  hosts: observability
+  become: yes
+
+  tasks:
+    # Two plain steps rather than one clever expression: first collect which
+    # units each host declares, then flatten that into endpoints.
+    - name: Collect the units each host declares
+      ansible.builtin.set_fact:
+        host_units: "{{ host_units | default([]) + [{'host': item, 'units': hostvars[item].monitored_services}] }}"
+      loop: "{{ groups['managed'] | sort }}"
+      when: hostvars[item].monitored_services | default([]) | length > 0
+
+    - name: Build one endpoint per unit
+      ansible.builtin.set_fact:
+        service_endpoints: "{{ service_endpoints | default([]) + [{
+          'name': (item.0.host | lower | regex_replace('[/_.,# +&]', '-')) ~ '/' ~ item.1,
+          'group': 'services',
+          'token': gatus_push_tokens[item.0.host],
+          'heartbeat': '16m'}] }}"
+      loop: "{{ host_units | subelements('units') }}"
+
+    - name: Register the service endpoints
+      ansible.builtin.include_role:
+        name: gatus_endpoint
+      vars:
+        gatus_endpoint_name: services
+        gatus_endpoint_external: "{{ service_endpoints }}"
+
+- name: Monitor systemd services on every host that has them
+  hosts: managed
+  become: yes
+  vars:
+    gatus_api: "https://{{ subdomains.gatus }}.{{ root_domain }}/api/v1/endpoints"
+    host_key: "{{ inventory_hostname | lower | regex_replace('[/_.,# +&]', '-') }}"
+
+  tasks:
+    - name: Is every deployed service running?
+      ansible.builtin.include_role:
+        name: healthcheck
+      vars:
+        healthcheck_name: service-health
+        healthcheck_description: "systemd services on {{ inventory_hostname }}"
+        healthcheck_check: systemd-units
+        healthcheck_units: "{{ monitored_services }}"
+        healthcheck_units_key_prefix: "services_{{ host_key }}"
+        healthcheck_interval: "5min"
+        healthcheck_boot_delay: "2min"
+        # The per-unit results go to keys under this collection; the role's own
+        # single-result push is unused here, so only the base is set.
+        healthcheck_push_base: "{{ gatus_api }}"
+        healthcheck_push_token: "{{ gatus_push_tokens[inventory_hostname] }}"
+      when: monitored_services | default([]) | length > 0

ansible/infra/402_public_monitoring.yml

@@ -0,0 +1,125 @@
+---
+# Domain expiry, DNS correctness, and public endpoint reachability.
+#
+# These are the first checks in the estate that PULL rather than push, and that
+# is the right way round for them: all three are about how the outside world
+# sees us, so they must be measured from outside. Gatus polls from the
+# observability host and needs nothing installed anywhere else - there is no
+# script, no timer and no token, because nothing is reporting in.
+#
+# That also means these have no heartbeat. A heartbeat answers "did the thing
+# that was supposed to report in do so"; when Gatus does the checking itself,
+# failure is immediate and self-evident.
+
+- name: Register the public-facing checks with Gatus
+  hosts: observability
+  become: yes
+
+  vars:
+    # Expected A records, derived from inventory rather than written down again.
+    # The estate's recurring bug is an address recorded in a second place and
+    # then left behind when the machine moved, so the check asserts against
+    # ansible_host - if a box is renumbered, inventory is the one edit.
+    dns_records:
+      - {sub: "{{ subdomains.gatus }}",              host: monitoring}
+      - {sub: "{{ subdomains.ntfy }}",               host: watchtower}
+      - {sub: "{{ subdomains.headscale }}",          host: spacey}
+      - {sub: "{{ subdomains.vaultwarden }}",        host: vipy}
+      - {sub: "{{ subdomains.forgejo }}",            host: vipy}
+      - {sub: "{{ subdomains.lnbits }}",             host: vipy}
+      - {sub: "{{ subdomains.ntfy_emergency_app }}", host: vipy}
+      - {sub: "{{ subdomains.personal_blog }}",      host: vipy}
+      - {sub: "{{ subdomains.memos }}",              host: vipy}
+      - {sub: "{{ subdomains.mempool }}",            host: vipy}
+      - {sub: "{{ subdomains.datum_gateway }}",      host: vipy}
+
+    # A public resolver on purpose: this must test what the internet sees, not
+    # what a local cache or the tailnet's MagicDNS happens to answer.
+    dns_resolver: "1.1.1.1"
+
+    # Expected status per site, checked live before being written down.
+    # 401 is the CORRECT answer for the two behind basic auth - asserting 200
+    # there would go green precisely when the auth broke.
+    public_sites:
+      - {name: gatus,       sub: "{{ subdomains.gatus }}",              path: "/",       status: 401}
+      - {name: ntfy,        sub: "{{ subdomains.ntfy }}",               path: "/",       status: 200}
+      - {name: headscale,   sub: "{{ subdomains.headscale }}",          path: "/health", status: 200}
+      - {name: vaultwarden, sub: "{{ subdomains.vaultwarden }}",        path: "/",       status: 200}
+      - {name: forgejo,     sub: "{{ subdomains.forgejo }}",            path: "/",       status: 200}
+      - {name: lnbits,      sub: "{{ subdomains.lnbits }}",             path: "/",       status: 200}
+      - {name: avisame,     sub: "{{ subdomains.ntfy_emergency_app }}", path: "/",       status: 200}
+      - {name: blog,        sub: "{{ subdomains.personal_blog }}",      path: "/",       status: 200}
+      - {name: memos,       sub: "{{ subdomains.memos }}",              path: "/",       status: 200}
+      - {name: mempool,     sub: "{{ subdomains.mempool }}",            path: "/",       status: 200}
+      - {name: datum,       sub: "{{ subdomains.datum_gateway }}",      path: "/",       status: 401}
+
+    # Ports published from the edge host by socket_proxy.
+    public_tcp:
+      - {name: bitcoin-p2p,    host: vipy, port: "{{ hostvars['knots_box_local'].bitcoin_p2p_port }}"}
+      - {name: fulcrum-ssl,    host: vipy, port: "{{ hostvars['fulcrum_box_local'].fulcrum_ssl_port }}"}
+      - {name: datum-stratum,  host: vipy, port: "{{ hostvars['knots_box_local'].datum_gateway_stratum_port }}"}
+
+  tasks:
+    # ── Domain expiry ────────────────────────────────────────────────────────
+    - name: Build the domain endpoint
+      ansible.builtin.set_fact:
+        domain_endpoints:
+          - name: "{{ root_domain }}"
+            group: domain
+            # Needs a scheme: Gatus derives the endpoint TYPE from the URL prefix
+            # (endpoint.Type()), and a bare domain is UNKNOWN and rejected. The
+            # apex points at the registrar's parking page, which is irrelevant -
+            # the only condition here is the WHOIS expiry, and no status check is
+            # asserted, so what the page serves does not matter.
+            url: "https://{{ root_domain }}"
+            # 24h, and upstream enforces a 5m minimum for DOMAIN_EXPIRATION
+            # anyway because it uses a free whois service that must not be
+            # hammered and whose data updates slowly.
+            interval: 24h
+            # 336h = 14 days. Renewal is manual at the registrar, so this needs
+            # enough runway to act on.
+            conditions:
+              - "[DOMAIN_EXPIRATION] > 336h"
+
+    # ── DNS ──────────────────────────────────────────────────────────────────
+    - name: Build the DNS endpoints
+      ansible.builtin.set_fact:
+        dns_endpoints: "{{ dns_endpoints | default([]) + [{
+          'name': item.sub ~ '.' ~ root_domain,
+          'group': 'dns',
+          'url': dns_resolver,
+          'interval': '24h',
+          'dns': {'query-type': 'A', 'query-name': item.sub ~ '.' ~ root_domain},
+          'conditions': ['[DNS_RCODE] == NOERROR',
+                         '[BODY] == ' ~ hostvars[item.host].ansible_host]}] }}"
+      loop: "{{ dns_records }}"
+
+    # ── Public HTTP ──────────────────────────────────────────────────────────
+    - name: Build the public HTTP endpoints
+      ansible.builtin.set_fact:
+        http_endpoints: "{{ http_endpoints | default([]) + [{
+          'name': item.name,
+          'group': 'public',
+          'url': 'https://' ~ item.sub ~ '.' ~ root_domain ~ item.path,
+          'interval': '5m',
+          'conditions': ['[STATUS] == ' ~ item.status,
+                         '[CERTIFICATE_EXPIRATION] > 168h']}] }}"
+      loop: "{{ public_sites }}"
+
+    # ── Public TCP ───────────────────────────────────────────────────────────
+    - name: Build the public TCP endpoints
+      ansible.builtin.set_fact:
+        tcp_endpoints: "{{ tcp_endpoints | default([]) + [{
+          'name': item.name,
+          'group': 'public',
+          'url': 'tcp://' ~ hostvars[item.host].ansible_host ~ ':' ~ item.port,
+          'interval': '5m',
+          'conditions': ['[CONNECTED] == true']}] }}"
+      loop: "{{ public_tcp }}"
+
+    - name: Register the public-facing endpoints
+      ansible.builtin.include_role:
+        name: gatus_endpoint
+      vars:
+        gatus_endpoint_name: public
+        gatus_endpoint_pulled: "{{ domain_endpoints + dns_endpoints + http_endpoints + tcp_endpoints }}"

ansible/roles/gatus_endpoint/defaults/main.yml

@@ -9,6 +9,11 @@ gatus_endpoint_name: ""
 
 # PULLED endpoints - Gatus makes the request and evaluates conditions.
 #   - {name, group, url, interval, conditions: [...], alerts: [...]}
+#
+# A DNS check adds `dns: {query-type, query-name}` - and note that for those,
+# `url` is the RESOLVER to ask, not the name being looked up.
+# A domain-expiry check is just `url: <domain>` with a [DOMAIN_EXPIRATION]
+# condition; it uses WHOIS/RDAP and needs no scheme.
 gatus_endpoint_pulled: []
 
 # EXTERNAL endpoints - the host pushes its own result. Gatus never reaches out,

ansible/roles/gatus_endpoint/templates/endpoints.yaml.j2

@@ -28,6 +28,18 @@ endpoints:
     group: {{ e.group }}
     url: "{{ e.url }}"
     interval: {{ e.interval | default('60s') }}
+{% if e.dns is defined %}
+    # A DNS endpoint: `url` is the RESOLVER to ask, not the thing being asked
+    # about. The name being queried lives in query-name, and [BODY] holds the
+    # resolved record for the conditions below.
+    dns:
+      query-type: {{ e.dns['query-type'] }}
+      query-name: {{ e.dns['query-name'] }}
+{% endif %}
+{% if e.client is defined %}
+    client:
+{{ e.client | to_nice_yaml(indent=2) | indent(6, true) }}
+{% endif %}
     conditions:
 {% for c in e.conditions %}
       - "{{ c }}"

ansible/roles/healthcheck/defaults/main.yml

@@ -28,6 +28,17 @@ healthcheck_boot_delay: "2min"
 healthcheck_push_url: ""
 healthcheck_push_token: ""
 
+# Some checks report MORE THAN ONE result - a host with four systemd services
+# needs four endpoints, or a single red light cannot tell you which one died.
+# Those set healthcheck_push_base to the endpoints COLLECTION and the check body
+# appends each key itself, the same way check-backups.sh reports per source.
+healthcheck_push_base: ""
+
+# Units for the systemd-units check. Each becomes its own Gatus endpoint.
+healthcheck_units: []
+# Prefix for the per-unit endpoint keys, e.g. "services_vipy" -> services_vipy-caddy.
+healthcheck_units_key_prefix: ""
+
 healthcheck_script_dir: /usr/local/bin
 healthcheck_log_dir: /var/log/healthchecks
 

ansible/roles/healthcheck/templates/checks/systemd-units.sh.j2

@@ -0,0 +1,33 @@
+    # One result PER UNIT, not one for the host. A host running four services
+    # needs four endpoints: a single red light would tell you "something on vipy
+    # is down" without saying which, and that is the question you actually have.
+    #
+    # Keys are host-qualified because unit names collide - caddy runs on four
+    # machines. Gatus builds the key as sanitize(group)_sanitize(name), so
+    # group "services" + name "vipy/caddy" becomes services_vipy-caddy.
+    local prefix="{{ healthcheck_units_key_prefix }}"
+    local down=""
+
+    for unit in {{ healthcheck_units | join(' ') }}; do
+        local state substate
+        state=$(systemctl is-active "$unit" 2>/dev/null || true)
+        substate=$(systemctl show -p SubState --value "$unit" 2>/dev/null || true)
+
+        if [ "$state" = "active" ]; then
+            report_key "${prefix}-${unit}" "true" "${unit} active (${substate:-running})"
+        else
+            # `failed` and `inactive` are different stories: one crashed, one was
+            # stopped. Both are down, but the message should say which.
+            local detail="${unit} is ${state:-unknown}"
+            [ "$state" = "failed" ] && detail="${detail} (${substate:-failed})"
+            report_key "${prefix}-${unit}" "false" "$detail"
+            down="${down}${down:+, }${unit}=${state:-unknown}"
+        fi
+    done
+
+    if [ -n "$down" ]; then
+        MESSAGE="down: ${down}"
+        return 1
+    fi
+    MESSAGE="all {{ healthcheck_units | length }} units active"
+    return 0

ansible/roles/healthcheck/templates/healthcheck.service.j2

@@ -9,6 +9,7 @@ User=root
 ExecStart={{ healthcheck_script_dir }}/{{ healthcheck_name }}-healthcheck.sh
 Environment=HEALTHCHECK_PUSH_URL={{ healthcheck_push_url }}
 Environment=HEALTHCHECK_PUSH_TOKEN={{ healthcheck_push_token }}
+Environment=HEALTHCHECK_PUSH_BASE={{ healthcheck_push_base }}
 StandardOutput=journal
 StandardError=journal
 

ansible/roles/healthcheck/templates/healthcheck.sh.j2

@@ -11,6 +11,8 @@ set -uo pipefail
 LOG_FILE="{{ healthcheck_log_dir }}/{{ healthcheck_name }}.log"
 PUSH_URL="${HEALTHCHECK_PUSH_URL:-}"
 PUSH_TOKEN="${HEALTHCHECK_PUSH_TOKEN:-}"
+# Set only by checks that report several results; see healthcheck_push_base.
+PUSH_BASE="${HEALTHCHECK_PUSH_BASE:-}"
 
 log() { echo "$(date '+%Y-%m-%d %H:%M:%S') - $*" >> "$LOG_FILE"; }
 
@@ -40,6 +42,18 @@ report() {
     fi
 }
 
+# Report to an arbitrary endpoint key under PUSH_BASE. Used by checks that
+# produce one result per item rather than a single verdict.
+report_key() {
+    local key="$1" success="$2" message="$3"
+    [ -n "$PUSH_BASE" ] || return 0
+    local encoded
+    encoded=$(printf '%s' "$message" | sed 's/%/%25/g; s/ /%20/g; s/&/%26/g; s/+/%2B/g; s/#/%23/g')
+    curl -s -o /dev/null --max-time 15 --retry 2 --retry-delay 3 -X POST \
+        -H "Authorization: Bearer ${PUSH_TOKEN}" \
+        "${PUSH_BASE}/${key}/external?success=${success}&error=${encoded}" 2>/dev/null || true
+}
+
 # ── the check ────────────────────────────────────────────────────────────────
 #
 # The blank line before the closing brace below is load-bearing. Jinja strips an