Skip to content

Public site

The WebService can serve a second, player-facing site beside the admin Command Center: a read-only statistics site for the community, optionally one view per hosting group. It shares the process, the port and the database with the console, but not its login — every page of it is public and answers without a session, a role or a capability. Visitors get the community's numbers, a leaderboard, a search over the pilots and a profile per pilot; nobody can change anything from it.

The site is switched on with frontend: true plus a sites: block in config/site.yaml — the file the console's Public site page owns. Without a served root site the console keeps the site root and nothing about the public site is served at /.

What it serves

Every address below exists in this build. <slug> is the site's own path segment — empty for the site named root, which is served at / when enabled.

Address What it is
/ or /<slug> The site's overview: hero figures, a live band of what is flying now, the server board, the weekly activity chart, the top pilots and — when CreditSystem has rows — the squadron standings
<site>/rankings The leaderboard: pilots ranked by one chosen metric, filterable by server and period, every column a sort
<site>/players Pilot search: a name search with paging, one row per pilot, each linking to the profile
<site>/player/<ucid> One pilot's profile: totals, airframes, the servers they flew, and their other UCIDs if one Discord account links several
<site>/me The signed-in visitor's own statistics — the same figures the profile already shows, reached from the shell's own chip
<site>/search.json The search box's suggestions, as JSON (see Read the numbers from the API)
<site>/language The language switch (see Carry the language)
<site>/theme.css The site's palette sheet, when one of theme.preset, theme.sheet or theme.expert is declared
<site>/logo.png The site's own logo, when branding.logo declares a raster image

Two addresses do not exist, and the site is deliberate about both:

  • <site>/console is not a site page: the Command Center is at /console while a site owns the root, and a visitor the console would refuse gets its refusal, not a public page.
  • squadrons is a nav id with no page in this build: the id is refused nothing at configuration time, but it renders no menu entry and <site>/squads answers 404.

Turning it on

Two ways in, and both write the same file:

  • In the console: Settings → Public site → Add a site. Pick what it shows (every managed_by token servers.yaml names, or The site root), type the name it carries in the file and the segment it answers on, and press Add the site. The console appends the entry to config/site.yaml and opens that site's card.
  • By hand: config/site.yaml is a small document holding one sites: block. This is complete and minimal:
# config/site.yaml
sites:
  root: {}

root: {} is the smallest valid block: the site with no group token, served at /, showing the servers that declare no managed_by, every page id in the menu and the shipped look. root: with nothing under it is YAML null and is refused (sites.root: must be a map (the site's own block)) — write the braces.

Both routes need frontend: true in config/services/webservice.yaml; that switch installs the web UI, and the site block decides what it serves.

The setup tour does not create a site

The tour ends with your servers configured. A public site is added afterwards — in the console, or by writing the file above.

The block's older home

An installation that has never used the console's page may still carry sites: under DEFAULT: in config/services/webservice.yaml. That block is read only while config/site.yaml does not exist; the moment the file exists it is the only source, and the console warns when both carry a block. The Public site page shows the legacy entries, and its first save writes them to config/site.yaml — so opening the page once and saving is the migration.

An enabled root site moves the console

The Command Center moves to /console only while an ENABLED site is SERVED at the site root — a root: block whose enabled key is absent or true. The console keeps / in every other case:

  • no sites: block at all;
  • an empty block;
  • a root: block with enabled: false — a switched-off site registers no route, so it owns no address;
  • a block that declares only group sites (their pages live at /<slug>, never at /).

With no enabled root site, / answers the console: an anonymous visitor is sent to the login page, and a bookmark to /console answers 404.

The sign-in page keeps its own address (/auth/login), and the installation's fixed surface never moves either: /login, /logout, /auth/*, /language, /theme, /i18n/* and /static.

enabled is a STRUCTURAL key: it decides which routes exist, so it takes effect through the console's apply rather than through a page load. A save on the Public site page offers that apply, and the apply rebuilds the site table in place, so a site switched off or on takes effect at once — no bot restart. See Applying the change.

The Sites page

Settings → Public site is the console page that owns config/site.yaml. Every site is one card, and the card's header leads with the address a visitor types — / for the root site, /guards for a site whose slug is guards — beside the name the file gives it.

  • The box is the site itself, drawn by the console: it follows the form before you save, so a palette, a mark, a section or a sentence can be judged without writing the file. A site added this session draws there too, although its own address only exists after the console's apply.
  • The fields are the keys below. enabled and slug are structural: a save that changes either offers the apply that rebuilds the site table in place — no bot restart (see Applying the change).
  • The mark: Upload an image stores one picture at four sizes and wins over everything; Remove falls back to branding.logo; with neither set the site draws the shipped product mark. See Where a site's logo is read from.
  • Expert mode, at the foot of the page, is this file in an editor with a live check: a document that does not parse, a key this release does not know (named with its path), a site the loader would refuse, and an empty document are all refused, and nothing is written on a refusal.
  • Add a site, beside it, appends one entry instead of replacing the file. Both blocks are the installation Admin's.

A managing scope may edit the content of its own site — the copy, the palette, the mark — while the structure of the installation's sites (adding one, changing an address, the file itself) stays the Admin's.

The keys of a site block

Key Type Default Meaning
site name map key — The block's own name. root is reserved for the site without a group token. Any other name needs a group_role.
slug string the site's name The URL segment the site is served under. The root site must leave it empty. Otherwise [a-z0-9][a-z0-9-]*, unique across sites, and not one of the shell's own paths (login, logout, console, static, auth).
group_role string none The managed_by token this site shows, exactly as servers.yaml writes it. Required for every site except root, which takes none.
enabled boolean true Whether the site is served at all. false keeps the block but registers no route, so the site owns no address — a switched-off root site does not move the console (see the warning under Turning it on).
branding map none Title, logo and footer.
theme map none Palette overrides.
nav list of ids [overview, leaderboard, squadrons, players] The menu entries, in order. Known ids: overview, leaderboard, squadrons, players. A single string is refused (it would iterate as characters).
tiles list of ids [active_week, online_now, playtime_total, avg_session, peak_hour] Known ids: active_week, online_now, playtime_total, avg_session, peak_hour. See the note below — this key is validated but nothing renders from it yet.
lang string en The language the site renders in when a visitor has chosen none. Must be a language that has a webui catalog, otherwise the block is refused.
character string pvpve The kind of community the site is: exactly pvp, pve or pvpve. It decides which figures a page leads with — never which figures exist. See the community's character.

A typo is never silent: an unknown key, a non-list nav or tiles, an unknown id, an unknown character, a duplicate or non-URL-safe slug, a theme block that names two palettes, or a group_role no server names all raise at startup and stop the web UI from installing. The message names the offending key.

tiles changes nothing yet

tiles: is parsed, validated against the five ids and carried on the site, but no page reads it in this build: the overview's readout strip is a fixed set of six figures. Setting tiles: is accepted and has no visible effect. Removing a readout from the overview is not configurable.

Which servers a site covers

Membership is not declared in the site block. servers.yaml is the only authority:

Site Shows
a site with a group_role every server whose managed_by names that token
the site named root every server that declares no managed_by at all
a server naming two tokens belongs to both of those sites, and to no other
# config/servers.yaml
Group-1:
  port: 10308
  managed_by:
    - 5th Guards
# config/services/webservice.yaml
DEFAULT:
  frontend: true
  sites:
    root: {}                                      # servers with no managed_by
    guards:
      slug: guards
      group_role: "5th Guards"                    # served at /guards

Two consequences worth knowing before a report looks like a defect:

  • On an installation where every server is tagged, root is empty. "Only one of my two servers shows" is usually the rule working.
  • A managed_by written as a scalar (managed_by: 5th Guards) is read as a one-element list, with one warning naming the server. Write it as a list.

A site with no servers

A site whose server set is empty is not refused. It is served normally: the membership is read from the live server registry at request time, so a site whose servers are not running, or whose token currently matches nothing, renders with an empty board — a 200 page whose figures are zero and whose lists are empty.

The one case that is reported is a missing registry: a process that has no ServiceBus logs

Sites: this process has no ServiceBus (...). Every public site shows no servers until the
service bus is available in this process.

once per process. The site still answers — it just has no servers to show.

The community's character

sites.<name>.character says what kind of community the site serves, and it is declared, never guessed: no share of PvP kills can tell a quiet PvP server from a mixed one, so the site never infers it from the data. Three values, nothing else — absent means pvpve:

Value The pilot page leads with It still shows
pvp the player-versus-player record: PvP kills, the PvP K/D ratio, and the PvP figures first in the strip the targets, the airframes and the competitive rating
pve the targets (air, ground, air defence) and the air kills in the strip the PvP record and the competitive rating
pvpve both combat panels side by side, the strip in the balanced order —

It changes emphasis, not content: the same figures are always on the page, in the same sections. The pilot page states the character in one line — PvP community, PvE community, PvP and PvE community — so a visitor knows why the page leads with those numbers.

The leaderboard's default metric does not follow the character in this build: a visitor's own choice always wins there, and the default stays the site's own.

# config/services/webservice.yaml
DEFAULT:
  sites:
    root:
      character: pve        # a PvE community: targets and airframe hours lead

Branding

Everything a site can wear lives in its own block. Every key is optional.

branding

Key Type Default Where the value goes
title string DCSServerBot The site's name in the top bar and in the document <title>.
logo string (path) none — the shipped mark /static/logo-mark.png The image in the top bar, instead of the shipped mark. Only raster images are served: .png, .jpg, .jpeg, .gif, .webp. An .svg — or any other suffix — is not refused, but no logo route is registered and the shipped mark is drawn instead.
footer_links list of {label, href} [] — no footer links The footer's anchors, in declaration order, right of the attribution. href must be an http:///https:///mailto: URL or a path beginning with /; no quotes, no </> and no protocol-relative //host form. Any other shape is refused at startup.

The footer's text is the product's attribution — Powered by DCSServerBot — written as a literal in services/webservice/templates/site_base.html. It is not configurable: no key sets it, no site can shorten it, and it is identical in every language. The only thing a site adds to the footer is its own anchors (footer_links), which sit right of the attribution. There is no branding.footer_html key and no site template is editable, so a site cannot replace the footer either.

branding.copyright is gone from the editor and from the footer. A site.yaml (or a legacy config/services/webservice.yaml block) that still carries it loads as before and the value is ignored — the key is accepted so an older file never refuses to start, and it renders nowhere.

The one thing outside site.yaml that can reach the page's look is the operator's own uploaded palette sheet (theme.sheet), which is served exactly as written — it is the operator's CSS, not a site setting, and it can of course hide any element. Nothing in the site's configuration, and no template, can remove the attribution.

footer_links entries are the whole vocabulary:

    branding:
      footer_links:
        - {label: Imprint, href: /imprint}
        - {label: Discord, href: "https://discord.gg/example"}

A link whose href starts with http is rendered with rel="noopener".

theme

A site's palette has three sources, and a theme block names at most one of them.

Key Type Default Meaning
preset string none — the shipped palette The id of a shipped preset: brass (the shipped look), slate, midnight or parchment. A light preset sets the page's colour scheme with it. An id this build does not ship warns once and falls back to the shipped palette.
sheet string none — the shipped palette The LEAF file name of a .css file in the site's asset folder (site/ under the configuration directory), uploaded on the Public site page — which always stores it as theme.css. It is served exactly as written, with no validation; see An uploaded sheet is served as written. A name whose file is not there is refused at startup, naming the key and the file.
expert string none — the shipped palette The body of one :root rule: --name: value; declarations, one per line, at most 8 KB. The server wraps the text in :root { … }, so a selector cannot be expressed at all.
light boolean false Sets data-theme="light" on the <html> element instead of data-theme="dark".

A block that sets two of preset, sheet and expert is refused at startup, naming both keys.

light on its own changes nothing visible

The shipped site stylesheet declares no light palette — the console's light palette lives in shell.css, which a public page does not load. light: true sets the attribute; to get a light site, name a light preset (slate or parchment): a preset answers the page's colour scheme with its sheet, so the page follows what it renders.

A palette is token-only

A preset and the expert block re-declare the site's custom properties and nothing else. The site's components are the shipped ones, and the shipped component selectors are an implementation detail rather than a published interface: an expert block that targets a component is refused at startup, because its grammar accepts declarations and nothing else, and the presets are files this build itself ships.

An uploaded sheet is served as written

theme.sheet is the third source, and the only one that is not checked. The Public site page (Public site → the palette block → Your own sheet) takes a .css file and stores it in the site's asset folder as theme.css; <site>/theme.css then answers that file byte for byte.

What the page does check is the upload, not the CSS:

  • the file must be named *.css;
  • it must be UTF-8 text (a file that is not is refused rather than served with replacement characters);
  • it must be at most 256 KB.

Everything else is the operator's. A sheet may therefore carry @font-face, url() and component rules, and a browser loading it will fetch what it names from this installation's own origin — so a sheet is code-adjacent content, exactly like a stylesheet you would add to the deployment yourself. Upload only what you would put there. The expert sheet is the validated path; the upload is for a sheet that already exists.

Re-uploading replaces the file (and the palette) at once; removing it clears theme.sheet and the site falls back to its preset. Both are per-site subjects in the console's write gate (site.write), and both carry the console's CSRF token.

The three sources stay exclusive in the save's direction too: picking a preset tile (or writing an expert sheet) while an uploaded sheet is still stored is refused, naming both keys — remove the sheet first, then choose.

A sheet does not decide the page's colour scheme: theme.light: true (or a light preset) does, since color-scheme is not something a sheet's own values can set.

These are the custom properties the shipped stylesheet declares. All of them may be re-declared.

Group Custom properties
Ground and ink --ground, --stage, --band, --panel, --panel-2, --line, --line-2, --black, --fg, --fg-2, --fg-3
Accent --brass, --brass-dim, --brass-wash, --brass-hi, --sage
Washes and glows --wash, --hair, --hero-glow, --hero-sage, --pulse-ring, --pulse-none, --pulse-edge
Medals --medal, --medal-font, --medal-ink, --medal-gold, --medal-silver, --medal-bronze
Type families --disp, --body, --mono
Type scale --t-2xs, --t-xs, --t-sm, --t-md, --t-lg, --t-xl, --t-2xl, --t-3xl, --t-display, --t-giant, --t-phead, --t-display-sm, --t-giant-sm, --t-phead-sm, --t-chart
Figure sizes --t-readout, --t-readout-sm, --t-panel-fig, --t-table-fig, --t-unit-giant, --t-unit, --t-unit-lg, --t-unit-xl
Spacing --sp-0x, --sp-0y, --sp-0f, --sp-0h, --sp-0e, --sp-0t, --sp-0q, --sp-0s, --sp-0u, --sp-0v, --sp-0w, --sp-0z, --sp-1, --sp-2, --sp-3, --sp-4, --sp-5, --sp-6, --sp-7, --sp-8, --gap
Page frame --max, --pad, --pad-y, --radius, --hair-w, --rule-w, --panel-pad
Top bar and nav --top-h, --top-h-sm, --top-gap, --nav-gap, --mark-h, --mark-h-brand, --mark-gap, --search-w, --monogram
Controls and marks --dot, --dot-offset, --bracket, --bar-h, --bar-h-lg, --focus-w, --focus-offset, --mock-h
Hero and masthead geometry --hero-cap, --hero-pad-top, --hero-pad-bottom, --hero-foot-gap, --hero-cap-sm, --hero-cut-sm, --giant-pad-top, --giant-pad-bottom, --giant-gap, --giant-side, --phead-cap, --phead-cap-sm, --phead-cut-sm, --phead-offset, --phead-gap, --phead-pad-top, --phead-pad-bottom
Chart, gauges and tables --tick, --tick-step, --tick-step-brass, --track-step, --hair-step, --p, --ring, --ring-inset, --ring-sm, --ring-sm-inset, --readout-h, --rank-w, --rank-w-sm

The eight values a rebrand actually needs are the ground and ink tokens plus the accent: --ground, --stage, --panel, --line, --fg, --fg-2, --fg-3 and --brass.

Where a site's logo is read from

branding.logo is resolved like this:

  • ~ and environment variables ($HOME, %USERPROFILE%) are expanded;
  • an absolute path is used as it stands;
  • a relative path is resolved against the bot's configuration directory — the config/ directory holding main.yaml (the --config option, default config).

So logo: site/crest.png is config/site/crest.png. Write the path relative to config/, not to the installation root: the schema's example spells config/site/crest.png, which would be read as config/config/site/crest.png. An absolute path avoids the question entirely.

A worked example

Two files: the site's block and its crest beside it.

# config/services/webservice.yaml
DEFAULT:
  frontend: true
  sites:
    guards:
      slug: guards
      group_role: "5th Guards"
      branding:
        title: 5th Guards
        logo: site/crest.png
        footer_links:
          - {label: Imprint, href: /imprint}
          - {label: Discord, href: "https://discord.gg/example"}
      theme:
        expert: |
          --ground: #10141c;
          --stage: #0b0f15;
          --band: #1b2230;
          --panel: #161d29;
          --panel-2: #1d2534;
          --line: #2b3648;
          --line-2: #3d4a60;
          --fg: #eef2f8;
          --fg-2: #ccd4e0;
          --fg-3: #a3aebf;
          --brass: #6ea8ff;
          --brass-hi: #9cc4ff;
          --brass-dim: #3f6ea8;
          --brass-wash: rgba(110, 168, 255, .12);
      lang: en
      character: pvp

The expert value is the body of the sheet: the server wraps it in :root { … } and serves it at <site>/theme.css. theme: {preset: slate} does the same job with a shipped sheet, in one line, and theme: {sheet: theme.css} serves the .css file the Public site page uploaded.

config/
├── services/webservice.yaml
└── site/
    └── crest.png

What each piece changes on the page: the top bar and the browser tab read 5th Guards, the top bar draws crest.png in place of the shipped mark, every accent — section rules, the live dot, the medals' surroundings, links — turns from brass to blue, and the footer reads Powered by DCSServerBot followed by the two links.

Applying the change

  • The sites: block's STRUCTURE — which sites exist, their slugs, their switches — is applied when the console's apply runs, not when a browser merely reloads. A save on the Public site page offers that apply, and the apply rebuilds the site table in place, so a site switched off or on, or re-addressed, takes effect at once — no bot restart. A hand edit of site.yaml on disk takes effect when the apply is pressed, and not before: a browser reload alone re-reads nothing.
  • The look's keys — branding.* and theme.* among them — were already resolved per request, so they never needed the apply and are unchanged by this. theme.preset, theme.sheet and theme.expert are resolved per request from the site file, and the sheet is answered with Cache-Control: no-cache, so a palette edit — or a re-uploaded sheet — shows on a normal reload. branding.logo is served with Cache-Control: max-age=3600, so a replaced crest can take up to an hour or a hard reload.

Embedding it in an existing website

Four honest options, from the least to the most work.

The simplest thing that works: a menu entry or a link on the site you already run, pointing at https://<host>/ (the root site) or https://<host>/<slug>. Nothing else is needed — the public site is anonymous, so a plain <a href> is enough.

Put it in a frame

The service sends no frame-blocking header. Verified on a live response: an X-Frame-Options header is absent, no Content-Security-Policy with frame-ancestors is sent, and the only headers the page carries are content-type and vary: Cookie. So an <iframe> renders the site inside your own page.

<div style="position:relative;width:100%;aspect-ratio:16/10">
  <iframe src="https://stats.example.org/guards"
          title="5th Guards statistics"
          loading="lazy"
          style="position:absolute;inset:0;width:100%;height:100%;border:0"></iframe>
</div>

The wrapper is what makes it responsive: aspect-ratio gives the frame a height derived from its width, and the absolutely positioned iframe fills it. title is the accessible name a screen reader announces; loading="lazy" keeps the frame from fetching until it is scrolled near.

Two limits come with a frame, and neither is a bug:

  • The site's own top bar comes with it — brand, menu, search, language pill and the sign-in chip, 72 px tall inside the frame. To embed just a table, use the API below instead.
  • A fixed-height frame cannot grow with the page. The site is one scrolling document; an iframe is a window of a fixed size onto it, so the reader scrolls twice. Size the frame for the content you expect to be looked at, not for the whole page.

Read the numbers from the API

The public site itself has no JSON twin for its pages. Exactly one JSON address exists on it — <site>/search.json — and the overview, the leaderboard, the pilots page and the profile have no JSON form at all. A site that wants the numbers in its own layout reads them from the REST API, which is the surface that answers JSON:

curl -H "X-API-Key: <auth.api_key>" "http://localhost:9876/serverstats"

The REST API is mounted on the same WebService process and port, at the prefix: of the restapi plugin's DEFAULT block — which defaults to "", so its paths live at the root.

Endpoint Method Returns
/serverstats GET Statistics of a server, or of every server when server_name is omitted
/server_attendance GET Detailed attendance statistics, per server
/leaderboard GET The leaderboard for one metric (what), optional server_name
/highscore GET Highscore for a period (period), optional server_name
/topkills GET Top kills
/topkdr GET Top KDR
/trueskill GET TrueSkill ratings
/stats POST One player's statistics
/modulestats POST One player's module statistics
/player_info POST One player's information
/weaponpk POST Weapon PK for one player
/traps POST Carrier traps
/traps/img GET One trap's image
/greenieboard POST The greenieboard
/events GET Mission events
/refuelings GET Refuelling statistics
/credits POST Campaign credits
/squadron_credits POST Squadron campaign credits
/squadrons GET Every squadron and its roles
/squadron_members POST One squadron's members
/player_squadrons POST One player's squadrons
/getuser POST Users by name
/current_server GET The server a player is flying on
/servers GET Every server, its active mission and its extensions

Every one of them is gated by the same rule, decided per request from the caller's address:

Scope Peer address Default requirement
local loopback none — no credential
trusted an address inside security.trusted (defaults: 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16, fc00::/7) api_key
remote anything else jwt
  • api_key is satisfied by an X-API-Key header carrying auth.api_key, or by a JWT.
  • jwt is satisfied by Authorization: Bearer <token>, verified against auth.jwt.jwks_url with audience dcssb and a shared key claim.
  • security.default.<scope> overrides the default for a scope; endpoints.<path>.security.<scope> overrides it for one endpoint, where <path> is the route without its leading slash — e.g. endpoints.serverstats.security.remote: none.
  • endpoints.<path>.enabled: false does not register the route at all (it answers 404).
  • A word that is not one of none, api_key, jwt or disabled falls back to jwt with a warning, never to an open door.

Two things to keep in mind:

  • The API is installation-wide, not site-scoped. A site's public pages show only its own servers; the API does not know about sites, so a caller sees every server the installation has. Filter by server_name where the endpoint accepts one, and keep remote at jwt.
  • Never put the API key in a page. A browser-side X-API-Key is a published key. Proxy the call from your own backend, or publish only the public site.

Carry the language

A visitor's stored language is the dcssb_lang cookie — one cookie, path /, valid for a year, shared with the Command Center, so a choice made on one surface applies on the other.

A page of the public site renders in, in this order:

  1. the dcssb_lang cookie, when it names a language that has a webui catalog;
  2. site.lang — the language the site's block declares.

Accept-Language and main.yaml are not consulted on a public page: a site that declares German stays German for a visitor who has chosen nothing.

A link may carry the choice. The switch is a plain GET to <site>/language, and it accepts lang and next:

<a href="https://stats.example.org/de/rankings">Rankings</a>
<a href="https://stats.example.org/de/language?lang=fr&next=/de/rankings">Français</a>
  • lang must be one of the languages with a catalog: en, cn, de, es, fr, it, pl, pt, ru, tr. Anything else changes nothing and just returns the visitor where they were.
  • next is validated against that site's own tree; anything else — an absolute URL, //host, a console path or another site's prefix — lands on the site's root instead.
  • The address is <site>/language, so the root site's switch is the installation's fixed /language, which is the same route with the same cookie.

Gotchas

The pages are public, and the templates treat that as a rule. Nothing that identifies a person or an account reaches a public page: no Discord ID, no e-mail address, no IP address or hash, no ban, watch or moderation state, no session detail. A pilot profile shows the pilot's figures, the airframes and servers they flew, and their other UCIDs — never the Discord account behind them, and never the wall-clock time of a session. That is why the /me page is a convenience over figures the profile already publishes, not an extra disclosure.

The figures are only as fresh as the last refresh. The site reads materialised views (mv_serverstats, mv_site_activity, mv_site_pilots, mv_site_pilot_totals, mv_site_pilot_aircraft, mv_site_missions, mv_site_aircraft, mv_site_hours), all of them owned and refreshed by the userstats plugin: its hourly task refreshes them every hour, concurrently, mv_statistics first (mv_serverstats is built from it). A fresh install — and any install that has just gained one of these views — shows an empty board, or zeros, until that refresh has run once. To populate them at once:

REFRESH MATERIALIZED VIEW CONCURRENTLY mv_statistics;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_serverstats;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_activity;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_pilots;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_pilot_totals;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_pilot_aircraft;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_missions;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_aircraft;
REFRESH MATERIALIZED VIEW CONCURRENTLY mv_site_hours;

mv_site_hours — the "When we fly" panel — is split across the hours a session actually spanned: each session's hop_on … hop_off span is cut at the UTC hour boundaries and the flight seconds are summed per hour, so a pilot who joins at 19:00 and flies until 23:00 fills four bars instead of one. The panel draws 24 bars, one per UTC hour.

An empty board with a populated statistics table is that refresh, not a broken page.

If a page answers 500 the reason is in the bot's own log, logs/dcssb-<node>.log — the site runs inside the bot's process, so there is no second log file. A 404 on a site address is usually a mistyped slug: a first path segment that names no configured site answers 404 and never falls back to the root site.