Remote configuration
Use Remote config to draft up to 50 typed keys for the selected game environment. Owners and admins can review the before-and-after diff and publish. Members and viewers can read configuration; incoming shared games cannot. An outdated version is refused so another editor's changes are preserved.
Keys start with a lowercase letter and use lowercase letters, digits and underscores, up to 64 characters. Values are strings, finite numbers, booleans or JSON, with an 8 KB limit per value and 20 nesting levels. Types are checked without implicit conversion: the string "false" is not a boolean. Each key can have up to 20 ordered rules; the first matching platform, build or behavioural segment supplies its value. Event properties and levels are unknown during a config fetch, so segments using them cannot be config rules.
Fetch GET /api/v1/ingest/config?deviceId=...&platform=...&gameVersion=... with the build's X-API-Key. Device IDs are required and limited to 256 characters; platform and build are optional and limited to 128. Game and environment come from the key, never a query parameter. The result contains values, experiments and etag. Cache the document and send its ETag in If-None-Match on the next fetch. A matching response returns 304 with no body. Use local defaults or the last cache if a fetch fails.
One successful non-cached config fetch counts as one event on the existing shared meter. A 304 is free. Config fetches have a separate usage line and do not create player analytics events. Ingest game/key/device limits and public-key origin, allowed-build and revocation checks apply. Config fetches additionally allow 30 requests per device per game per minute. Disabled games refuse fetches. Config values are public to anyone holding the build's ingest key; use them for game settings, never credentials or server secrets.
Editing a referenced platform/build segment rebuilds the snapshot and advances the configuration version. A referenced segment cannot be removed or changed to property/level rules. Audit history lists key names and versions, excluding values. Disabled or deleted keys disappear on the next successful fetch, so clients should use their local defaults for missing keys.
Scheduled values. Owners and admins can schedule a value for one published key: a start time, an optional end time up to 90 days later, and a note. The fetch compares the times with its own clock, so a schedule takes effect to the second. While active it replaces the key's rules and default for every player, and when two overlap the later start wins. A game can have 50 upcoming or active schedules, each up to 400 days ahead. Cancel one at any time; creating and cancelling are in the audit log. Schedules cost nothing extra: a cached client keeps its free 304 until a schedule actually changes a value it receives.
Staged rollouts. Roll a new default out to a rising share of devices, such as 10%, 50% then 100%, with 1 to 168 hours per stage. A device's place is a stable hash, so a device inside 10% stays inside at 50%. Segment and experiment rules still apply first. Every quarter hour Ravensight halts the rollout if one of the game's monitors for errors, progression failures, sessions, active players or session length has moved the wrong way since it started, or if errors per player inside the rollout during the stage exceed players outside it over the same window by more than your limit (10% to 500%), with at least 50 players inside. The comparison never treats fewer than 0.02 errors per player as the baseline, so one early error does not stop a rollout. At 100% the comparison is with the same length of time just before the stage. A publish that changes the key's default or removes the key cancels its rollout, and completion never overwrites a default someone published meanwhile. Halting returns every player to the previous value and records why. After the last stage the value becomes the key's default. A game can run five rollouts at once, one per key.
MCP list_config_schedules reads schedules and rollouts, with their status and why a halted rollout stopped. It creates and cancels nothing.
MCP list_remote_config reads names, types, enabled flags, versions and rule counts at the normal programmatic rate. It does not return values or publish changes. Configuration writes use the authenticated website API. All nine SDKs include typed config access, scoped ETag caching and explicit experiment exposure labels. Running experiments add deterministic assignments and matching variant rules to the same response.