📝 document runtime sink state surface

docs/api/index.md

@@ -342,9 +342,12 @@ BitLogger API navigation.
 
 ## Configured logger runtime
 
+- [runtime-sink.md](./runtime-sink.md)
 - [runtime-sink-flush.md](./runtime-sink-flush.md)
 - [runtime-sink-drain.md](./runtime-sink-drain.md)
 - [runtime-sink-close.md](./runtime-sink-close.md)
+- [runtime-sink-pending-count.md](./runtime-sink-pending-count.md)
+- [runtime-sink-dropped-count.md](./runtime-sink-dropped-count.md)
 - [configured-logger.md](./configured-logger.md)
 - [configured-logger-flush.md](./configured-logger-flush.md)
 - [configured-logger-drain.md](./configured-logger-drain.md)

docs/api/runtime-sink-dropped-count.md

@@ -0,0 +1,76 @@
+---
+name: runtime-sink-dropped-count
+group: api
+category: runtime
+update-time: 20260613
+description: Read the cumulative dropped-record count from a RuntimeSink.
+key-word:
+    - runtime
+    - sink
+    - queue
+    - public
+---
+
+## Runtime-sink-dropped-count
+
+Read the cumulative dropped-record count from a `RuntimeSink`. This helper exposes the direct sink-level loss metric for queue-backed runtime variants.
+
+### Interface
+
+```moonbit
+pub fn RuntimeSink::dropped_count(self : RuntimeSink) -> Int {
+```
+
+#### input
+
+- `self : RuntimeSink` - Runtime sink whose dropped-record metric should be inspected.
+
+#### output
+
+- `Int` - Number of dropped records reported by the runtime sink.
+
+### Explanation
+
+Detailed rules explaining key parameters and behaviors
+
+- Queue-wrapped runtime variants forward to the wrapped queue sink's `dropped_count()` value.
+- Plain console and plain file runtime variants return `0` because they do not track queued record drops.
+- The counter is cumulative for the lifetime of the concrete runtime sink value.
+- This method is the direct sink-level API used by `ConfiguredLogger::dropped_count(...)`.
+
+### How to Use
+
+Here are some specific examples provided.
+
+#### When Need Direct Queue Loss Visibility
+
+When a runtime sink may discard records under pressure:
+```moonbit
+if sink.dropped_count() > 0 {
+  println("runtime sink dropped records")
+}
+```
+
+In this example, direct runtime queue loss becomes visible without inspecting inner fields manually.
+
+#### When Compare Queue Tuning Changes
+
+When queue overflow behavior should be checked operationally:
+```moonbit
+ignore(sink.dropped_count())
+```
+
+In this example, the helper exposes the metric needed to compare different queue settings.
+
+### Error Case
+
+e.g.:
+- If the runtime sink is not queue-backed, the method simply returns `0`.
+
+- If callers need the higher-level logger wrapper API, use `ConfiguredLogger::dropped_count()` instead.
+
+### Notes
+
+1. This helper reports cumulative loss, not the reason for that loss.
+
+2. Pair it with `pending_count()` and queue configuration when investigating pressure.

docs/api/runtime-sink-pending-count.md

@@ -0,0 +1,75 @@
+---
+name: runtime-sink-pending-count
+group: api
+category: runtime
+update-time: 20260613
+description: Read the current queued backlog count from a RuntimeSink.
+key-word:
+    - runtime
+    - sink
+    - queue
+    - public
+---
+
+## Runtime-sink-pending-count
+
+Read the current queued backlog count from a `RuntimeSink`. This is the direct sink-level queue metric behind configured logger pending-count behavior.
+
+### Interface
+
+```moonbit
+pub fn RuntimeSink::pending_count(self : RuntimeSink) -> Int {
+```
+
+#### input
+
+- `self : RuntimeSink` - Runtime sink whose current queue backlog should be inspected.
+
+#### output
+
+- `Int` - Current number of pending queued records.
+
+### Explanation
+
+Detailed rules explaining key parameters and behaviors
+
+- Queue-wrapped runtime variants forward to the wrapped queue sink's `pending_count()` value.
+- Plain console and plain file runtime variants return `0` because they do not own a pending queue here.
+- This is a point-in-time metric and may change immediately after it is read.
+- This method is the direct sink-level API used by `ConfiguredLogger::pending_count(...)`.
+
+### How to Use
+
+Here are some specific examples provided.
+
+#### When Need Direct Queue Backlog Visibility
+
+When code is working with a `RuntimeSink` value directly and wants queue pressure information:
+```moonbit
+let pending = sink.pending_count()
+```
+
+In this example, backlog is read from the runtime sink without going through a logger wrapper.
+
+#### When Verify Manual Drain Progress
+
+When drain loops should observe remaining queue backlog directly:
+```moonbit
+ignore(sink.drain(max_items=8))
+ignore(sink.pending_count())
+```
+
+In this example, the metric helps confirm whether manual queue draining reduced backlog.
+
+### Error Case
+
+e.g.:
+- If the runtime sink is not queue-backed, the method simply returns `0`.
+
+- If callers need the higher-level logger wrapper API, use `ConfiguredLogger::pending_count()` instead.
+
+### Notes
+
+1. Use this helper for simple direct queue visibility on `RuntimeSink` values.
+
+2. Pair it with `dropped_count()` when investigating queue pressure and loss.

docs/api/runtime-sink.md

@@ -0,0 +1,85 @@
+---
+name: runtime-sink
+group: api
+category: runtime
+update-time: 20260613
+description: Public runtime sink enum used by config-built synchronous loggers to unify console, file, and queued sink variants.
+key-word:
+    - runtime
+    - sink
+    - type
+    - public
+---
+
+## Runtime-sink
+
+`RuntimeSink` is the public runtime sink enum used by config-built synchronous loggers. It unifies console, file, and queued sink variants behind one runtime-facing sink surface.
+
+### Interface
+
+```moonbit
+pub(all) enum RuntimeSink {
+  Console(ConsoleSink)
+  JsonConsole(JsonConsoleSink)
+  TextConsole(FormattedConsoleSink)
+  File(FileSink)
+  QueuedConsole(QueuedSink[ConsoleSink])
+  QueuedJsonConsole(QueuedSink[JsonConsoleSink])
+  QueuedTextConsole(QueuedSink[FormattedConsoleSink])
+  QueuedFile(QueuedSink[FileSink])
+}
+```
+
+#### output
+
+- `RuntimeSink` - Public runtime sink union used by `ConfiguredLogger` and config-built logging helpers.
+
+### Explanation
+
+Detailed rules explaining key parameters and behaviors
+
+- This is a public enum root type, not a type alias.
+- The variants cover plain console sinks, plain file sinks, and queue-wrapped forms of those sink families.
+- `build_logger(...)` and related config-driven construction paths use this type as the concrete sink behind `ConfiguredLogger`.
+- The type exposes direct runtime helpers such as `flush()`, `drain()`, `close()`, `pending_count()`, and `dropped_count()`.
+- Queue-specific metrics are meaningful only for queued variants, while plain variants usually report `0` for queue counters.
+
+### How to Use
+
+Here are some specific examples provided.
+
+#### When Need A Typed Config-built Runtime Sink Surface
+
+When code should keep the runtime sink union explicit after config assembly:
+```moonbit
+let logger = build_logger(LoggerConfig::new(target="svc"))
+let sink : RuntimeSink = logger.sink
+```
+
+In this example, the concrete runtime sink remains available for direct runtime inspection.
+
+#### When Need Direct Runtime Variant-specific Branching
+
+When behavior should differ between plain and queued runtime sink variants:
+```moonbit
+match sink {
+  QueuedFile(inner) => ignore(inner.pending_count())
+  File(inner) => ignore(inner.is_available())
+  _ => ()
+}
+```
+
+In this example, the enum shape makes runtime sink-specific handling explicit.
+
+### Error Case
+
+e.g.:
+- `RuntimeSink` itself does not have a runtime failure mode.
+
+- Actual file behavior, queue backlog, and close results still depend on the active variant and current backend state.
+
+### Notes
+
+1. `ConfiguredLogger` is the main higher-level alias that wraps `Logger[RuntimeSink]`.
+
+2. Use the direct `RuntimeSink` helpers when code owns the sink value itself instead of only the logger wrapper.