diff --git a/corrscope/config.py b/corrscope/config.py
index 29212e9..b1788f8 100644
--- a/corrscope/config.py
+++ b/corrscope/config.py
@@ -1,3 +1,8 @@
+"""
+The most important class in this module is `DumpableAttrs`.
+See its docstring for details.
+"""
+
 import pickle
 import warnings
 from enum import Enum
@@ -113,8 +118,73 @@ def copy_config(obj: T) -> T:
 
 class DumpableAttrs:
     """ Marks class as attrs, and enables YAML dumping (excludes default fields).
-    - Subclass parameter `always_dump` contains
-      whitespace-separated list of fields to always dump.
+
+    It is subclassed for
+    (statically-typed key-value objects which will be dumped/loaded as YAML files).
+
+    ## Subclassing `DumpableAttrs` and converting to YAML
+
+    - This class works like `dataclasses` or `attrs.dataclass` decorators,
+      and wraps the latter.
+    - Subclass `DumpableAttrs`, then add class-level type annotations.
+      These annotations are converted into `__init__(...)` constructor parameters.
+
+    ```py
+    class Config(DumpableAttrs):
+        path: str
+        priority: int = 0
+    ```
+
+    Unlike many other libraries and usual JSON,
+    the YAML string representation of a `DumpableAttrs` object
+    encodes what Python type the object is.
+    For example, Config("foo.wav", 1) is dumped as:
+
+    ```yaml
+    !Config
+    path: foo.wav  # Required
+    priority: 1  # If not present, defaults to 0
+    # See DumpableAttrs docstring for details.
+    ```
+
+    ## Polymorphism
+
+    The YAML file determines what type is loaded,
+    so the config type can be used to pick a object type at runtime.
+
+    - For example, putting `!CorrelationTriggerConfig` in YAML
+      loads a `CorrelationTriggerConfig` config object,
+      which tells corrscope to create a `CorrelationTrigger` algorithm object.
+    - Putting `!NullTriggerConfig` in YAML
+      instead loads a `NullTriggerConfig` config object,
+      which tells corrscope to create a `NullTrigger` algorithm object.
+      (This is only used for unit tests, and is incompatible with GUI.)
+
+    ## `KeywordAttrs` are similar, with 2 differences:
+
+    - Subclasses can have non-default arguments after default arguments.
+    - Its constructor can only be called with keyword (a=1, b=2) arguments,
+      not positional (1, 2).
+
+    ## Optional Parameters
+
+    class Config(DumpableAttrs, always_dump="", exclude=""): ...
+    - `always_dump` contains whitespace-separated list of fields to always dump
+      (if equal to default).
+    - If always_dump="*", `exclude` contains whitespace-separated list of fields
+      to not dump (if equal to default).
+
+    ## Loading from YAML: Alias and Ignored
+
+    YAML loading uses __getstate__ and __setstate__.
+
+    If `old = Alias("new")`,
+    then loading a YAML file `old: value` initializes `new = value`.
+
+    If `old = Ignored`,
+    then loading a YAML file `old: value` silently discards `value`.
+    Ignored is unused in my code,
+    since __setstate__ automatically discards unrecognized fields (with a warning).
     """
 
     if TYPE_CHECKING:
diff --git a/corrscope/designNotes.md b/corrscope/designNotes.md
index 8ea4ca9..9464863 100644
--- a/corrscope/designNotes.md
+++ b/corrscope/designNotes.md
@@ -1,3 +1,47 @@
+# Module/Class Structure and Organization
+
+When corrscope is launched, it first executes `__main__.py` and `cli.py`.
+
+## cli.py
+
+If -w is present, it writes a .yaml file. If -p are present, it runs `corrscope.CorrScope` directly. If neither is present, it imports and runs the `gui` subdirectory.
+
+## CorrScope, Config
+
+`corrscope.py` defines classes `CorrScope`, `Config` (and `Arguments`).
+
+- `CorrScope` is the main loop of the program, and only communicates with the GUI through `Arguments`. `CorrScope` requires a `Config` and `Arguments`.
+    - `Arguments` is constructed by the GUI and used to update rendering progress dialogs, etc.
+    - `Config` is a dataclass (see `config.py`) which can be edited through the GUI, or loaded/saved to a YAML file.
+- When `cli.py` creates new configs, `default_config()` is used as a template to supply values. When loading existing YAML files, only dataclass default values are used.
+
+-----
+
+`Config` holds `channels: List[ChannelConfig]`, which store all per-channel settings.
+
+`CorrScope` turns `channels` into a list of `Channel` objects. Each channel uses its own `ChannelConfig` parameters to create:
+
+- self.trigger_wave: Wave
+- self.render_wave: Wave
+- self.trigger: MainTrigger
+
+-----
+
+Each frame:
+
+`CorrScope` reads data from Wave objects, asks MainTrigger
+
+Each channel has its own `trigger_wave`, `render_wave`, and `trigger` instance. For each channel, `trigger_wave` is used by `trigger` to find a "trigger time" near the frame's center time.
+
+The `Renderer` and `Output` instances are shared among all channels.
+
+- `Renderer` turns a list of N-channel "wave file portions" (from `render_wave`) into a RGB video frame.
+- `Output` sends a RGB video frame to ffmpeg, ffplay, or (in the future) an interactive GUI player window.
+
+## config.py
+
+See docstring at top of file.
+
 # Design Notes (cross-cutting concerns)
 
 ## Renderer to Output