diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md
new file mode 100644
index 0000000..90bcaba
--- /dev/null
+++ b/ARCHITECTURE.md
@@ -0,0 +1,273 @@
+# Architecture and Design of Spectator
+
+This document explains the structure and design decisions behind Spectator.
+It is broken up into the logical components of Spectator:
+
+- [Terms](#terms)
+- [DSL](#dsl) - Domain Specific Language. Macros and methods that build a spec.
+- [Matchers](#matchers)
+- [Examples and groups](#examples-and-groups)
+- [Runner and harness](#runner-and-harness)
+  - [Hooks](#hooks)
+- [Mocks and doubles](#mocks-and-doubles)
+  - [Stubs](#stubs)
+  - [Doubles](#doubles)
+- [Formatting](#formatting)
+
+## Terms
+
+The following are terms and concepts frequently used in the project.
+They are listed in alphabetical order,
+but you may find it useful to jump around when learning what they mean.
+
+**Assertion**
+
+An *assertion* is a fundamental piece of a test.
+It checks that a condition is satisfied.
+If that condition isn't met, then an exception is raised.
+
+**Builder**
+
+A *builder* is a type that incrementally constructs a complex object.
+Builders are primarily used to create *specs* and *example groups*.
+See: https://sourcemaking.com/design_patterns/builder
+
+**Config**
+
+Short for *configuration*, a *config* stores information about how to run the *spec*.
+The configuration includes parsed command-line options, settings from `.spectator` and `Spectator.configure`.
+
+**Context**
+
+A *context* is the scope or "environment" a test runs in.
+It is a part of an *example group* that provides methods, memoized values, and more to an example block.
+From a technical standpoint, it is typically an instance of the class defined by an `example_group` block.
+It can thought of as a closure.
+
+**Double**
+
+Stand-in for another type.
+*Doubles* can be passed to methods under test instead of a real object.
+They can be configured to respond to methods with *stubs*.
+*Doubles* also track calls made to them.
+An important note: a *double* is _not_ the same type (nor does it inherit) the replaced type.
+
+**DSL**
+
+*DSL* stands for **D**omain **S**pecific **L**anguage.
+It is the human-like language that comprises a *spec*.
+Keywords in the *DSL*, such as `describe`, `it`, and `expect`, are macros or methods.
+Those macros and methods make calls to Spectator to describe the structure of a *spec*.
+They are combined in such a way that makes it easy to read.
+
+**Example**
+
+An *example* is essentially a *test* and metadata.
+Spectator makes a distinction between *test* and *example*.
+An *example* can have a description, *group*, *context*, *result*, and *tags*.
+That is to say: an *example* is the *test* and information for execution.
+An *example* is a type of *node*.
+
+In the *DSL*, an *example* is created with `example` and `it` blocks and their variants.
+
+**Example Group**
+
+An *example group* (or *group* for short), is a collection of *examples*.
+*Groups* can be nested in other *groups*, but can only have one parent.
+*Groups* can have *hooks*.
+*Groups* have extra properties like a name and metadata.
+A *group* is a type of *node*.
+
+In the *DSL*, an *example group* is created with `example_group`, `describe`, and `context` blocks and their variants.
+
+**Expectation**
+
+An *expectation* captures a value or behavior and whether it satisfies a condition.
+*Expectations* contain *match data*.
+They are bubbled up from the *harness* to the runner.
+*Expectations* can be thought of as wrappers for *assertions*.
+
+In the *DSL*, an *expectation* is the code: `expect(THIS).to eq(THAT)`
+An *expectation target* is just the `expect(THIS)` portion.
+
+**Formatter**
+
+A *formatter* takes *results* and reports them to the user in a specific format.
+Examples of *formatters* are XML, HTML, JSON, dots, and documentation.
+The runner will call methods on the *formatter*.
+The methods called depend on the type of *result* and the state of the runner.
+For instance, `#example_started` is called before a an example runs,
+and `#dump_summary` is called at the end when all *results* are available.
+
+**Harness**
+
+A *harness* is used to safely wrap *test* code.
+It captures *expectations* and creates a *result* based on the outcome.
+
+**Filter**
+
+A *filter* selects *nodes* to be included in a running *spec*.
+There are multiple types of *filters*.
+
+**Hook**
+
+A *hook* is a piece of code to execute at a key time.
+For instance, before a *test* starts, or after everything in a *group* completes.
+*Hooks* can be run in the same *context* as a *test*.
+These are known as "example hooks."
+*Hooks* that don't run in a *context*, and instead run independent of *examples*, are called "example group hooks."
+*Hooks* are attached to *groups*.
+
+**Label**
+
+A *label* is a string from the *spec* that identifies a expression.
+*Labels* are captured to improve the readability of *results* and *match data*.
+
+In the following code, the labels are: `does something useful`, `the_answer`, and `42`.
+
+```crystal
+it "does something useful" do
+  expect(the_answer).to eq(42)
+end
+```
+
+**Matcher**
+
+A *matcher* defines an expected value or behavior.
+*Matchers* are given an "actual" value from a test and produce *match data*.
+The *match data* contains information regarding whether the value or behavior was expected (satisfies a condition).
+They behave similarly to an instance of a `Regex`.
+
+In the following code, the `eq(42)` portion returns an instance of a *matcher* expecting the value 42.
+
+```crystal
+expect(the_answer).to eq(42)
+```
+
+**Match Data**
+
+*Match data* is produced by *matchers*.
+It contains information regarding whether an *expectation* is satisfied and values from the match.
+The values are key-value pairs identifying things such as "expected value" and "actual value."
+*Match data* is similar in concept to `Regex::MatchData`.
+
+**Mock**
+
+A *mock* is a type that can have its original functionality "swapped out" for a *stub*.
+This allows complex types to be "mocked" so that other types can be unit tested.
+*Mocks* can have any number of *stubs* defined.
+They are similar to *doubles*, but use a real type.
+
+**Node**
+
+A *node* refers to any component in a *spec*.
+*Nodes* are typically *examples* and *example groups*.
+A *node* can have metadata associated with it, such as a *label*, location, and *tags*.
+
+**Procsy**
+
+A *procsy* is a glorified `Proc`.
+It is used to wrap an underlying proc in some way.
+Typically used to wrap an *example* when passed to a *hook*.
+
+**Profile**
+
+A *profile* includes timing information for *examples*.
+It tracks how long each *example* took and sorts them.
+
+**Report**
+
+A *report* is a collection of *results* generated by running *examples*.
+It provides easy access to various metrics and types of *results*.
+
+**Result**
+
+A *result* summarizes the outcome of running an *example*.
+A *result* can be passing, failing, or pending.
+*Results* contain timing information, *expectations* processed in the *example*, and an error for failing *results*.
+
+**Spec**
+
+A *spec* is a collection of *examples*.
+Conceptually, a *spec* defines the behavior of a system.
+A *spec* consists of a single, root *example group* that provides a tree structure of *examples*.
+A *spec* also contains some *config* describing how to run it.
+
+**Stub**
+
+A *stub* is a method in a *double* or *mock* that replaces the original functionality.
+*Stubs* can be attached to a single instance or all instances of a type.
+
+**Tag**
+
+A *tag* is an identifier with optional value.
+*Tags* can be used to group and filter *examples* and *example groups*.
+Some *tags* have special meaning, like `skip` indicating an *example* or *group* should be skipped.
+
+**Test**
+
+The word "test" is overused, especially when using a testing framework.
+We make an effort to avoid using the word "test" for everything.
+However, *test* has a technical meaning in Spectator.
+It refers to the code (block) executed in an *example*.
+
+```crystal
+it "does a thing" do
+  # Test code starts here. Everything inside this `it` block is considered a test.
+  expect(the_answer).to eq(42)
+  # Test code ends here.
+end
+```
+
+## DSL
+
+The DSL is made up of methods and macros.
+What look like keywords (`describe`, `it`, `expect`, `eq`, etc.) are just macros and methods provided by the DSL.
+Those macros and methods are defined in multiple modules in the `Spectator::DSL` namespace.
+They are logically grouped by their functionality.
+
+Each module is included (as a mix-in) to the base Spectator context that all tests use.
+The `SpectatorTestContext` class includes all of the DSL modules.
+
+The DSL methods and macros should be kept simple.
+Their functionality should be off-loaded to internal Spectator "APIs."
+For instance, when the DSL creates an example,
+it defines a method for the test code and calls another method to register it.
+While Crystal macros are powerful, excessive use of them makes maintenance harder and compilation slower.
+Additionally, by keeping logic out of the DSL, testing of internals becomes less dependent on the DSL.
+
+*TODO:* Builders...
+
+*TODO:* Tricks...
+
+## Matchers
+
+*TODO:* Base types
+
+## Examples and groups
+
+*TODO*
+
+## Runner and harness
+
+*TODO*
+
+### Hooks
+
+*TODO*
+
+## Mocks and doubles
+
+*TODO*
+
+### Stubs
+
+*TODO*
+
+### Doubles
+
+*TODO*
+
+## Formatting
+
+*TODO*