diff --git a/dbnames.md b/dbnames.md
new file mode 100644
index 0000000..da8109a
--- /dev/null
+++ b/dbnames.md
@@ -0,0 +1,112 @@
+The website schema needs to be discoverable. The `sqlite_schema` table (or the `table_list` pragma) and the `table_info` pragma are essential.
+
+The following facts need to be discoverable:
+- The name of each node type.
+- The fields in each node type.
+- The type of each field in a node.
+- The subfields in each field type.
+- The type of each subfield in a field.
+- Whether a field or subfield is repeated, optional, or single.
+- Whether a field has an application-primitive, but non-sqlite-primitive, type, such as datetime.
+- Discriminated union fields.
+
+Field types need to be acyclic. *Do they really?*
+
+Field values need to be deleted when the node that contains them is deleted.
+
+If field types are namespaced to a node type, then SQLite schema object naming will get harder.
+	Maybe just let them be shared.
+
+```
+(node blogpost)
+(composite-field car)
+(node "car review")
+(field cars)
+(field car)
+node:blogpost
+field:
+
+(: road-review node)
+(: car composite-field)
+-- Entity names must be qualified by their type/kind, or there can't be e.g. a node type and field type that share a name.
+-- I think qualification is best, because then SQLite will prevent duplicates by the requirement that table names be unique.
+(list (: car composite-field))
+
+
+create table "car:composite-field" (
+	id integer primary key,
+	"Make" text not null,
+	"Model" text not null,
+	"Year" text not null,
+	"Test distance" int not null,
+	"Test date:date" text not null,
+);
+
+create table "road review.cars:(list-field (road review:node))" (
+	"Road review:node" int primary key references "road review:node",
+	"Car:composite-field" int primary key references "car:composite-field",
+);
+
+create table "Road review:node" (
+	id integer primary key,
+	"Title" text not null,
+	"Cars:(list car)"
+);
+
+(field car)
+(node review)
+(list (field car) (node review)) -- If we parameterize each list relation by both the listed field and the containing node or field, then we've basically monomorphized it.
+-- However, that won't cut it if a node or field contains multiple lists of the same type. We also need to parameterize list relations by the field name in the containing node or field.
+```
+
+```
+primitive-field-type := int | text | date | 
+record-field-type := (record <field-type-name> ((<field-name> <field-type>) ...))
+variant-field-type := (variant <field-type-name> ((<field-name> <field-type>) ...))
+field-type := (field <field-type-name>)
+node-type := (node <node-type-name>)
+parent-object-type := <field-type> | <node-type> 
+list := (list <element-type> <parent-object-type> <field-name>)
+```
+
+
+```
+(record car)
+	(id integer)
+	(make text)
+	(model text)
+	(year text)
+	(test-distance integer)
+	(test-date date)
+
+(list car)
+	(node road-review)
+	(car car)
+
+(node road-review)
+	(id integer)
+	(title text)
+	(cars (list car))
+```
+
+```
+(node <node-type-name>)
+	id integer primary key
+	(<composite-type-name> <field-name>) <db-type> ...
+(list <composite-type-name>)
+	id
+(list-item <composite-type-name>)
+	list-id references (list <composite-type-name>) primary key
+	sequence-number primary key
+	item-id references (composite-value <composite-type-name>)
+(composite-value <composite-type-name>)
+	id
+	-- This indirection is so that SQLite will prevent duplicate field type names in table names.
+	<record|variant>-value-id references (<record|variant>-value <composite-type-name>)
+(record-value <composite-type-name>)
+	id integer primary key
+	(<composite-type-name> <field-name>) <db-type> ...
+(variant-value <composite-type-name>)
+	id integer primary key
+	(<composite-type-name> <field-name>) <db-type> ...
+```
\ No newline at end of file