diff --git a/README.md b/README.md
index c5b07fa..862871a 100644
--- a/README.md
+++ b/README.md
@@ -1,16 +1,19 @@
# Table of Contents
-1. [Project Goals](#org863a759)
-2. [Architecture](#orgcd91fd0)
- 1. [Scraper](#org3256ad3)
- 2. [Storage](#org7a9a92c)
- 3. [Analysis](#org6ed72dc)
-3. [Roadmap](#org416f14d)
+1. [Project Goals](#org5acb669)
+ 1. [Document and analyze sentiment](#org9291576)
+ 2. [Make data available](#org8054421)
+ 3. [Generalize](#orgdda4b6f)
+2. [Architecture](#org1d6bc40)
+ 1. [Scraper](#org4298028)
+ 2. [Storage](#org1cd413c)
+ 3. [Analysis](#orgaea450e)
+3. [Roadmap](#org6b7660d)
-
+
# Project Goals
@@ -21,7 +24,30 @@
3. Generalize to other public comment tools.
-
+
+
+## Document and analyze sentiment
+
+- Scrape the data, parse, clean, and store. Clearly separate scraper from sentiment analyzer for maximum auditability.
+- Build tests for identifying abuse, such as spam and account fraud
+- Identify any patterns connecting measured sentiment against VA decisions
+
+
+
+
+## Make data available
+
+- Pick a good visualization tool
+
+
+
+
+## Generalize
+
+- Identify scalable ways to apply this toolset to similar problems
+
+
+
# Architecture
@@ -31,7 +57,7 @@
4. Display: TBD
-
+
## Scraper
@@ -42,14 +68,14 @@ Scrapy provides a simple mechanism for browsing and
3. Individual comment page: \`viewcomments.cfm?commentid=X\` - shows regulation title + brief description at the top, plus the comment
-
+
## Storage
One JSONL file per forum/bill.
-
+
## Analysis
@@ -121,7 +147,7 @@ Google and Amazon both return generic sentiment (tone of writing: positive/negat
-
+
# Roadmap