diff --git a/README.md b/README.md
index 87ff5fe..677bd99 100644
--- a/README.md
+++ b/README.md
@@ -2,7 +2,11 @@
 
 ![Rust](https://github.com/tonytins/corten/workflows/Rust/badge.svg) [![Build Status](https://travis-ci.org/tonytins/corten.svg?branch=master)](https://travis-ci.org/tonytins/corten) [![Build status](https://ci.appveyor.com/api/projects/status/ffru6ik26j2b87ko?svg=true)](https://ci.appveyor.com/project/tonytins/corten) [![codecov](https://codecov.io/gh/tonytins/corten/branch/master/graph/badge.svg)](https://codecov.io/gh/tonytins/corten)
 
-A stack-based virtual machine initially based on [So you want to build a language VM](https://blog.subnetzero.io/post/building-language-vm-part-01/) tutorial.
+Corten is a stack-based virtual machine written in Rust.
+
+## Specifications
+
+See [specifications](docs/spec.md) page.
 
 ## Requirements
 
diff --git a/docs/spec.md b/docs/spec.md
new file mode 100644
index 0000000..06e2fa3
--- /dev/null
+++ b/docs/spec.md
@@ -0,0 +1,20 @@
+# Specifications
+
+## Iridium
+
+Corten is based on Fletcher Haynes's [So you want to build a language VM](https://blog.subnetzero.io/post/building-language-vm-part-01/) tutorial. His virtual machine used for the tutorial is known as [Iridium](https://github.com/fhaynes/iridium). Despite it's origins, it does aim to be full a fledged virtual machine and is already on it's [third iteration](https://gitlab.com/fletchercp/iridium3) with support for SSH, PIDs and Strings. Corten aims to be Iridium 1-compatible.
+
+## Instruction Set 
+
+| Register | Opcode |
+| --- | --- |
+| 0 | LOAD |
+| 1 | ADD |
+| 2 | SUB |
+| 3 | MUL |
+| 4 | DIV |
+| 6 | HLT |
+| 5 | JMP |
+| 8 | JMPF |
+| 9 | JMPB |
+| _ | IGL |
\ No newline at end of file
diff --git a/src/instruction.rs b/src/instruction.rs
index c99c802..7f3e701 100644
--- a/src/instruction.rs
+++ b/src/instruction.rs
@@ -6,6 +6,7 @@ pub enum Opcode {
     IGL,
     LOAD,
     ADD,
+    SUB,
     MUL,
     DIV,
     JMP,
@@ -14,12 +15,13 @@ pub enum Opcode {
 impl From<u8> for Opcode {
     fn from(vm: u8) -> Self {
         match vm {
-            0 => Opcode::HLT,
-            1 => Opcode::LOAD,
-            2 => Opcode::JMP,
-            3 => Opcode::ADD,
-            4 => Opcode::MUL,
-            5 => Opcode::DIV,
+            0 => Opcode::LOAD,
+            1 => Opcode::ADD,
+            2 => Opcode::SUB,
+            3 => Opcode::MUL,
+            4 => Opcode::DIV,
+            6 => Opcode::HLT,
+            7 => Opcode::JMP,
             _ => Opcode::IGL,
         }
     }
diff --git a/src/vm.rs b/src/vm.rs
index 5eabc11..5e77305 100644
--- a/src/vm.rs
+++ b/src/vm.rs
@@ -1,3 +1,4 @@
+#![allow(dead_code)]
 use crate::instruction::Opcode;
 
 pub struct VM {
@@ -53,7 +54,7 @@ impl VM {
         match self.decode_opcode() {
             Opcode::LOAD => {
                 let reg = self.next_8_bits() as usize;
-                let num = self.next_16_bits() as u32;
+                let num = self.next_16_bits() as u16;
                 self.registers[reg] = num as i32;
             }
             Opcode::HLT => {
@@ -83,6 +84,11 @@ impl VM {
                 let target = self.registers[self.next_8_bits() as usize];
                 self.pc = target as usize;
             }
+            Opcode::SUB => {
+                let reg1 = self.registers[self.next_8_bits() as usize];
+                let reg2 = self.registers[self.next_8_bits() as usize];
+                self.registers[self.next_8_bits() as usize] = reg1 - reg2;
+            }
         }
         true
     }
@@ -92,6 +98,14 @@ impl VM {
 mod vm_tests {
     use crate::vm::*;
 
+    fn get_test_vm() -> VM {
+        let mut test_vm = VM::new();
+        test_vm.registers[0] = 5;
+        test_vm.registers[1] = 10;
+
+        test_vm
+    }
+
     #[test]
     fn test_crate_vm() {
         let test_vm = VM::new();
@@ -100,37 +114,37 @@ mod vm_tests {
 
     #[test]
     fn test_opcode_hlt() {
-        let mut test_vm = VM::new();
-        let test_bytes = vec![0, 0, 0, 0];
-        test_vm.program = test_bytes;
-        test_vm.run_once();
-        assert_eq!(test_vm.pc, 1);
+        let mut vm = get_test_vm();
+        let test_bytes = vec![6, 0, 0, 0];
+        vm.program = test_bytes;
+        vm.run_once();
+        assert_eq!(vm.pc, 1);
     }
 
     #[test]
     fn test_opcode_igl() {
-        let mut test_vm = VM::new();
+        let mut vm = get_test_vm();
         let test_bytes = vec![200, 0, 0, 0];
-        test_vm.program = test_bytes;
-        test_vm.run_once();
-        assert_eq!(test_vm.pc, 1);
+        vm.program = test_bytes;
+        vm.run_once();
+        assert_eq!(vm.pc, 1);
     }
 
     #[test]
     fn test_jmp_opcode() {
-        let mut test_vm = VM::new();
-        test_vm.registers[0] = 1;
-        test_vm.program = vec![7, 0, 0, 0];
-        test_vm.run_once();
-        assert_eq!(test_vm.pc, 1);
+        let mut vm = get_test_vm();
+        vm.registers[0] = 1;
+        vm.program = vec![7, 0, 0, 0];
+        vm.run_once();
+        assert_eq!(vm.pc, 1);
     }
 
-    /*#[test]
+    #[test]
     fn test_load_opcode() {
-        let mut test_vm = VM::new();
+        let mut vm = get_test_vm();
         // 500 is represented this way in little endian format
-        test_vm.program = vec![0, 0, 1, 244];
-        test_vm.run();
-        assert_eq!(test_vm.registers[0], 500);
-    }*/
+        vm.program = vec![0, 0, 1, 244];
+        vm.run();
+        assert_eq!(vm.registers[0], 500);
+    }
 }