diff --git a/Cargo.toml b/Cargo.toml
index 5d07ce1..0e0abea 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -1,11 +1,14 @@
 [package]
 name = "pow"
-version = "0.1.0"
+version = "0.1.1"
 authors = ["Andrew Dirksen <andrew@dirksen.com>"]
 description = """
-Tag arbitrary data types with proof of work based on sha256.
+Generate or verify sha256 based proofs of work over arbitrary typed data.
 """
 edition = "2018"
+documentation = "https://docs.rs/pow"
+tags = ["proof of work", "pw set", "powtag", "tag"]
+readme = "readme.md"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
diff --git a/readme.md b/readme.md
new file mode 100644
index 0000000..3657cfc
--- /dev/null
+++ b/readme.md
@@ -0,0 +1,5 @@
+# Pow
+
+Typed proof of work tags.
+
+[docs](https://docs.rs/pow)
diff --git a/src/lib.rs b/src/lib.rs
index 57f1bbe..9f3050a 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -8,7 +8,10 @@
 //!
 //! ```
 //! use pow::Pow;
-//! let difficulty = u128::max_value() / 2; // very easy mode
+//!
+//! // very easy mode
+//! let difficulty = u128::max_value() - u128::max_value() / 2;
+//!
 //! let phrase = b"Phrase to tag.".to_vec();
 //! let pw = Pow::prove_work(&phrase, difficulty).unwrap();
 //! assert!(pw.score(&phrase).unwrap() >= difficulty);
@@ -22,14 +25,74 @@
 //! #     SystemTime::now().duration_since(SystemTime::UNIX_EPOCH).unwrap().as_secs()
 //! # }
 //! # use pow::Pow;
-//! let difficulty = u128::max_value() / 10_000 * 9_999; // more diffcult, around 10_000 hashes
+//!
+//! // more diffcult, takes around 100_000 hashes to generate proof
+//! let difficulty = u128::max_value() - u128::max_value() / 100_000;
+//!
 //! let now: u64 = get_unix_time_seconds();
 //! let pw = Pow::prove_work(&now, difficulty).unwrap();
 //! assert!(pw.score(&now).unwrap() >= difficulty);
 //! ```
 
-//! Expected computional cost scales something like
-//! O(u128::max_value() / (u128::max_value() - difficulty))
+//! # Score scheme
+//!
+//! To score a proof of work for a given (target, Pow) pair:
+//! Sha256 is calculated over the concatenation SALT + target + Pow.
+//! The first 16 bytes of the hash are interpreted as a 128 bit unsigned integer.
+//! That integer is the score.
+//! A constant, SALT, is used as prefix to prevent pow reuse from other systems such as proof
+//! of work blockchains.
+//!
+//! In other words:
+//!
+//! ```
+//! # use serde::Serialize;
+//! # use pow::Pow;
+//! # use core::any::Any;
+//! # const SALT: &'static str = "not the actual salt used";
+//! # fn serialize<T: Serialize>(_: &T) -> u8 { 0 } // not the actual serialize function
+//! # fn deserialize(_: &[u8]) -> u128 { 0 } // not the actual deserialize function
+//! # fn sha256(_: &u8) -> [u8; 32] { [0; 32] } // not the actual sha256 function
+//! fn score<T: Serialize>(target: &T, pow_tag: &Pow<T>) -> u128 {
+//!     let bytes = serialize(&SALT) + serialize(target) + serialize(pow_tag);
+//!     let hash = sha256(&bytes);
+//!     deserialize(&hash[..16])
+//! }
+//! ```
+//!
+//! # Serialization encoding.
+//!
+//! It shouldn't matter to users of this library, but the bincode crate is used for cheap
+//! derterministic serialization. All values are serialized using network byte order.
+//!
+//! # Threshold scheme
+//!
+//! Given a minimum score m. A Pow p satisfies the minimum score for target t iff score(t, p) >= m.
+//!
+//! # Choosing a difficulty setting.
+//!
+//! Difficulty settings are usually best adjusted dynamically a la bitcoin.
+//!
+//! To manually select a difficulty, choose the average number of hashes required.
+//!
+//! ```
+//! fn difficulty(average: u128) -> u128 {
+//!     debug_assert_ne!(average, 0, "It is impossible to prove work in zero attempts.");
+//!     let m = u128::max_value();
+//!     m - m / average
+//! }
+//! ```
+//!
+//! Conversely, to calculate probable number of hashes required to satisfy a given minimum
+//! difficulty.
+//!
+//! fn average(difficulty: u128) -> u128 {
+//!     let m = u128::max_value();
+//!     if difficulty == m {
+//!         return m;
+//!     } 
+//!     m / (m - difficulty)
+//! }
 
 mod proof_of_work;