diff --git a/README.md b/README.md
index 9b9aeeb..9f9efe1 100644
--- a/README.md
+++ b/README.md
@@ -25,3 +25,45 @@ Try it [on Compiler Explorer](https://godbolt.org/z/T-MFoh).
**Known issues:**
* With C++17 GCC, `to_string` must be used to initialize variables; otherwise, the integer-string conversion is done at run-time.
+
+# How it works
+
+The basic structure of `to_string` is shown below:
+
+```cpp
+template
+struct to_string_t {
+ char buf[]; // Size selection explained later.
+ constexpr to_string_t() {} // Converts the integer to a string stored in buf.
+ constexpr operator char *() {} // These allow for the object to be implicitly converted
+ constexpr operator const char *() {} // to a character pointer.
+};
+
+template
+to_string_t to_string; // Simplifies usage: to_string_t() becomes to_string.
+```
+
+Since the number and base are template parameters, each differing `to_string` use will get its own character buffer.
+
+The integer/string conversion is done using a simple method I learned over the years, where the string is built in reverse using `n % base` to calculate the value of the lowest digit:
+
+```cpp
+constexpr to_string_t() {
+ auto ptr = buf + sizeof(buf) / sizeof(buf[0]);
+ *--ptr = '\0';
+ for (auto n = N < 0 ? -N : N; n; n /= base)
+ *--ptr = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"[n % base];
+ if (N < 0)
+ *--ptr = '-';
+}
+```
+
+As you may have noticed, `buf` needs to be given a size for all this to work; in fact, the above code relies on the buffer having a size equal to the generated string (or else `buf[0]` would still be uninitialized). This is actually the case: a lambda is used within `buf`'s declaration to count how many characters long the string will ultimately be. This counting is done in a manner similar to conversion loop shown above:
+
+```cpp
+char buf[([] {
+ unsigned int len = N >= 0 ? 1 : 2; // Need one byte for '\0', two if there'll be a minus
+ for (auto n = N < 0 ? -N : N; n; len++, n /= base);
+ return len;
+ }())];
+```