Dynamic Programming Is a Sparse Walk Through a Number Counter

In my last post I made an argument that felt, to me, like stating the obvious — but one I couldn’t find anyone else putting into words: backtracking is just counting. If your search space is a product of small independent choices, a depth‑first walk through it is the same object as incrementing a mixed‑radix number — an odometer. And a “prune” is just you refusing to descend into a number you can already see is doomed. I wrote the full version of that up: backtracking as a bijection to the integers. ...

September 21, 2026

Beyond Recursion: Implementing Backtracking as a Mixed-Radix State Space

The Combinatorial Gap When we approach a backtracking problem, we are essentially dealing with two different magnitudes of space. First, there is the Total Theoretical State Space. If we have $N$ decision variables, each with a cardinality (base) of $B_i$, the total number of possible configurations is the product: $\prod_{i=1}^{N} B_i$. This represents an astronomical number of potential states. Second, there is the Effective Search Space—the set of paths that remain viable under specific constraints. ...

September 8, 2026

Processing Gigabytes of JSON: Why Streaming is the Only Way to Scale

Processing Gigabytes of JSON: Why Streaming is the Only Way to Scale The Massive Config Audit In large-scale cloud or storage environments, the “state of the world” is often exported as a giant JSON dump (e.g., an inventory of every virtual disk, volume, and node in a cluster). These files can be tens of gigabytes when uncompressed. If we try to open these with a standard JSON library, our application will crash because those libraries typically build a DOM (Document Object Model)—a tree structure in RAM that is often 5-10x larger than the file itself. ...

July 12, 2026

Processing Billions of Integers: A Lesson in Memory Efficiency and Bit Manipulation

Processing Billions of Integers: A Lesson in Memory Efficiency and Bit Manipulation The Challenge In storage systems, we often deal with raw binary dumps of metadata tables. When debugging consistency issues, you may need to scan billions of 32-bit identifiers to find duplicates or unique entries. Because these files can be massive, loading them into a standard hash map is impossible. The task seems simple on the surface: given a binary file containing 1 billion 32-bit integers, count how many are unique and how many appear exactly once. ...

July 7, 2026