An independent developer has documented the construction of a custom binary serializer and schema language designed to reduce the space occupied by JSON-style data. The project began as an experiment in representing primitive values more compactly, then expanded into a typed format with generated encoders and decoders. Its author reports payload reductions of about 80% in the examples tested, though the claim is a project result rather than an independent benchmark.
The design starts from a familiar inefficiency: a number written as text can require several bytes even when its numeric value fits in one. Simply storing the compact value is not enough, because a decoder must also know its type and where the next field begins. The project addresses that problem with headers that carry type and length information. For larger lengths, it uses a variable-length integer scheme in which seven bits of each byte hold data and another bit indicates whether the header continues.
That foundation allows strings and integers to share one stream without fixed-width overhead for every value. The author then extends the scheme to signed numbers by mapping positive and negative integers into an unsigned sequence, a common technique associated with zigzag encoding. Repeated field names create another source of waste in ordinary JSON, so the format separates a reusable schema from individual records. Once sender and receiver share the schema, the encoded data can identify fields through their declared structure instead of repeating textual keys.
The experiment grew into a small schema language supporting basic types and more complicated structures. The author describes adding arrays, optional values, enumerations and unions, then generating serialization code from those declarations. That shifts complexity toward tooling: compact bytes require both sides to agree precisely about the schema, and changes must be managed so old and new readers do not silently disagree. Human readability and the broad compatibility of JSON are also traded for smaller wire representations.
The project is best understood as a technical exploration rather than a universal replacement recommendation. JSON remains convenient for debugging, interoperability and systems where payload size is secondary. A custom binary format introduces maintenance, compatibility and security obligations that established formats and libraries have already spent years addressing. The published account does not present cross-platform performance tests, independent validation or comparisons against mature alternatives.
Still, the work provides a concrete walkthrough of why binary serialization becomes complicated. Saving bytes is not just a matter of removing punctuation: a format must encode types, lengths, signs, collections and evolution rules while keeping decoding unambiguous. The experiment’s main contribution is showing how those requirements emerge one by one as a compact representation develops into a language and toolchain.
