whitepaper

TORPC - whitepaper

LaTeX source for Token Optimized RPC: Reading a Blockchain Without Drowning an Agent in Tokens (draft v0.9).

Authors: Mike Kondratev, Alexander Kolesov, Roman Fasakhov, Stanley Wu. Same author list as the spec and the paper's own title page.

The paper is self-contained - TikZ diagrams are inline, no external images or bibliography files.

Build

Needs Tectonic.

brew install tectonic   # macOS
./build.sh

Output: torpc.pdf.

You can also run tectonic torpc.tex directly.

Files

• torpc.tex - the paper source (edit this)
• torpc.pdf - build artifact. Any edit to torpc.tex requires a rebuild (./build.sh), otherwise the committed render no longer matches the source
• build.sh - build wrapper

What the paper cites, and where it lives

• Spec and per-method rulesets: ../specs/, conformance vectors: ../conformance/ (both in this repository).
• Benchmark corpus, accuracy suite, and the 2026-07-17 live production measurement (script plus per-method table): the bench/ folder of the code repository, https://github.com/w3tech/torpc-js.

Numbers in the paper are quoted from those published artifacts with their run label. If a figure changes, change it in the artifact first, then here.

The measurements the paper is allowed to quote

Four sources, each with its own scope. They are not interchangeable, and every figure in the paper names its own.

• Live production run, 2026-07-17. 21 measured methods (of 23 transformable), 525 raw/T1/T2 triples over 25 random ETH mainnet blocks, o200k_base counted over the full HTTP response body: -35.3% at T1 and -48.4% at T2, token-weighted, or -17.4% and -24.5% unweighted per method. Per-method figures come from bench/results/live-token-savings-2026-07-17.md as published, never from arithmetic performed here.
• Earlier six-method run, 2026-05-27. -34.0% at T1, -45.6% at T2, against the pre-canonical alias header. Quoted only as a consistency check on the newer run.
• Public bench corpus. -45.0% overall at T2 over 114 pinned fixtures, plus the three-model retrieval-accuracy suite: bench/results/BENCHMARK.md and bench/results/SUMMARY.md.
• The spec's worked receipt. 696 tokens raw, 342 at T1, 292 at T2, from ../specs/methods/eth_getTransactionReceipt.md. A single response measured against the production endpoint, outside both runs.

Never attribute a bench-corpus number to production traffic, or the reverse, and never quote a per-method figure that is not in one of those tables.

Licence

CC0 1.0 Universal, like everything else in this repository. See ../LICENSE.

Contributing

Open a PR against main with your changes to torpc.tex. Rebuild torpc.pdf before pushing so the render matches the source.