Daily Edition Sources +6

Builder Work: Zerodecimal Makes Vibe Coding Prove Arithmetic

Alexandros Kyriakakis publicly said zerodecimal was built with Claude Code, Ultracode, and Fable 5; the repo is worth studying because the generated arithmetic is surrounded by proof scaffolding.

Diagram Punk poster showing zerodecimal proof scaffolding around AI-assisted arithmetic with cards for HN proof, div10.go, table tests, allocation gates, and benchmark caveat.
Diagram Punkvibe coded arithmetic needs proofs.
repo AlexandrosKyriakakis/zerodecimal evidence
6 source signals 1 repo 6 source signals
Evidence: 6 source signals / July 8, 2026 / Daily Edition
Open Edition Evidence below

The scout source is a Show HN post for zerodecimal, a Go fixed-point decimal library that claims zero allocations, no big.Int fallback, reciprocal division by powers of ten, differential checks, fuzz targets, and committed benchmark comparisons.

The Engineering Problem

Decimal arithmetic is a bad place for casual generated code: overflow, rounding, parsing, formatting, SQL/JSON codecs, and performance shortcuts can all lie quietly. Zerodecimal's interesting move is to make those lies harder to survive the test suite.

The Repo Evidence

The README points to allocation gates, crosschecks, fuzz targets, and benchmark files. div10.go explains the reciprocal-division path, while tables_test.go re-proves generated constants against big.Int.

The Building Skill

The transferable skill is not "let an agent write hard code." It is to make generated hard code prove its own invariants: exact allocation counts, independent table recomputation, differential arithmetic checks, fuzz corpora, and benchmark methodology in the same repo.

What To Inspect Next

The caveat is that committed benchmarks are not independent certification, and the repo has low adoption today. A constructive question for the builder: which proofs would fail first if another contributor changed the reciprocal table or rounding path?

Evidence Trail

Receipts below the story

The article above is the public narrative. This section keeps the source trail and limits on the same page.

Edition
DateJuly 8, 2026
LaneDaily Edition
Confidence87%
Sources6
ReposAlexandrosKyriakakis/zerodecimal

Primary Evidence

Evidence Limits

  • The evidence does not independently certify benchmark claims.
  • The evidence does not prove production adoption or long-term maintenance.
  • What to watch: whether outside contributors can change arithmetic paths without weakening the proof gates.
Letters & Corrections

Send a note to the desk

Corrections, missing context, or a follow-up lead.