open-standards/wai/quantum-conformance/landscape/stated-result.md

# The stated result

This is a result anyone can reproduce from the scripts here, and the script that checks it.

## The statement

On the commit that holds this file, built with Rust 1.99.1 (`rust-toolchain.toml`), on any machine:

1. The twelve-talk program (`wai-quantum landscape examples talks`) has 531,441 arrangements. Its lowest
   energy is −10, held by 2 of them. The sampling mode, at the default start, reaches −10 at **all 1,000
   seeds from 1 to 1,000**, spending exactly **13,440 accumulates** a run. Seed 1's replay digest is
   `6c40700b29e6193abe0a54d8424f2c65f6658a55849176b21f2e2cce4bca05fc`, and the 1,000 lines
   of `wai-quantum landscape replay` hash, with SHA-256, to
   `9ad4367cf31a607c3edfc3732d740824e4b7d56ae141c5bf772fd087347209dd`.
2. Across every worked example, at its own temperature, at 0 and at 0.5, 200 seeds each, **3,637 of the
   7,800 runs** that run are solved, and the lines hash to
   `d9db5db3fc24f35a8d8c389e720ccbe36e7c3c26bcb16672db064cddc892e95b`. The examples marked as failing fail.
3. The scale table, [`table.json`](table.json), rebuilds to SHA-256
   `aea3a97c927fbce24197d0065f63e1f36a2bda6c359fa67f956c4ab752c32dd6`. In it, against the best of three
   classical methods at the same budget, the sampling mode at the default start is **ahead in 2 rows, tied
   in 45 and behind in 21** of 68. Starting at the largest row of couplings, it is **ahead in 0, tied in 38
   and behind in 30**.

## The check

```sh norun=usage
git clone https://git.transaction.science/Transaction-Science/open-standards
cd open-standards/wai/quantum-conformance/landscape
./reproduce.sh
```

It prints each stated value beside what the machine got, and exits 0 only if every one matches. It needs
Rust through rustup and nothing else. With Node, wasm-bindgen and the `wasm32-unknown-unknown` target, it
also compares the browser build's answers with the native ones. It takes about a minute and a half on a
machine of 32 cores.

## Who has reproduced it

Each row ran `reproduce.sh` on 2026-11-06, on the code at the commit that holds this file.

| who | machine | result |
|---|---|---|
| the author | Linux, x86-64 | every value |
| the author | Linux, arm64 | every value, with the same bytes |
| the author | macOS, Apple M3 Ultra (arm64) | every value, with the same bytes, and the browser build's answers the same as the native ones |

No one outside the project has run it. Until someone does, this is a result that can be reproduced, not one
that has been. The machines above show that the bytes do not depend on the processor or the operating system.

## What it shows, and what it does not

It shows that the emulator's answers, counts and digests are exact and portable, that its failures are
reported as failures, and that a classical method on the same instance leads it on most of the larger rows.
It does not show anything about a machine's speed or energy. The joule line is priced from one FPGA board's
rates, and nothing here ran on any device but the machine that ran the script.