drewstone

Braid PR #153 installed-package proof

Revision: 2c32fe4fe3bdb536dacd66c9952ed2d32f7ca3cc (after merging main PR #152).

Complete gate beelink1-7e1b3dcb1d06722716083e0c745e85b7 passed, exit 0, Node v22.23.3 with frozen dependencies. Repository checks, remaining suite, native process cleanup matrix, release contract, packed public first-task RPC proof and the installed real Codex turn all passed in that gate.

Measure	Baseline 8a2bce8	Verified 2c32fe4
Named test files	133	37
Test lines	75,923	20,087

We paid an AI $2.48 to manage other AIs. The managed team beat the unmanaged one 15–1.

A measured result on AI supervision, with receipts: the plan and budget were frozen before launch, a fixed program graded every answer, a second session replicated the scoring from raw files, and both mistakes we made are published at the top of their documents.

The problem

Quantum computers run circuits, and every two-qubit gate in a circuit is a chance for the hardware to fail — so compilers work hard to remove them. The standard tool, qiskit's level-3 transpiler, is randomized: run it 20 times, keep the best. That best-of-20 defines our benchmark: given 20 circuits and a real device layout, produce a rewrite with fewer two-qubit gates (or the same number, arranged shallower) that provably computes the same thing.

"Provably" is the load-bearing word. A fixed grading program — committed to the repository before any run, untouchable by the AIs — checks each delivered circuit for exact mathematical equival

Screenshots for blueprint-agent PR: partner onboarding no-payment outcome.

UI proof for tangle-network/tax-agent PR #464 — the Tax Year Overview, captured on a local dev run.

	{
	"base": "92cbbc42faf3e034a61dc494396b92643fa15087",
	"syntax": "PASS both changed TypeScript sources",
	"flagsAndHelp": "token-identical",
	"publicationInputObjects": "token-identical",
	"otherCommands": "byte-identical",
	"timeouts": "original SDK150ms and CLI500ms retained with timeout messages; no inflated timeout",
	"receiptLaw": "exact observed pending job+nonzero code, progress->SIGINT/SIGTERM, ready/resume/noPOST assertions retained or strengthened",
	"cleanup": "Promise.finally removes listeners before outer catch exits; test awaits physical close before JSON parse",
	"files": [

	{
	"source": "products/sandbox/web/scripts/generate-sandbox-api-reference.ts",
	"results": [
	{
	"guard": "readMount",
	"entry": "worker.ts",
	"error": "products/sandbox/api/src/worker.ts must mount sandboxesRoutes exactly once; found 0"
	},
	{
	"guard": "readImageMount",

	#!/usr/bin/env node
	// One-off delivery evidence, deliberately outside product source.
	import assert from 'node:assert/strict';
	import { execFileSync } from 'node:child_process';
	import { readFileSync } from 'node:fs';
	import { createRequire } from 'node:module';
	import { resolve } from 'node:path';

	const [base, sdkArgument] = process.argv.slice(2);
	assert(base && sdkArgument, 'Usage: check-error-classes-move.mjs <base> <sdk-directory>');

	#!/usr/bin/env node
	/** Check the egress move, including comments, bindings and deduplicated catches. */
	import assert from "node:assert/strict";
	import { execFileSync } from "node:child_process";
	import { readFileSync } from "node:fs";
	import { resolve } from "node:path";
	import { createRequire } from "node:module";


	const sdk = resolve(process.argv[3]);