Hey hey HN,

I built business-use to catch the difficult bugs, those that happen when your code executes fine, but your business logic is broken.

It's a Python/Node.js library (sdk) that lets you assert business invariants in your actual running services:

Some code:

from business_use import initialize, ensure

Somewhere in your config

intialize()

Define your business flow expectations, e.g. in your public API

ensure( id="payment_created", flow="checkout", run_id=body.payment_id, data={"amount": 10001, "status": "pending", "created_by": user_id} )

Somewhere else in your codebase, e.g. in an internal microservice: Auto-approval should be by different user and only for amount <= $100

ensure( id="payment_auto_approval", flow="checkout", run_id=event.payment_id, data={"approved_by”: someone_id}, validator=lambda data, ctx: ctx["data"]["created_by"] != data["approved_by"] and ctx["data"]["amount"] <= 10000, description="Auto-approval finance policy check" )

Why not just use DataDog, New Relic, HoneyComb, etc?

- APM tracks if services are up.

- This tracks if the right business events actually happened.

i.e. your webhook can return 200 with low latency, but if the split job never queued, APM is blind. Those were not built to understand business logic, but infra.

Why not tests?

- Units/integration tests run with mocks in isolation.

- This runs in production/staging with real data, ensuring that your business requirements are met.

Vision:

You can add a rule to your AI agent's system prompt: "When implementing any feature that modifies business logic, add business-use assertions." The agent gets feedback not just on code correctness, but business correctness. As we ship more AI-generated code, we need new kinds of guardrails. This is one of them. (In the repo you can find a Claude Code command for it :D)

Part of the motivation to build this comes from TLA+ (the theory behind it). In particular how it would look like in an actual production environment.

A few examples:

- Payment flows (creation -> risk check -> approval -> notification -> refund / ...)

- Onboardings (signup -> email verification -> welcome email sent -> weekly reminder)

- E-commerce workflows (cart creation -> order creation -> fulfilment API called -> inventory adjustment)

In the repo you can see a demo of the e-commerce case using medusa.js.

I'd love feedback on:

- Is there something out there that we are missing? We saw tracetest, but it was too married to otel... - Is "runtime assertions for business logic" the right framing? - What's missing?

Open source, built in a weekend from BCN by Taras & Ez.-

OpenBSD Jumpstart