Four stages. Each one produces an artefact — a plan, a diff, a deployment record, a changelog — so you are never asked to trust output you cannot inspect.
The order matters more than any single stage. Most tools open at Build; opening at Scope is the whole reason the rest gets cheap.
You bring a sentence. It comes back with a plan: what it thinks you meant, what it would cut, and the questions it refuses to answer for you. You edit the plan, not the code.
Contracts where state genuinely has to settle in the open. Everything else off-chain, because putting a leaderboard on a chain is a decision, not a default.
Testnet first, every time, with the test plan run against it. Mainnet is prepared and handed to you. It never holds a key, and there is no mode where it does.
Failed transactions, gas drift, an interface that broke because an RPC moved. It opens the fix as a plan — the same loop it started with.
Not chain loyalty. A consumer interface has a latency budget and a cost budget, and for most of a decade open settlement did not fit inside either one.
Think about what “one tap” actually requires. The action has to resolve faster than a person notices a pause — a few hundred milliseconds — and cost so little that neither of you thinks about the cost at all. Miss either and you get the pattern everyone recognises: a batch job, a confirmation screen, a spinner, an apology.
BNB Chain settled both questions at once. An open chain that works — and economics that fit a consumer product. Sub-second finality at roughly a cent and a half, on a network that stays public. No closed ledger. No broker in the middle.
The figures above are why that trade is no longer forced. Sub-second finality at roughly a cent and a half sits inside a consumer budget, on a chain that stays open. It is not the fastest number anyone has ever quoted and it is not a promise about next year — it is what the network was doing on the day we measured it.
Flux Ape will still argue the chain out of your plan when it has not earned its place. That is not a hedge. It is the same discipline, pointed at the chain.
Stated up front, because an agent with no published limits is one whose limits you discover at the worst possible moment.
Mainnet transactions are prepared and handed to you. There is no version of this where a deploy happens without a human signature.
State goes on-chain when open settlement is the point. A profile, a leaderboard, a comment thread — those are a database, and the plan will say so.
Issuance is a business decision. An agent quietly adding one to your plan is a failure, not a feature.
Every build opens with a cut list. Approve a plan with everything left in and it will tell you what that will cost first.
