The first instinct on a blockchain project is to put as much logic on-chain as possible. It feels principled: transparent, trustless, final. In institutional systems, that instinct is usually wrong.

A smart contract is the hardest component in the system to change, the most exposed to attack and the most expensive to audit. Every line it contains is a line you will defend for years. So we ask a narrower question: what must be on-chain for the system to be trustworthy, and what merely could be?

What belongs on-chain

  • Custody of the asset and the rules that constrain its movement.
  • Invariants that participants must be able to verify independently.
  • Events that downstream systems rely on for reconciliation.

What usually does not

  • Pricing, fee schedules and business rules that change with commercial terms.
  • Compliance decisions that depend on off-chain data and human review.
  • Workflow state, such as approvals, retries and exception handling.

These belong in services that can be versioned, monitored and fixed on a Tuesday afternoon.

Upgradeability is a governance question

Upgradeable contracts are sometimes necessary. They are never free. An upgrade path is a privileged action, and it needs the same treatment as any other: a named owner, a quorum, a time delay where appropriate, and a public record. We design that governance before we write the proxy.

Invariants first

Before the first feature, we write the invariants the contract must never violate: total supply equals the sum of balances, restricted addresses can never receive, settlement cannot complete twice. These become property-based and invariant tests that run on every change. Features are then built to pass them, not the other way around.

The best contract we can deliver is the smallest one that still makes the system trustworthy.