Conform
Posts the refund as a ledger event.
Decreen shows the reviewer which law the change follows. Nobody argues it again.
Decreen makes already-made engineering decisions reusable for humans and coding agents.
Runs in your execution environment. No repository access required.
Before Decreen
After Decreen
BeforeA senior explains the same decision in review. Then again next week.
AfterThe decision is approved once. Review starts from the decisions a change touches.
BeforeWhy refunds work this way lives in one senior’s head.
AfterIt’s written down as a law. Your team and your agents read the same one.
BeforeAn agent’s pull request quietly changes a rule. Nobody decided that.
AfterA change that needs a new decision is flagged. A senior rules on it.
You don't write them by hand.
Agents turn each ticket into a plan against your decisions.
A reviewer approves the plan before any code is written.
Agents build it. Not everything in the plan always lands.
Judge checks the result against every decision, including ones nobody planned to touch.
The reviewer rules on each mismatch, and the verdict goes back into your decisions.
A decision your team approved (a “law”)
Refunds are posted as ledger events, never as balance adjustments.
Orders may still call adjustBalance() for goodwill credits. A lint rule can't tell a credit from a refund.
A coding agent opens a pull request: “Add order cancellation”.
Conform
Posts the refund as a ledger event.
Decreen shows the reviewer which law the change follows. Nobody argues it again.
Violation
Calls adjustBalance() to refund the order. Build and tests pass.
Decreen flags the conflict and cites the law. It doesn’t block the merge. The reviewer decides.
New decision
Offers store credit instead of a refund. Is that a refund or a credit?
No approved decision covers this. A senior rules once. The ruling becomes a law, and the next similar change is checked against it.
A linter says pass or fail. It never says: nobody has decided this yet.

Alina Powała, co-founder of Decreen
PhD in Computer Science (multi-agent systems). Building software since 2008. You talk to me, not a sales team.
LinkedIn