Dotwave AI-proposed cleaning operations
Dotwave's AI never changes your data on its own. When you ask it to perform a cleaning task, it responds with a proposal — a card describing exactly what it intends to do — and waits for you to approve it. This confirm-before-acting design keeps you in control while still letting the AI do the heavy lifting of working out the operation. This article explains how proposals are presented, how to confirm or reject them, where to find the reasoning, and what happens once you approve.
How proposals work
When you ask the AI to do something — say, "fill nulls in age with the median" — it does not execute. Instead it presents the operation in an orange card that lays out three things clearly:
- What it will do, described in plain English so there is no ambiguity about the change.
- How many rows or values will change, so you know the exact scope of the impact before anything happens.
- A risk badge — Low, Review carefully, or High — summarising how confident the AI is that this is a sound move.
The card is the AI showing its work before acting. You get the operation, its blast radius, and its risk level in one place, which is everything you need to make an informed call.
Confirming or rejecting
Two actions sit on the proposal card. Clicking "Apply this →" executes the operation against your data. Clicking "Reject" dismisses the proposal and changes nothing. That is the whole contract: the AI never touches your data without your click. If a proposal is not what you meant, reject it and rephrase your request; the AI will propose again based on the new instruction.
Reading the reasoning
Dotwave deliberately puts the operation front and centre. The proposed operation card is shown first — what will happen, to how many rows, at what risk. The AI's underlying reasoning is one click away, under a "Why this recommendation" toggle below the card. This ordering means you see the decision before the justification: skim the card, and if you want to understand why the AI chose this approach, expand the toggle to read its full reasoning. You are never forced to wade through an explanation to find out what the AI actually wants to do.
A "High risk" badge means the AI had concerns but is proposing anyway (because you asked it to). Read the explanation before clicking Apply.
What happens after you confirm
Once you click Apply, three things follow. The operation is added to your cleaning recipe, so it becomes a repeatable step rather than a one-off edit. The audit trail logs it as "applied by ai", keeping a clear record that this change originated from an AI proposal you approved. And the profile updates to reflect the new state of your data — null counts, statistics, and outlier figures all recalculate — so any question you ask next is answered against the freshly changed dataset.
The result is a workflow where the AI proposes and you dispose. It handles the mechanics of figuring out the operation and quantifying its impact; you keep final say over every change, with the reasoning always available and every applied step recorded.