Methods is the shelf of everything your business has worked out how to answer. You'll find it in the sidebar, alongside Reports and Scheduled.
A method is a saved way of answering a question — the figures to pull, the order to pull them in, and the checks to run on the way. Some are saved analyses the assistant runs when you ask something it has answered before. Others are saved report recipes that build the same report the same way every time. When a chat answer is saved as a report, the report recipe is saved with it so the report and its method stay together. When that report is published, the matching method is put into service with it.
What you see
The left side lists the methods available to you. Each one shows what it is for and whether it is in service. The right side explains the chosen method in plain terms: the steps it runs, the figures each step uses, and how the steps connect.
Two kinds of method sit side by side here. Some come with the product and are shared by everyone. Others were worked out on your own data and belong only to your business — a method of yours is never shared into another customer's account.
If a method has gone out of date it is flagged on this screen, and the assistant stops reaching for it until it has been re-checked. You'll see the reason it was set aside.
Methods that run a report
Some methods answer by running one of your reports. The Trial Balance method is the first of these: ask for the trial balance in chat and the assistant runs the same Trial Balance report you'd open from the Reports screen — same figures, same checks, same permissions.
A report method never works out numbers of its own. It names the report; the report's own definition — the one reviewed and approved on the Reports side — does all the calculating. That means a figure from a report method always matches the report screen exactly, and if the underlying data has shifted in a way the product hasn't confirmed is safe, the method refuses to answer rather than showing figures that might be stale. If your role isn't allowed to see the report, the method won't show it either.
Publishing a report puts its method on the shelf
When an administrator publishes a report from chat, the report's method is placed on this shelf in the same act — publishing is the approval, so there is no second sign-off to chase. From that moment, asking for the report by name in chat answers through the method, with the same figures and the same permissions as the report screen.
The method's recipe is written by the product itself from the report you approved — it simply names the report and hands over to it; nobody, including the assistant, writes any part of the recipe by hand. It is tested against the same rules as any other method before it is stored, and if the test fails nothing lands on the shelf and you are told why.
If the shelf placement fails for any reason, the report itself still publishes — chat will tell you plainly that the method did not land, and publishing the report again retries it. Publishing the same report twice never creates a duplicate: if its method is already in service, it is left exactly as it is.
What makes a method go out of date
Every method is built from a fixed set of trusted building blocks, and it names the exact version of each one it was written against. A method goes out of date when one of those building blocks is no longer available to it — it has been retired, withdrawn, or is waiting to be signed off again.
This is checked continuously rather than only when someone looks. The moment a building block is retired, every method in service that depends on it is set aside, with the reason naming the block. Nothing has to be noticed for the method to stop being used.
It clears the same way. If the building block is put back and signed off, the methods that depend on it return to service on their own. If you want to know where a method stands right now, Check it re-tests it against everything as it stands today and clears the flag if it holds up.
There is no way to wave a flagged method through, and that is deliberate. A method missing one of its building blocks would not produce a slower answer or a rougher one — it would produce an answer with a hole in it. The way forward is to restore the block or change the method.
Being set aside is also recorded, so it is always possible to see when a method went out of service and what took it out.
What an administrator can do
If you're an administrator, each method your business owns carries these controls:
- Check it — runs the method through the product's rules without changing anything. Safe at any time: it never puts a method into service and never alters one that already is. The result is a readable inspection report: every rule the method was tested against, in plain words, with a tick where it passed and the finding where it did not. The report is produced by the checking machinery itself, so it always reflects exactly what was tested.
- Approve and put in service — signs the method off and makes it available immediately. There is no second step; approving is putting it live.
- Take out of service — withdraws a method so the assistant stops reaching for it. This is never permanent. A withdrawn method shows a Bring it back button, and bringing it back returns it to service exactly as it was.
- Draft the amendment — describe a change in plain language and the assistant drafts it for you. The assistant only proposes: its draft is tested against the same rules as any method, and if it passes it is saved as a new version, as a draft awaiting your approval. The version in service is never touched, and a draft that fails the rules is not saved at all — you are shown why instead. The draft records that the assistant wrote it, what you asked for, and which version it came from.
Approving, withdrawing, restoring, and amending are recorded against your name, so there is always a record of who changed what and when.
The methods that come with the product are the one exception. They are governed by the platform, not by any one business — your business can use them freely, but approving or withdrawing one is not done from here, and the screen says so instead of offering buttons that would be refused.
Why some people see fewer methods
Everyone except administrators sees only the methods that are actually in service, signed off, and still current. A method that has gone out of date is set aside until it is checked again. That matches the promise made everywhere else in the product: nothing reaches you until a person has approved it. Administrators see everything, including drafts and withdrawn methods, because deciding what goes live is their job.
If the shelf is empty
An empty shelf simply means no method has been saved yet. Methods appear here once a way of answering a question has been worked out and kept. Ask the assistant a question in chat; when it settles on an approach worth keeping, it is offered for approval and then shows up here.