What happens between a request and a result.
One piece of work, followed from start to finish, including the part where it stops.
Worked example
Worked example with sample data. Not a live system and not a customer account.
-
Request
A client cancels the 2:00 PM appointment. The booking tool records it.
-
Context
Three people on the waitlist asked for an earlier time. Two have agreed to receive texts. One opted out yesterday. It is within business hours.
-
ExceptionStopped
The person who opted out is taken off the list. She is left off before anyone is asked.
-
DecisionHeld
Offering a slot by text is on the list of things the owner approves. The owner gets one message: "2:00 PM just opened. Offer it to Maya and Luis? Reply Y to send, N to skip."
-
PermissionDone
The owner replies Y at 1:48 PM. The approval covers these two people and this slot, and it expires.
-
ActionDone
The offers go out. Whoever accepts first gets the slot. The others are told it is taken.
-
OutcomeDone
Slot filled at 1:52 PM. The request, the person left off, the approval, the two messages sent and the result are on record.
How we design it.
What stops the work.
| Situation | What happens |
|---|---|
| The action involves money, a refund or a price | HeldIt waits for a named person |
| The message is to someone who has not been contacted before | HeldIt waits for a named person |
| The approval has expired, or the details changed after approval | HeldIt asks again |
| The person opted out | StoppedThe message is not sent. |
| The system is not sure | HeldIt asks, or it hands the matter to a person. It does not guess |
| Something fails | StoppedThe work stops, a person is told, and the failure is recorded with its cause |
We run our own company this way.
-
A rule on paper cost us $13.24. Now it is a check.
A written rule said a check had to pass before we spent money on a job. Nobody had put the check in code. We paid $13.24 for work we could not use. The check now runs in code before the spending step.
-
AI agents told us work was done when it was not.
False "done" reports from AI agents are a recurring, documented problem in our own work. That is why we define done in writing before building, and why we built Kijestic Acceptance.
These are accounts of Kijestic's internal work. They are not client results.
Where each part stands
| Part | Status |
|---|---|
| Approval before sensitive actions, and a record of every action | Tested in our own staging with sends simulated. Not yet running for a client. Set up per project. |
| The owner's text line | Built and tested with simulated sends. Not yet running for an owner |
| Phone and text front desk | Built and tested. Not yet running for a client |
| Content review and claims checks | In operation for a skincare brand |
| On-device skin guidance | Live on a client's store. Guidance, not diagnosis |