How TaxiD is engineered

A quote, a payment, an allocation, a collection, a failure — each is recorded as an event that cannot be edited after the fact. Work is then driven from those events: notifications, reconciliation, escalations and reporting all read the same history instead of keeping private copies of the truth.

That is what makes it possible to answer "what actually happened on this booking?" months later, in order, with references.

Rules such as when a vehicle may be offered, when full payment is required, or when an approval is needed are held as policies with a defined point at which they are evaluated. Changing a rule is a business decision recorded in the platform, not a code change nobody can trace.

Where a rule has not been decided yet, the platform says so and blocks the action rather than guessing a default.

We use AI to read across our own operational records — to summarise a situation, rank what needs attention, and draft a recommendation. It is a reasoning layer over verified data; it is not the source of truth and it does not approve money, contracts or access.

Answers are expected to distinguish a fact from an estimate, and to cite the records behind them.

Each capability has a register of controls, and a control only passes when a real test against real infrastructure has produced a recorded result. Passing a unit test, or a developer's confidence, is explicitly not enough — which is why parts of the platform are visibly marked as not yet certified.