Doing it, step by step, answered in full.

No. Four of the seven channels carry the same score whatever audience detail you add — audio 65, addressable TV 55, DOOH 50 and cinema 40 — because their targeting vocabulary cannot express more than that however good the input. The one line of a brief that does move them is geography: naming a city the UK table cannot map costs fifteen points on every channel, floored at ten. Only CTV, search and social move, and they move on specific inputs rather than on length: marketplace segment matches for CTV, intent topics for search, interest terms for social.

There is no universal figure, and anyone quoting one is guessing. It depends on your baseline variance, your conversion volume, how many independent markets you have and how large an effect you would need to detect. That is precisely what the power analysis computes before the test runs, which is why it runs first. A multi-catchment retailer can often power a test that a single national campaign cannot at any budget.

No. Nothing has run live, and a geo-lift is graded causal only under conditions sandbox data cannot meet by construction. The engine, the power gate and both refusals are real and tested; the result is not yet evidence about a real campaign, and no lift figure appears anywhere on this site as though it were.

Not in the way client money in a segregated account would be, and it is worth asking rather than assuming. A wallet balance is a claim on Tenhaw LTD. Reservations against live campaigns are held until spend reconciles to the provider invoice, which limits how far ahead of delivery the balance runs, but it does not make the balance bankruptcy-remote. Fund what you are about to spend rather than a quarter in advance, and raise it in the contract if it matters to your finance team.

Nothing that commits money. Eleven consequential operations — approvals, provider-attempt resolution, mandate replacement and revocation, mission closure, workflow acknowledgement, asset-job resolution, dossier saving and verification, and issuing or revoking a live execution grant — are human-only signed REST actions with no agent tool at all. What an agent can do is prepare, propose, draft and surface, all of it recorded. The design intent is that an agent removes work rather than removing accountability.

No, and it does not claim to. It predicts what the clearance bodies and publishers will flag, using deterministic rules against the UK codes. It cannot read a substantiation document and says so explicitly: where you declare that you hold evidence, it records the declaration beside the claim and states that it has not assessed quality. The bodies decide, and they want the documents.