Product
Admin
The verdict engine, attention states and the surfaces an operator works from.
What the surface does
Admin is where an operator works. It opens on a verdict rather than a wall of metrics, and branches into one area per part of the business, each reading the same resolvers the rest of the product reads.
The verdict engine
The home screen synthesises one state for the day. It has exactly two shapes.
- Quiet. Nothing changed today; there is nothing here that needs you. Quiet is a stated conclusion carrying the receipt for why we believe it — the age of the evidence, and when the price was last checked. It is not an empty screen and not a fabricated all-clear.
- Attention. A ranked list of things that changed or need a decision. Each item carries the receipt that justifies it: the resolver that produced it and the underlying values. An item that cannot name its receipt is not emitted.
A dashboard that always finds something to report is a metrics dump wearing a verdict’s clothes, so this one is allowed to return nothing — and on a day when no evidence refreshed, nothing is both the honest answer and the most common one.
Ranking is fixed rather than discretionary. A withdrawn or stalled offer outranks everything, because it is the thing that changed most. Next comes a price whose latest re-check did not confirm, which means you are being shown an ageing figure. Then evidence that has aged past the point where its selling angle may still be asserted.
A stall is never allowed to become quiet. Where a stall has no written diagnosis, the item is still raised and marks itself undiagnosed. Silence about a stall is worse than an imprecise description of one, because the operator stops looking.
Stalls are diagnosed, not flagged
Every stalled state resolves to three lines — what happened, why, and what happens next — and the last line names who acts. Most name us. That distinction is what stopped refusal screens from offering a “try again” button for states the operator cannot affect. An unrecognised state has to be added to the diagnosis table deliberately; there is no generic fallback, because “something went wrong” is exactly the vague flag this structure replaces.
The areas
| Area | What it holds |
|---|---|
| Home | The verdict, attention items, key figures, funding state and the next actions |
| Product | Identity, exclusivity status, how the market moved since you claimed, and the analyse tool |
| Growth | Organic briefs and approvals, the ad artefact, referral progress |
| Website | Store truths, preview, section order with reasons, domain result |
| Supplier | Price, lowest observed listing, margin basis, expected unit cost and profit, listings |
| Activity | Orders, satisfaction sample, support agent status |
| Settings | Account and claim state, referral link, and the engine’s stated reasons |
One number, one definition
Every figure the dashboard shows is read from the same function that serves it elsewhere, and each carries the name of that source. The score, the freshness, the price and the headline are single definitions, so two surfaces cannot quietly disagree — and the test suite can assert agreement mechanically rather than comparing numbers that happen to match today.
Cost before action
Every control that changes something declares what it costs and whether it can be undone before you use it. Reversible surfaces are named explicitly; anything that spends money or reaches outside the system is not, and says so first.
Honest limits
- The verdict reports on your own claimed products. It is not a market monitor.
- Revenue and order figures come from your own founder-scoped ledger. An empty ledger reports zero as a fact, not as a chart.
- Campaign launch and domain purchase are initiated from here, by you.
For team use, see enterprise. For what the agent may do without you, product security.