The design brief

What is there today. A prototype, in the lab, marked unbuyable on its own face.

The one the store most needs and has never had. A level page argues about what a policy is; a product page shows the thing. Amazon-shaped: description left, price and buy on the right, a screenshot carousel, specifications, and a reviews section that is honest about being empty.

It has to carry

• A media carousel with many slots, of which only some have a real image — the rest are labelled frames naming what belongs there.
• A right-hand rail that stays with the reader: price at 42px, delivery estimate and the clock it runs from, the buy button, and for the top two levels the deposit split (a fifth now, the rest on delivery).
• A specifications table — product code, what it is, how many shapes, delivered in, licence, formats, whether an account is needed.
• A reviews section as an empty state.
• An edition switcher: the same product described five ways, one per audience, switching in place without a page load.

The hard part. The edition switcher has to change the argument without changing the product. Five audiences, one SKU, one price, five reasons.

What is there today. Fourteen rows in four groups across five columns, including a free one.

Features down the left, five columns across — free, then the four levels. It has to work as a page, as something printable, and as a thing a reader screenshots and sends to somebody else.

It has to carry

• Five columns including the free one, which comes first.
• Row groups with headings, so a reader can find the row they care about.
• A phone version that does not become a horizontal scroll nobody uses.
• A print-shaped version: this is explicitly wanted as a brochure and a PDF.

The hard part. Five columns at 390px. Most comparison tables give up here and become five stacked cards, which loses the comparison — which is the entire point of the page.

What is there today. One per level, plus two add-ons.

The page a buyer lands on from a card. What this level is, what arrives, what does not, who does the work, and the way to buy it. Shorter than the product page and pointed at one decision.

It has to carry

• What arrives and what does not, as two lists rather than a paragraph.
• For the top two: who does the work, with the professional's name and face.
• The delivery estimate and the clock it runs from — from your payment for the first two, from your reply for the top two.
• The buy button, and the deposit split where there is one.

The hard part. 'What does not arrive' has to be as prominent as what does, and has to not read as an apology.

What is there today. An order built in the reader's own browser, no account, nothing stored anywhere else.

Everything up to the moment the reader leaves for the payment provider. No form fields — the provider's page collects the name, the email and the card, and this store never sees any of them. There is no input, textarea or select anywhere in this site's output and a check holds that line, so draw a cart made of buttons.

It has to carry

• The order reference, generated in the browser and shown on the page.
• A discount code applied and its effect visible.
• The hand-off explained before it happens: what the provider will ask for, and what this store will never see.

The hard part. A checkout with no form. It is a real constraint, not a stylistic one, and it is the thing this store can say that most stores cannot.

What is there today. One per level.

The instruction was 'when you have sold a vault, you should just see the vault'. So the screen after payment is not a receipt with a link on it — the thing bought is on the screen, open, with the receipt beside it.

It has to carry

• The vault itself, embedded and open, above anything congratulatory.
• For a buyer whose own vault does not exist yet, a published example labelled as an example in its first sentence.
• What happens next and when, with the date.
• For the top two levels: who is doing the work, and how they will make contact.

The hard part. The reader has paid and is now looking at somebody else's example vault. That has to be obviously an example without feeling like a bait and switch.

What is there today. Five of them, one per audience.

Where a persona door leads. The same four levels, led with the one that fits this reader and the rest quieted rather than hidden.

It has to carry

• Nothing hidden. A check refuses display:none on an offer in an audience view — every level stays reachable from every audience, and what changes is which one leads and what is said about it.
• The reader's own question in their own words, at the top.
• One level foregrounded, the others at reduced emphasis and still clickable.

The hard part. 'Quieted, not hidden' is a visual problem. Opacity alone reads as broken; a different treatment entirely reads as a different product.

What is there today. Two prototypes in the lab — a canvas and a sequence.

A free page, and the one the project lead expects to use most in conversation. The agent sits in the centre; the things it can reach sit around it — a database, the internet, part of the internet, a file system, the ability to send email. The reader clicks or drags them onto the agent and builds up a description of what it can actually do.

It has to carry

• The agent at the centre and its capabilities around it, clickable and draggable.
• Twenty-three capability primitives on a verb.object.reach grammar, which exist already and are not to be invented.
• A running description of what has been assembled, in plain words.
• An export at the end: the description, in a form that can be handed to a model session with a template vault, which then produces the buyer's own vault.
• It has to work on a phone and in a room, over somebody's shoulder.

The hard part. This is a toy that has to produce a document. Show at least two different interactions for it — the canvas and something else entirely — because which one works in a live conversation is not known yet.

What is there today. One, for the person doing the work today.

Who does the £500 and £1,500 work. Built for a list: more professionals are being recruited, so it is a template with one entry today, not a personal page.

It has to carry

• Enough for a buyer to decide whether this person's signature is worth £1,500.
• Every biographical line sourced, because a check refuses one that is not.
• An index that works with one entry and with twelve.

The hard part. A page of one person's credentials that does not read as a CV.

What is there today. Fifty-two claims, each with a state and a date, cited from the pages that make them.

Every factual claim this site makes, the state it has earned, and the date. Pages cite a claim and the chip links back here. It is the store's whole argument for being believed, and it currently looks like a table.

It has to carry

• Every claim state visually distinct and legible at chip size. The page lists them; two currently share a class and render identically.
• The join in both directions: from a claim to the pages citing it, and from a page's chip to the claim.
• It should be possible to skim for the claims that are weakest, not just the ones that are strongest.

The hard part. Making a page of caveats feel like confidence rather than hedging.