Stories A told story · first draft 2026-11-18. Read, not built. Every state below is read from origin/dev e0cee5383b and none from production
Checkout
Every shop on ONE gets one checkout engine whose levers raise order value and conversion, each lever measured, none of them deceiving the buyer.
The seven beats
- Groups
- A shop on ONE, inside the group that owns it, and the eight doors to its checkout
- Actors
- The buyer, a person or an agent; the seller; the tuning agent; the person who approves a price
- Events
- A buyer presses Buy and the sheet slides in, uncounted
- Things
- One engine: a clear total, both rails, one honest offer before payment and one after, each measured
- Paths
- door → started → bump → paid → offer after → refund window; marked, warned, faded
- Outcome
- Open: the mint is real; the counting and the offers are not
- Learning
- Open. Nothing learned yet
Open the sales room and look at the funnel. Its fifth stage is called Checkouts started, and it holds no number. It says why in one line:
Nothing posts a checkout event to /api/storefront/event yet.
That line is components/sales/sales-math.ts:185, and it is true. The
store's own stats do count checkouts (storefront:stats,
lib/resolvers/commerce.ts:166). They count rows in storefront_events
where the event is checkout. Nothing on any page writes one. A
separate intent signal, ui:checkout:open, fires when a card form
mints inside EmbeddedCheckout (components/pay/EmbeddedCheckout.tsx:181),
which the chat sheet's card rail, the cart drawer and the buy button use.
It does not fire on the shop page, on a USDC rail, or when a buyer
arrives and leaves before choosing a rail, and it lands in the click
stream, not in the store's table.
So ONE can take a buyer's money eight different ways, and it cannot say how many buyers reached a checkout and left. Every lever in this story (an order bump, an offer after payment, a cart line, a member price) is a claim about that number moving. A lever that cannot be measured is a guess with a button on it. This story starts with the count.
Status: red at the making, by law. None of the nine rows in the schedule above is built. Each goes green when its test does.
What Tony asked for
Tony asked for the lot: the slide-in checkout, the design, conversion, upsells, order bumps, cross-sells, and everything ClickFunnels, the hotel and airline sites and the Shopify apps use to raise order value.
This file takes all of it and sorts it into two piles. One pile is levers that work because they are true: an add-on the buyer wanted anyway, a total with nothing hidden, a saved card that means one press. The other pile is levers that work because the buyer was misled: a timer that resets, "2 left" with no stock count, a box ticked for them. ONE builds the first pile and refuses the second, in code. § The line says where the line falls and why it is already law in the places we sell.
The checkouts ONE has today
Eight doors and one mint per rail. Every card payment goes through
mintConnectCheckout (lib/storefront-session.ts:746), which reads the
seller's Stripe account and fee from the seller's own row and never from
the request. Every USDC payment goes to a wallet the seller holds. ONE
holds neither.
a buyer presses Buy
│
├─ the chat sheet ─────── ChatCheckout
├─ the shop page ──────── CheckoutPage
├─ the cart drawer ────── CartDrawer
├─ a buy button, a block BuyCTA · CartCheckout
├─ a funnel step ──────── CheckoutStep → PayPanel
├─ a payment link ─────── checkout:link
├─ a workflow sell step ─ checkout:create
└─ ChatGPT ────────────── checkout_sessions (ACP)
│
▼
mintConnectCheckout card, to the seller's Stripe
the crypto rail USDC, to the seller's wallet
│
▼
an orders row → the receipt (a receipt prints only
from a settled row)
Surface by surface
| Surface | Where | What it does | What it lacks |
|---|---|---|---|
| The slide-in sheet | lib/home/home-chat.ts:282 openOfferChat slides the chat in (wide, or full screen on a phone) and mounts components/chat/pay/ChatCheckout.tsx:104 |
The one payment surface allowed in a transcript. A model names a price id; a trusted component does the mint, so no amount ever comes from the model. Sells credits, products, services and subscriptions; subscriptions are card only (offerableRails, ChatCheckout.tsx:191). Four doors into it: a #buy= link, the shop grid, a product card, the cart card |
No count of arrivals. No order summary with a total before the rail buttons. No add-on line. No offer after payment. On the receipt, nothing comes next |
| The home specimen | components/home/CheckoutSpecimen.tsx:65, words from SALE_DEMO in lib/sale/showcase.ts:619 |
A real Stripe test sale of the $100 strategy call in the demo shop. The test card types itself onto a card with no inputs. Refuses outright on a live key | A demo by design. It sells the call at its showcase price; the call is live: false (showcase.ts:73) |
| The shop page | components/storefront/CheckoutPage.tsx:100, route pages/u/[slug]/shop/checkout.astro:29 |
Order summary, Stripe's card form inline beside a USDC tile, both on one screen. Buy goes straight here with buy=1, so it takes 2 presses (5 with the cart). The tile appears only when the USDC rail is really open |
No bump. No offer after payment. Apple Pay is open (the domain is not registered for the Connect account, per the 2026-11-14 lane) |
| The cart drawer | components/storefront/CartDrawer.tsx:155, cart store lib/cart.ts:185, receivers cart:* (receivers.ts:5802) |
The server cart: survives a refresh and a device change, re-priced at mint, N lines into one payment by its cart id. Heading fixed, lines scroll, total and button pinned. Discount codes bind code and seller together (lib/resolvers/storefront/discounts.ts) |
No free-shipping line (the threshold exists: lib/storefront/shipping.ts:141). No "goes with" line. No discount field in the drawer itself |
| A buy button | components/pay/BuyCTA.tsx:164 → components/pay/EmbeddedCheckout.tsx:87; product page components/commerce/evaluate/ProductBuyPanel.tsx |
The buy affordance any page block renders. Mints only with a seller and a bound price. The only door that fires ui:checkout:open |
Same gaps as the sheet. ui:checkout:open fires here only once the card form mints, not on arrival |
| The page-block checkout | components/commerce/convert/CartCheckout.tsx (BasketPanel, line 431) |
The cart and checkout blocks a seller drops on a page; both rails in place, nothing typed is money | Same gaps |
| A funnel step | components/funnel/CheckoutStep.tsx → components/pay/PayPanel.tsx |
Takes a payment, then moves the funnel on through funnel:advance |
Sells one of a fixed list of plans (components/pay/PriceCards.tsx:46). A seller cannot put their own product here |
| The funnel upsell | components/funnel/UpsellStep.tsx:47 |
A button that says "One-tap add to my order" and a "No thanks" | It charges nothing. Both buttons send the choice to funnel:advance (lib/resolvers/funnel.ts:88), which runs the next step. The "payment method" it reuses is a plan id (lib/funnel/types.ts:55), not a saved card, and no one-time payment on ONE saves a card for a later charge (setup_future_usage appears in no code in the repo; subscription mode keeps a card only for its own renewals) |
| The upsell block | lib/puck/archive/config.archived.tsx:985, placed by lib/puck/template-pages/order-confirmation.ts:85 |
A heading and a link | It draws. The link adds nothing to any order, and the template types "save 20% on your next bottle" with no price behind it, against the standing ruling (§ The line) |
| Cross-sell and shop-the-look blocks | components/commerce/offer/CrossSellView.tsx:70, ShopTheLookView.tsx |
Real products bound by price id; the button adds real lines to the server cart | The pairing is typed by the seller. Nothing reads which products are really bought together |
| Payment links | checkout:link (receivers.ts:11651), mintCheckoutLink (lib/storefront-session.ts:1148), drawn by components/home/PayLink.tsx:74; checkout:notify posts one into a run's thread |
A link for any price, sent on any channel, counted by channel through via |
No bump: a link names one price. No offer after payment |
| The workflow sell step | checkout:create / checkout:resume (receivers.ts:9681, 9705) |
A run waits at a sell step until the buyer pays, then the webhook wakes it. Card only |
Nothing after the pay except the next step a person wired |
| ChatGPT | pages/api/checkout_sessions.ts, lib/storefront/acp/session.ts:592 |
ONE's agentic checkout door for ChatGPT (text/sell-in-chatgpt-plan.md) |
Outside this story's first waves; an offer there is data in a protocol, not a screen |
| Not counted as doors | components/pay/BuyCredits.tsx, components/pay/PayCard.tsx; lib/shop-checkout.ts:49 |
Credit packs, the in-place checkout block, and a course or playbook redirect | Credits settle on a balance and join the count in wave 3. shopCheckout has no caller on origin/dev: an orphan, to be wired or removed |
| Sold | lib/sold/sold-offer.ts:12 (price_sold_humans, price_sold_community_monthly), enrol through components/chat/offers/CourseOfferCard.tsx:149 → the sheet |
The $100 course is the one live Sold product. Words are fixed; price and open state are read | The community's card checkout fails "No such price" (text/one.md:389): a fix in review, owned elsewhere (goal 1). The 2026-11-14 probe found the course asks a stranger to sign in before paying |
| What the store watches | lib/store-admin/checkouts.ts:4 (abandoned = an unordered cart quiet for 3,600 s), page pages/u/[slug]/studio/checkouts.astro |
The seller can see abandoned carts | Read only. No reminder is sent ("a recovery email is a later append", line 2) |
| The words | lib/sale/showcase.ts |
One source for every surface that pictures a sale: four offers, two rails, six channels | Pictures only. Three of the four offers are live: false |
Two more facts sit under all of it. A was-price is already guarded:
compareAtPrice (lib/pay/checkout.ts:647) shows a struck-through price
only when the seller's row has a higher one, never zero, never a default.
And the prompts exist without the wiring: six elevate-upsell-* and
cart-abandonment skills write the words for a bump, an offer after payment
and a reminder. Their agent, upsell-strategist, subscribes to
upsell:build-needed and emits upsell:assets-ready. Neither is a
receiver. The words are written and nothing puts them on a checkout.
What the others do
What the funnel builders, the travel sites, the airlines and the Shopify apps do to raise order value, in their own terms. Sources are listed at the foot. Vendor numbers are their own, unaudited.
Funnel builders (ClickFunnels and its kind)
| Pattern | What it is |
|---|---|
| Order bump | One checkbox on the checkout, beside the main product, before payment. Priced low; one vendor's rule of thumb is 10–30% of the main product [3] |
| One-click upsell (OTO) | A one-time offer shown after payment. It charges the card the buyer just used, with no new card entry. ClickFunnels shows its checkout in "OTO mode" when the buyer arrives from a purchase [1] |
| Downsell | A smaller offer, shown only to a buyer who said no to the upsell just before it [4] |
| Tripwire | A small first purchase that turns a visitor into a buyer, so the next sale is to a customer |
| Continuity | A recurring product after the first sale: a membership, a community |
| Two-step order form | Name and email first, payment second, so a buyer who leaves after step one can still be reached |
Hotel sites (Booking.com, Agoda)
| Pattern | What it is | Honest when |
|---|---|---|
| Member prices | Booking.com's Genius: a free account unlocks a discount at level 1; more stays unlock breakfast and upgrades on some properties [8]. Agoda has VIP tiers and app-only "insider" prices [9] | the member price is real and the same for every member at that level |
| Free cancellation | The refund term, said beside the price | the term is the term |
| Pay later / at the property | Agoda offers pay now, book now pay later, or pay at the hotel [9] | the later charge is the same amount, said now |
| Social proof | Review scores; "booked N times" | the count is read, with the period |
| Scarcity | "Only 1 room left" | it is the real count. In 2019 the UK's CMA took binding commitments from Booking.com, Agoda, Expedia, Hotels.com, ebookers and trivago to stop giving "a false impression of the availability or popularity of a hotel" [10] |
| Price anchoring | A struck-through price | it compares like with like. The same commitments stopped comparing weekday rates with weekend rates [10] |
| Full price up front | The sites committed to show compulsory costs in the first price [10]; in the US the FTC's fees rule now requires it for hotels and live-event tickets from 12 May 2025 [12] | always |
| Saved details, progress steps | A signed-in guest's details filled in; "step 2 of 3" | always |
Airlines
Ancillaries are the extras sold beside the fare: bags, seats, priority, insurance. IdeaWorks estimates them at $157 billion worldwide in 2025 [7]. Bags are the largest single source; seat fees rival them at many carriers, and at United seat fees ($1.3 billion) passed bag fees ($1.2 billion) in 2023 [7]. The lesson for ONE is timing: each extra is offered at the moment it matters, a seat when the map is open, a bag when the trip is booked. The warning is unbundling. A fare that looks cheap because the rest comes later is drip pricing, and ONE does not do it.
Shopify apps
| App kind | What it does |
|---|---|
| After-payment offers (Zipify One Click Upsell, AfterSell) | One offer right after checkout, one press to accept, a cheaper downsell on a no. Zipify reports a 16.2% average take rate across 50,000+ stores, its own figure [5] |
| Thank-you page (ReConvert, now Upsell.com) | The receipt page becomes a page with offers, a survey and a reorder button [5] |
| Personalised offers (Rebuy) | Reads the catalogue, the cart, the customer and the stock to choose what to offer in the cart, at checkout and after [5] |
| Cart drawer (Upcart, iCart and others) | The cart slides in with a free-shipping bar ("$13 away"), a suggested item that closes the gap, and tiered rewards. One vendor claims 10–20% higher order value, its own figure [6] |
| Bundles and subscriptions (Bold) | Buy-together prices, and subscribe-and-save |
What makes a checkout convert at all
Before any lever, the checkout itself. Baymard puts average cart abandonment at 70.22% across 50 studies from 2006 to 2025, finds that checkout design alone is often the only reason a buyer leaves, and estimates that a large shop can gain up to 35% in conversion from checkout design changes alone [11]. The usual causes are costs that appear late, a forced account, a long form and a payment method the buyer does not have. ONE already does two of these right: the full price comes from the server, and the shop page needs two presses, not five. The other two are open: a signed-in step before Sold, and Apple Pay.
The line
ONE uses no lever that works by misleading the buyer. That is not a new ruling. It is three old ones, said once for the checkout:
text/pay-components-plan.md§ The blocks: "a claim with no data source is deleted, not defaulted." That pass already turned a frozen countdown into "a real deadline; withdraws the CTA at zero", a free-shipping bar drawn from a basket it never saw into one read from the live cart, and a typed "Save $96" into a saving "derived or absent".text/books-offer.md:193: "Honest scarcity: none. Don't invent any."text/product.md:119: "Not manufactured scarcity — just the truth that the first ones in get the best terms."
The law agrees, where ONE's sellers sell. In the EU a paid extra needs the buyer's express consent, and a pre-ticked box is not consent; a buyer charged that way gets the money back (Consumer Rights Directive, Art. 22) [13]. "Falsely stating that a product will only be available for a very limited time" is banned in all circumstances (Unfair Commercial Practices Directive, Annex I, point 7) [14]. The UK enforces both. A crawl of 11,000 shopping sites found 157 deceptive countdown timers on 140 sites, timers that ran out on offers that did not end [15].
Two notes on reach. The FTC's fees rule binds only hotels and live-event tickets; ONE keeps its principle (the full price first) everywhere, because it is right, not because the rule covers us. The FTC's click-to-cancel rule was vacated in July 2025 and is being rewritten; ROSCA still applies to anything that renews [16]. ONE is a Hong Kong company, and a seller on ONE answers to the law where its buyers are.
Honest, and how ONE proves it
| Lever | Allowed when | The row that proves it |
|---|---|---|
| Order bump | unticked; its own price shown; same press to skip | the bump's product price row |
| Offer after payment | one, then at most one smaller; skip is one press | the offer's price row |
| One-press card charge | the buyer agreed to save the card, in words, before paying | the saved-card consent on the order |
| Bundle price | the saving is the difference of real prices | both price rows |
| Was-price | the seller's row holds a higher price | compareAtPrice (lib/pay/checkout.ts:647) |
| Member price | every member at that level gets it | the group's membership |
| Deadline | a real end: a cohort start, an event, a launch close; the offer goes at zero | a deadline row |
| Seats left | a real cap, e.g. Sold's founding group of 20–50 | the cap and the member count |
| Stock left | stock is tracked | product_variants.stock; NULL reads "Not tracked" |
| "N bought this" | read from orders, with the period | a count of paid orders |
| Free-shipping line | the seller has a rate with a threshold | freeShippingThreshold (lib/storefront/shipping.ts:141) |
| Refund term | it is the seller's real term | the seller's policy |
| Reminder of a left cart | the buyer gave an address and agreed to email | the consent door (text/crm.md card 7) |
Refused, and why
| Pattern | Why it is refused |
|---|---|
| A pre-ticked add-on | not consent; refundable by law in the EU [13] |
| A timer with no deadline, or one that resets | banned outright in the EU [14] |
| "Only 2 left" with no count | the CMA's 2019 commitments [10]; books-offer.md:193 |
| "12 people are looking at this" | a popularity claim with no read |
| A cost that appears at the last step | drip pricing; the CMA [10], the FTC fees rule as principle [12] |
| A subscription inside an add-on | a renewal the buyer did not choose; ROSCA [16] |
| Shaming the no ("No, I like paying more") | pressure, not information; the no is plain words |
| A was-price that was never charged | compareAtPrice refuses it already |
| Sold-out items shown to push others | the CMA's 2019 commitments [10] |
| A third offer after two noes | the buyer said no twice |
| Persuading an agent buyer | an agent buyer gets the offer as data: product, price, terms. No copy aimed at it |
Each refusal runs at save, in code (deliverable 2), so it cannot be argued round by a prompt, a seller or an agent tuning for a number.
The engine
One engine every shop on ONE gets. The eight doors stay; each becomes a view of one checkout. They share one summary, both rails where both are true, the bump line, the receipt and the offer after it.
What the buyer sees
┌──────────────────────────────────┐
│ Sold, the course $100 │
│ $100, once │
│ ──────────────────────────────── │
│ [ ] add: Sold, the book $20 │
│ the playbook, every chapter │
│ ──────────────────────────────── │
│ Total $100 │
│ paid to ONE · nothing renews │
│ │
│ [ Pay by card ] [ Pay with USDC ]│
│ [x] save my card for one-press │
│ buys at this shop │
└──────────────────────────────────┘
The total is the first number, and it is the whole number. The bump is one
line, unticked, with its own price, and ticking it changes the total in
place. The card box is the buyer's choice, said in words. The book line is
an example: the book is live: false today (showcase.ts:68), so on
Sold's real checkout that line does not appear (deliverable 9). The box
says nothing about the community month and names no refund term: the
community waits on its own checkout (text/one.md:470), and text/sold.md
states no refund term. A term appears when the seller has one.
After payment
paid
│
▼
the receipt, printed from the row
│
├─ one offer, if one is set and live
│ │
│ ├─ yes, card saved ──► one press,
│ │ charged off-session
│ │ └─ bank asks ──► a normal
│ │ checkout
│ ├─ yes, USDC ────────► a second
│ │ signature
│ └─ no ───────────────► one smaller
│ offer, or done
▼
done. Nothing after two noes.
Two facts limit "one click", and the story keeps both:
- A card is one press only when it was saved with consent. Stripe saves
a card for later with
setup_future_usage: off_session; the later charge is a server-side payment withoff_session: true. A card saved through Stripe Checkout does not reappear on a later Checkout, so the one-press charge has to be its own server call, and when the bank wants the buyer to confirm, the buyer comes back on-session [17]. On ONE that call is proposed and not in the registry. Before a receiver is written, thereceiverskill's first question applies: does a path already do it? The closest is a secondcheckout:create, which works today at the cost of the buyer entering the card again. - USDC has no one click. Every payment is the buyer's own signature. The offer after a USDC payment says plainly that it is a second signature. A subscription stays card only, everywhere.
The levers
| Lever | Where | First number it moves | Guard number |
|---|---|---|---|
| Order bump | the summary, before pay | order value | conversion |
| Offer after payment | the receipt | order value | refunds |
| Smaller offer | after a no | take rate | refunds |
| Bundle | the product and the summary | order value | conversion |
| "Goes with" | the cart drawer | items per order | conversion |
| Free-shipping line | the cart drawer | order value | conversion |
| Member price | the summary, signed in | conversion, repeat orders | margin |
| Saved card | the summary | conversion of the second order | none |
| Deadline / seats left | the summary, only when real | conversion | refunds |
| Reminder of a left cart | email, inside consent | recovered orders | unsubscribes |
Every lever has a guard number. A bump that lifts order value and lowers conversion is a loss with a nice average. An offer after payment that comes back as refunds sold nothing. The read for each lever (deliverable 6) shows both, against a holdout: a share of buyers, set by the group, who never see the lever, so its lift is a difference and not a hope.
The numbers
| Number | What it is |
|---|---|
| Checkouts started | buyers who reached a checkout, per door |
| Conversion | paid orders ÷ checkouts started, per door |
| Order value | revenue ÷ paid orders, with and without each lever |
| Take rate | taken ÷ shown, per lever |
| Revenue added | the lever's lines on paid orders |
| Refunds in window | refunds on the lever's lines before the refund term closes |
| Lift | the lever's group minus the holdout, on conversion and order value |
| Revenue per visitor | revenue ÷ visitors, the one number all levers share |
Each with its source and the time it was read. A number with no source says
so; it is never shown as zero (the sales room's rule, sales-math.ts:172).
Where it fits in the six
No new entity and no new thing-type. The enum already holds product,
service and sku (schema/one.tql:322).
| Dimension | What the checkout puts there |
|---|---|
| Groups | The shop's group holds the dials: which levers are on, the holdout share, the refund window, the guard limits. Two groups can run them differently, the way they already set their own fade rate |
| Actors | The buyer, a person or an agent with a wallet and a limit; the seller; the agent that tunes the levers; the person who approves a price. Same verbs for each |
| Things | An offer is a product the seller already sells. A bump is not a new kind of thing; it is a product shown in a new place |
| Paths | A lever is a path between two products, tagged with its kind (lever:bump, lever:after, lever:smaller, lever:goes-with, lever:bundle). A paid take marks it, a decline or a refund warns it, and an unshown pairing fades. The strongest path is the offer shown next |
| Signals | Checkout started, offer shown, taken, declined, paid, refunded. The paid ones and the refunds are transactions |
| Learning | Each test is a hypothesis: "this bump lifts order value in this shop without lowering conversion". The holdout confirms or rejects it, and the result is written back where the next shop can read it |
Mark on a paid take, never on a click. The event vocabulary already rules
that reaching a checkout is intent, not a sale
(lib/event-vocabulary.ts:246).
Agents and people
What an agent may do. Turn a lever on or off; choose between offers the
seller approved; set the order of offers; start a test with a holdout;
write the words through the elevate-upsell-* skills; read every number.
Each change is a signal, and the lever's outcome marks or warns the change.
refine-analyst reads the numbers, upsell-strategist writes the offer,
cro routes, and the seller keeps the switch.
What only a person does. Change a price. Create a discount. Approve a
new product as an offer. Lift a guard limit. A tuning agent that wants
any of these sends human:notify and waits, the same way a workflow's
human step waits.
What nobody does. Switch off the honesty law. It is a check at save, not a setting.
An agent as buyer. Agents with wallets buy on ONE. An agent buyer sees an offer as structured data (product, price, terms, refund) and decides under its owner's limit. A lever that only works on a person's impatience does nothing to an agent. That is a good test of whether a lever is honest.
What we have, sorted
Legend:
●built and read in the code on origin/dev, not run this session◐partly there○a prompt, a picture or a design only·not started
Nothing here is marked as seen on production.
A · The checkout itself
| Power | State | Where |
|---|---|---|
| One mint for every card door, fee and account from the seller's row | ● | mintConnectCheckout, lib/storefront-session.ts:746 |
| Both rails, USDC only where it is really open, card-only subscriptions | ● | offerableRails, lib/pay/checkout.ts:753; CheckoutPage.tsx:100 |
| A server cart that survives a refresh and pays N lines at once | ● | cart:*, lib/cart.ts, CartDrawer.tsx:155 |
| Discount codes bound to the seller | ● | cart:apply-discount, lib/resolvers/storefront/discounts.ts |
| Tax and shipping on the session | ● | automaticTaxFor, resolveShipping, lib/storefront-session.ts:510,580 |
| A receipt only from a settled row | ● | ChatCheckout.tsx (receipt phase) |
| A was-price only when real | ● | compareAtPrice, lib/pay/checkout.ts:647 |
| The full total as the first line on every door | ◐ | the shop page has a summary; the sheet does not |
| Apple Pay | · | domain not registered for the Connect account |
| One look across the eight doors | · | each door its own layout |
B · Counting
| Power | State | Where |
|---|---|---|
| Checkouts started, per door | · | sales-math.ts:185: "Nothing posts a checkout event" |
| A table that would hold it | ◐ | storefront_events, read by storefront:stats; nothing writes checkout |
| An intent signal when a card form mints | ◐ | ui:checkout:open, EmbeddedCheckout.tsx:181; not on the shop page, not on USDC, not on arrival |
| Paid orders per door and channel | ● | orders, via, the sales room |
| Abandoned carts | ● read only | lib/store-admin/checkouts.ts:4 |
| Take rate, lift, refunds per lever | · | nothing |
C · Before payment
| Power | State | Where |
|---|---|---|
| Order bump | · | nothing; the skill writes words only |
| Bundle with a derived saving | ◐ | BundleOffer block (archived barrel); saving derived or absent |
| "Goes with" from real products | ◐ | CrossSellView.tsx:70, seller-typed pairing |
| Pairing from orders that held both | · | nothing reads it |
| Free-shipping line | ◐ | threshold read in shipping.ts:141; not in the drawer |
| Member price | · | rewards:enroll exists; no price reads membership |
| A real deadline that withdraws its offer | ◐ | CountdownBar block; not on any checkout |
| Seats left from a real cap | · | nothing |
D · After payment
| Power | State | Where |
|---|---|---|
| A card saved for later, with consent | · | setup_future_usage is in no code in the repo; only a subscription keeps a card, for its renewals |
| A one-press charge on that card | · | proposed; not in the registry |
| An offer on the receipt | · | the receipt ends the sheet |
| The funnel's upsell button | ○ | UpsellStep.tsx:47: records a choice, charges nothing |
| The upsell block | ○ | draws a link; typed "save 20%" in a template |
| A second checkout as the offer, today | ◐ | checkout:create works; the buyer enters the card again |
| A reminder to a buyer who left | · | "a recovery email is a later append" |
E · Agents
| Power | State | Where |
|---|---|---|
| The words for a bump, an offer after, a bundle, a reminder | ○ | six elevate-* skills |
| An agent that writes them | ○ | upsell-strategist; its two signals are not receivers |
| An agent that tunes a lever | · | no lever exists to tune |
| An agent buyer that pays | ● | the vault spender (text/one.md § The treasury) |
F · Sold
| Power | State | Where |
|---|---|---|
| The course, $100, live | ● | price_sold_humans, sold-offer.ts:16 |
| The community, $100 a month | · | "No such price" fix in review (goal 1) |
| The book, the call | ○ | live: false in showcase.ts |
| Sign-in before paying | ● found by the 2026-11-14 probe | a question for Tony, below |
What did not land
- ONE's only upsell button charges nothing. "One-tap add to my order" records a choice and moves the funnel on. A buyer who presses it believes they bought something. This is the first fix, before any new lever (deliverable 8).
- The prompts talk to nobody.
upsell-strategistwaits forupsell:build-neededand answersupsell:assets-ready. Neither is a receiver, so a signal to either dissolves. - A one-press card charge needs a door that does not exist. It needs a saved card (with the buyer's yes) and a server-side charge. Both are new. Until then, the offer after payment is a second checkout.
- Sold has one live product. An honest bump or offer for Sold needs a second live product. The community comes first, when its checkout works; the book and the call are not on sale. Until then Sold's checkout shows no offer, and this story improves Sold only through counting, design and fewer steps.
Two questions for Tony
These are his calls. The story builds either way.
- Sign in before paying for Sold. The 2026-11-14 probe found the $100 course asks a stranger to sign in first. Baymard names a forced account as a top reason buyers leave [11]. If the key comes first on purpose, the story keeps it; if not, enrolment can follow payment by email.
- The founding group's cap. Goal 1 names a founding group of 20 to 50. If the cap is real, "seats left" is honest scarcity, read from the member count. If it is a target, the checkout says nothing about seats.
The order we build in
Planned in waves, never in days. Sold first, because goal 1 is the first
real Sold sale (text/goals-2026-11-18.md).
W1 count, fix what lies, and Sold's sheet:
checkout started on every door; the
drawn upsells charge or go; the sheet
shows the full total first
│
▼
W2 the honesty law at save, then the
order bump on every door
│
▼
W3 the lever read: take rate, lift,
refunds, against a holdout; on the
sales room, the CLI, the chat and MCP
│
▼
W4 the offer after payment: a second
checkout first; then the saved card
with consent and the one-press charge
│
▼
W5 the cart drawer: free-shipping line,
"goes with" from real orders, bundles
│
▼
W6 agents tune: on/off, choose, test;
a person approves any price
│
▼
W7 member prices, real deadlines and
seats, the reminder inside consent
Wave 1 is the smallest set that lets Tony watch one buyer reach Sold's checkout and be counted. Three rows, to be filed once Tony has read this file:
- Count every door. One "checkout started" row from each of the eight
doors, on either rail, naming the door, so the sales room's fifth stage shows a number.
Deliverable 1. Tags:
checkout,sales. - Stop the two upsells that lie.
UpsellStepsays what it does or charges through a real checkout; the upsell block binds a price or is removed; the typed "save 20%" goes. Deliverable 8. Tags:checkout. - Sold's sheet. The slide-in sheet shows the full total before the
rail buttons, and shows no offer while Sold has one live product. The
community month is said only once the community's checkout works
(
text/one.md:470). Deliverable 9, and the first read of conversion on ONE's own checkout. Tags:checkout,sold.
Every surface moves with each row: the receiver contract, the CLI verb, the
MCP tool, the sales room, the chat and llms.txt (CLAUDE.md § Every
surface moves together). A money-path row goes review branch → security
PASS → dev.
What it defends, and what it accepts
| Surface | Defends | Accepts |
|---|---|---|
| The total | the first number is the whole number | a lower-looking price than a site that hides fees |
| The bump | never ticked for the buyer; skip is one press | a lower take rate than a pre-ticked box |
| The offer after | one, then at most one smaller | leaving money a third offer might take |
| The saved card | saved only on the buyer's yes; the bank can still ask | a second checkout for buyers who say no |
| USDC | every payment is the buyer's signature | no one-click on USDC |
| Scarcity and deadlines | only from a count or a date | fewer urgent-looking pages |
| The tuning agent | tunes inside the law; a person moves a price | slower price tests |
| The holdout | a lever's lift is measured, not assumed | a share of buyers who never see a lever |
| The money | card to the seller's Stripe, USDC to the seller's wallet; ONE holds neither | ONE earns only its fee |
Where this file is looser than the code
- "Eight doors and one mint per rail." The card doors share
mintConnectCheckout. The crypto rail has its own path (mintCheckoutLink, the inline escrow rail), and credits settle on a balance. "One mint" means one per rail. - "The slide-in checkout." Read here as the chat sheet
(
openOfferChat→ChatCheckout). The cart drawer also slides in; both are in scope, and the drawer gets wave 5. - "A lever is a path between two products." A design. No path is
written today for any product pair, and where a lever's settings are
stored is a decision for
checkout-plan.md. - "Checkout started" receiver. Not named here, and none is invented. The table exists; the writer and its name are wave 1's to decide.
- The one-press charge. Proposed only. The
receiverskill's first question (does a path already do it?) is unanswered. - Outside numbers. Every figure from Baymard, IdeaWorks, Zipify and the cart-drawer vendors is theirs. ONE has measured none.
- Locked words used: signal, mark, warn, fade, follow; the
sellandhumanstep kinds by name. Dead names: none.
Sentences that would be false if shipped
- "ONE's checkout raises your order value." (No lever exists.)
- "See your conversion rate." (Checkouts started has no source.)
- "One-click upsells on every shop." (No card is saved; USDC has no one click.)
- "One-tap add to my order." (It adds nothing today.)
- "Customers who bought this also bought…" (No pairing is read from orders.)
- "Recover abandoned carts automatically." (The list is read only.)
- "Our agents optimise your checkout." (Their signals are not receivers.)
- "Zipify-level take rates." (Their figure, not ours.)
Sources
- ClickFunnels support, Adding a one-click upsell / downsell / OTO page — https://support.myclickfunnels.com/docs/adding-a-one-click-upsell-downsell-oto-page.md
- ClickFunnels support, How to add a one time offer (OTO) product in a funnel step — https://support.myclickfunnels.com/support/solutions/articles/150000156993-how-to-add-a-one-time-offer-oto-product-in-a-funnel-step
- CartMango, Order bump — https://cartmango.com/order-bump/ ; SamCart, Order bumps vs upsells — https://www.samcart.com/blog/order-bumps-vs-upsells
- FunnelKit, Order bump vs one click upsells — https://funnelkit.com/order-bump-vs-one-click-upsells
- Zipify, The best upsell app for Shopify Plus in 2026 (vendor) — https://zipify.com/blog-best-shopify-plus-upsell-app-in-2026/ ; Fudge, Best Shopify apps for post-purchase upsells — https://www.fudge.ai/blog/best-shopify-post-purchase-upsell-apps/
- GrowthSuite, Free shipping progress bar (vendor) — https://www.growthsuite.net/resources/shopify-upsell-cross-sell/cart-drawer-upsell/free-shipping-progress-bar ; Upcart — https://apps.shopify.com/quick-upsell
- IdeaWorks, Global estimate 2025 — https://ideaworkscompany.com/wp-content/uploads/2025/11/Press-Release-197-Global-Estimate-2025.pdf ; IdeaWorks reports — https://ideaworkscompany.com/reports/ ; View from the Wing on United — https://viewfromthewing.com/united-airlines-makes-11-5-billion-from-extras-but-paying-more-doesnt-mean-youre-getting-more/
- Booking.com Genius — https://www.booking.com/genius.html ; Booking.com for Business — https://business.booking.com/make-business-travel-more-rewarding/
- Agoda payment options and VIP tiers, as described by third parties (not Agoda's own pages) — https://www.shopback.com/agoda ; https://www.rankandstyle.com/deals/agoda-promo-code
- CMS on the CMA's 2019 commitments from six booking sites — https://cms.law/en/gbr/legal-updates/cma-secures-commitments-from-holiday-booking-sites-on-sales-practices ; TravelPulse — https://www.travelpulse.com/news/travel-technology/six-booking-sites-promise-to-stop-misleading-travelers.html
- Baymard, Cart abandonment rate — https://baymard.com/lists/cart-abandonment-rate ; Baymard, Checkout usability — https://baymard.com/research-articles/ecommerce-checkout-usability-report-and-benchmark
- FTC, Trade Regulation Rule on Unfair or Deceptive Fees (final rule) — https://www.ftc.gov/system/files/ftc_gov/pdf/r207011_udf_rule_2024_final_0.pdf ; Reed Smith summary — https://www.reedsmith.com/en/perspectives/2025/01/junk-fees-no-more-the-ftc-releases-final-rule-on-unfair-or-deceptive-fees
- European Commission, Consumer Rights Directive — https://commission.europa.eu/law/law-topic/consumer-protection-law/consumer-contract-law/consumer-rights-directive_en ; Wikipedia, Consumer Rights Directive 2011 — https://en.wikipedia.org/wiki/Consumer_Rights_Directive_2011
- Directive 2005/29/EC, Annex I — https://www.legislation.gov.uk/eudr/2005/29/annex/I/adopted
- Mathur et al., Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites (CSCW 2019) — https://arxiv.org/pdf/1907.07032 ; https://webtransparency.cs.princeton.edu/dark-patterns
- Gibson Dunn, FTC restarts negative option rulemaking after Eighth Circuit vacatur; enforcement under ROSCA continues — https://gibsondunn.com/ftc-restarts-negative-option-rulemaking-after-eighth-circuit-vacatur-enforcement-under-rosca-continues/
- Stripe, Save a card and charge it later — https://docs.stripe.com/payments/cards/save-and-reuse ; Stripe, Use saved payment methods (Checkout) — https://docs.stripe.com/checkout/custom-checkout/use-saved-payment-methods ; Stripe, Save during payment — https://docs.stripe.com/payments/advanced/save-during-payment
Where the depth lives
| For | Read |
|---|---|
| The sales path a checkout sits at the end of | text/sales.md (draft, not on dev yet) |
| What Sold is and what it costs | text/sold.md |
| The goals this serves | text/goals-2026-11-18.md |
| The ruling that a claim with no source is deleted | text/pay-components-plan.md § The blocks |
| The store's blocks, graded | text/ecommerce.md |
| Selling inside ChatGPT | text/sell-in-chatgpt-plan.md |
| The consent door for any reminder | text/crm.md |
| What the money means once it lands | text/finance.md |
Connected to
Every link here was checked against the repo when the page was built. 2 names the story uses could not be found and are shown without a link.
People
- the buyer
- the seller
- the owner of the group
- the person who approves a price
Agents
Each opens a chat with that agent.
- upsell-strategist
- refine-analyst
- cro
- sales-closer
- an agent buyer paying from its own wallet
Skills
Each opens the skill's page.
Paths
The flows this story walks — trigger → step → step.
- trigger: a Buy press on any door → tool: checkout started (one row, the door named) → condition: a bump is set and honest? → tool: the bump line → sell: checkout:create → condition: paid?
- trigger: paid → condition: an after-payment offer is set and live? → condition: card saved with consent? → tool: the off-session charge (PROPOSED, not in the registry) | sell: checkout:create (a normal second checkout) → condition: taken? → no: one smaller offer, then the receipt
- trigger: the refund window closes → tool: the lever read → condition: the lever beat its holdout without raising refunds? → human: the seller keeps it on, or an agent turns it off
Tools
Receivers and scripts it signals.
- checkout:create
- checkout:resume
- checkout:link
- checkout:notify
- cart:get
- cart:add
- cart:update
- cart:remove
- cart:apply-discount
- cart:remove-discount
- products:get
- products:list
- orders:status
- orders:send-link
- storefront:stats
- funnel:advance
- links:create
- rewards:enroll
- workflow:step-stats
- human:notify
- human:resolve
Events
Every event is a signal; each names the path it marks or warns.
- checkout started (VERIFY: no receiver; storefront_events takes a 'checkout' row nobody writes) · from the door the buyer pressed door → started; marked by a paid order, warned when the session expires unpaid tags: checkout, intent
- checkout:create · from every card door started → paid; the run waits at a sell step tags: checkout, sale
- offer shown / taken / declined (VERIFY: none is a receiver) · from the bump line and the receipt product → product, tagged with the lever; marked on a paid take, warned on a decline or a refund tags: checkout, lever
Transactions
- the main order, by card to the seller's Stripe or by USDC to the seller's own wallet
- a bump, inside the same payment
- an after-payment offer, as its own payment
- ONE's platform fee on each, computed from the seller's fee row, never from the request
Analytics
- checkout started per door — VERIFY: not counted
- paid per door — read today from orders
- take rate per lever — not computed
- order value with and without each lever — not computed
- refunds per lever inside the window — not computed