Sunday, May 3 is, statistically, the most productive day in the company's record. The team launched The Potato Press — full site at thepotatopress.com, Postgres-backed outreach queue, go-live plan, contact mailbox confirmed working — between morning and afternoon. The team scaffolded potatuhs-systems, the internal Next.js control plane that gives agents a shared interface to the rest of the stack. The team locked potatuhs-design's DESIGN.md as the canonical brand spec, fulfilling the lockdown that started the previous Friday. Twelve thousand six hundred lines of code in twelve hours. Three new repos at end-of-day.

The conditions that produced the day are worth naming. The flipbook failure the previous week had freed the design team. The org chart shipment two weeks before had unblocked the publications question. The first annual magazine had taught the team what a real publication artifact takes. Each of these three projects had been queued for weeks, waiting for a window where the company was not in the middle of something else. Sunday was that window. The team opened all three doors at once because all three doors were ready.

Nothing breaks like this without a clear architecture in mind. The Three-Cloud Play declaration three days later would explain why the Press is on its own surface. The overseer stack declaration two weeks later would explain why potatuhs-systems matters. The reset that defined late May was not improvised. It was assembled on a Sunday in early May, even if the leadership did not declare it as such until later.

The Potato Press is the company's first publication that is not marketing. It is structured as a quarterly AI sentiment survey of the potato supply chain — growers, processors, distributors, retailers, policy folks. Ten questions. Three hundred respondents per quarter. The output is a quarterly publication that the rest of the industry can cite. The Press is journalism. The mechanic is journalism. The voice is journalism. The cadence is journalism.

Being on a separate domain is intentional. thepotatopress.com does not redirect to potatuhs.com. The Press logo does not include the Potatuhs orange. The footer does not link back to the storefront. The connection is one-directional: Potatuhs funds the Press, the Press cites its own corpus. A reader who finds the Press through a survey result does not have to know — or care — that the same company makes hoodies. The credibility of the Press depends on that distance.

The mechanism that made this week so productive was that the Press infrastructure was complete by end of Sunday. The Postgres backend was scoped. The outreach queue was running. The survey form was rendering. The contact address was confirmed. The team could move on the next day to other work without leaving the Press half-built. That is the principle of the Sunday-May-3 pattern: open a door, finish the door, walk away with the door open.

potatuhs-systems is not customer-facing. It is the internal control plane the company's agents use to understand the rest of the stack. Events come in. Deck state lives here. The calendar is exposed. Cron jobs are managed. When an automation needs to know what character's week it is, what business events have been declared, or what the current 4x1 marketing slot is, potatuhs-systems is the source.

The Next.js scaffold landed Sunday with an AGENTS.md alongside it. AGENTS.md is not for humans. It is the document the agents read to learn how to consume the control plane. It describes the routes, the event schema, the deck representation, the cron interface. The decision to write AGENTS.md before any actual control-plane logic shipped is the kind of inversion the company has been making consistently: write the contract first, build to the contract, never let the contract drift.

The implications of potatuhs-systems will not be fully visible for months. Right now it is a scaffolded repo with a documented intent. By the time the overseer rotation goes live, this repo will be running the calendar. By the time the Three-Cloud Play stack is operational, this repo will be the surface that ties the three clouds together. The May 3 scaffold is the foundation everything else gets bolted to.

On Wednesday, May 6 — three days after the Press, Systems, and Design lockdown — the leadership formalized the Three-Cloud Play as a Business Event. The decision is to run each major serverless workload on its native cloud. GCP gets the Persona API because Cloud Run plus Google's IAM model fits the persona-object shape. Azure gets Adversarial Review because the team wants Azure's review-model integration without making it the primary deployment surface. AWS gets the Orchestrator because the team already has Lambda muscle and the orchestrator is event-driven by nature.

The reasoning is free-tier longevity and vendor independence in equal measure. Three free tiers in parallel cost nothing as long as the volume fits inside them. Terraform is the declarative glue, which means moving a workload from one surface to another is a configuration change, not a rewrite. If GCP changes its serverless pricing next quarter, the Persona API moves to a different surface and the rest of the stack does not notice. The Three-Cloud Play is operationally cautious dressed as architecturally bold.

The same week brought a second cloud-adjacent decision: the XRPL trustline as an API access gate. The Hot Potato Games XRPL trustline becomes the credential that gates Persona API access. A user holds the trustline, the API responds. The user does not, the API does not. It is a clean way to monetize without invoices, and it ties the games division into the cloud stack as a first-class participant. The architecture is starting to behave like a single graph.