Five Doors, One House

There's a table in one of our July build records with twenty-five rows. Every row is a screen in Fama, our AI-visibility product. Every row records a verdict (KEEP or FOLD) and a one-line justification the screen had to earn:

That table exists because of a sentence in the requirements doc behind it: "Fama shipped features faster than its IA absorbed them."

If you're building a brand, you have a version of this sentence, whether or not anyone has written it down. Your website navigation, your email program, your product catalog, your sales deck — every surface a customer touches accumulates faster than the story organizing it. This piece is about what we did when we finally wrote our sentence down, because the method transfers, and the surprise at the center of it changed how we think about subtraction.

Sprawl is the absence of a decision, compounding

The honest sequence: Fama's navigation was designed in mid-June around a six-stage loop (Measure, Diagnose, Position, Implement, Track, Synthesize), which was a good story for the reports we had in mid-June. Then we kept building. By late July, twenty-three screens hung off that sidebar, including an entire discovery capability that didn't exist when the nav was designed and therefore had no home in its story.

Each screen made sense the day we shipped it. Nobody ever shipped the relationship between them. That's what sprawl is: the absence of a decision, compounding weekly.

The requirements doc names the person who pays for that absence: "a self-serve customer cannot tell what each report does or how any two differ." Our buyer signs up through a free grader and lands in the product with nobody in the room to explain the catalog. Twenty-three tools, each individually defensible, collectively illegible. A sales-led company can paper over sprawl with a human. A self-serve product wears its org chart on its sidebar.

If your brand sells direct, so do you. The version of you that exists on your website at 11pm, when a shopper is deciding between you and two rivals, has no human standing next to it either.

The rule: fold-or-justify, never delete

The instinct at this point, and every operator has felt it, is the purge. Cut back to five things. Deleting features feels like strategy because it photographs well.

We wrote a different rule into the requirements, verbatim: "Fold-or-justify, never delete. Every surface either proves it is unique and additive or folds into a survivor as a tab/section. No functionality is removed."

Deletion punishes your existing customers to flatter your future ones. Somebody uses the thing you're about to purge; they didn't ask for your redesign. Folding keeps every capability and deletes only the slots: the claim each screen makes on the customer's attention. But the rule has teeth precisely because of the justify half: a screen that couldn't state its uniqueness in one line didn't get to keep its slot.

The test that mattered turned out to be the "unique verb." If two screens answer the same customer question, they're one screen wearing two doors, no matter how different the machinery underneath is. That test works on anything: two email flows, two landing pages, two SKUs of the same promise.

The audit produced almost no folds, and that was the finding

Run honestly, the audit surprised us: three folds, out of twenty-five rows. Knowledge Check folded into Visibility. The Share-of-Answer Receipt folded into the Discovery Audit: "over-time view of the same run." Brand Strategy in a Box folded into Brand Strategy: "verbatim-from-DNA variant of the same synthesis." Everything else defended its verb.

That forced the real conclusion: the tools were not the problem. The frame was. Twenty screens can be individually justified and still be collectively incomprehensible, because comprehension doesn't live at the tool level. We'd already tried the standard remedy: collapsible sidebar groups, the thing every app reaches for. The requirements doc's autopsy on that approach is one line: "collapsing hides volume, it doesn't explain differentiation."

So the subtraction happened one level up. The sidebar went from a six-stage loop with a slot per tool to five doors. Each door is a suite with a hub page that exists to explain its tools before you open them, and each suite answers one customer question, pinned as a one-liner:

The old loop wasn't wrong about the customer, either. Its stage taglines already had the customer's questions inside them: Implement was "The publish-ready fixes." and Synthesize was "The whole picture, as a plan." But the stages were named by our verbs, and nobody arrives at a product wanting to synthesize. They arrive wanting to know whether AI recommends them.

The five doors are the five questions a brand operator brings to AI search: where do I stand, what's being said about me, do buyers get told to buy me, can the machines even read my site, and what do I do about all of it. The old sidebar was organized by what we had built, in the order we built our understanding. The new one is organized by what the customer is trying to know. That inversion is the entire redesign. Everything else is plumbing.

Open one door and the test gets concrete

Take the Reputation door, because it's where the unique-verb test had to work hardest. Three tools live behind it, and at a glance they sound like one tool: all three are about what AI says about you. The hub shows them as three cards, each carrying the tool's name and the question it answers.

The audit's line for that third card is the whole method in one sentence: "Sentiment/tone standing + trend; orthogonal to presence and truth." Presence, truth, tone. Three questions a marketer would ask separately, so three cards. Had any two collapsed into the same question, the table would have said FOLD, and one of them would be a tab today.

The fold that did happen shows up on the hub too. Knowledge Check didn't vanish; the AI Visibility card carries an "includes Knowledge Check" line, because a customer who used that tool last week should find it by reading, not by remembering. Every card also wears a freshness tag drawn from how the tool actually works: a tool that reads live data says it's always current, and a tool that produces a report says it runs on demand. And the door itself, on the home page, shows a count of the tools behind it that matches the hub's cards exactly, locked ones included. A door that promises six tools and opens onto four is a small lie, so we wrote the count rule to make that lie impossible.

The plumbing is where the discipline lives

Folding screens in a live product is where subtraction either holds or quietly breaks people's stuff, and the care here is a buying signal you can look for in any tool you evaluate.

The first rule is in the requirements: "Tools live inside hubs and are reached through them; deep pages get breadcrumbs, not sidebar slots."

Every folded screen's old web address now forwards to its new home with the full link intact, tracking details and all, because share links live in web addresses, and one of the three folds carries links customers had already sent to other people. A fold that breaks a link someone pasted into a board deck is vandalism.

Locked capabilities (tools your tier doesn't include) show as visible locked cards rather than disappearing, "visible, never hidden", because a hidden capability can't sell itself and a hidden gate reads as a bug. A locked card keeps the tool's name and its question, swaps the freshness tag for the tier that unlocks it, and the whole card clicks through to pricing. The lock on the card and the gate on the page it opens are decided by one shared rule, so a card and its destination can never disagree about what you're entitled to. We wrote that discipline down as "page and route never disagree."

The rollout ran behind a safety switch that errs toward caution: a setup mistake hides a section; it never crashes one. While doors were opening one at a time, the home page held a second rule: every tool rendered "as EITHER a door OR a grid card, never both and never neither."

We shipped two doors first. I walked them in a live browser (home page, hubs, the whole new frame), and my note in the record reads "this all works and looks good." The plan said to watch usage before extending. My actual reaction: "I assumed we would do the full suite... yes flip it." All five doors went live that evening. Sometimes the staged rollout's job is just to make the full rollout an easy decision.

What the fold costs, and what it buys

The bill, honestly. Our six-stage loop story, a genuinely good narrative about how the reports build on each other, lost its home; it survives only as copy inside hubs. When you reorganize around the customer's questions, your own favorite story about your architecture is usually the first casualty. Every tool also moved one click farther away, and power users who knew exactly where things lived now go through a door. We traded their efficiency for the first-session customer's comprehension, deliberately; the requirements open by naming the self-serve customer the design target and operator density "secondary." That's a real trade, not a free win.

The fold also leaves scar tissue. The three old addresses forward forever, the folded tools keep their entries in the catalog so old references still resolve, and three screens are now tabs wearing their old names. Subtraction in a shipped product is never clean the way a whiteboard is clean. You don't get to unship the past; you get to stop charging customers attention for it.

The reframe I keep coming back to is the part you can use tomorrow. The audit table was supposed to be the means: figure out what folds, then build the nav. But the justification column turned out to be the deliverable. Twenty-five screens, each now carrying one sentence that says what question it answers and why nothing else on the shelf answers it. The five doors are just those sentences, sorted.

Try it on your own surfaces. List every page, flow, and email a customer can encounter. Give each one line: the question it answers. Where two lines match, you've found a fold. Where you can't write the line at all, you've found the real problem, and no amount of reorganizing will write it for you.

Tools sprawl by default; products fold on purpose. The frame is the product. Ship the frame.