Masterclass · World Economics

Worlds That Pay For Themselves.

A world is a machine you can version, move and re-run. That is a technical claim and it is made elsewhere. This page is about the money. What a world costs to keep, what it costs to move, why the part you actually differentiate on is close to free, and the four things you can charge for once the unit of delivery stops being a repository.

10 GB10 TB
the current ComputeDriven world tiers — an envelope, not a definition
0.21%
of a 100 GB world is the semantic layer, at its enforced ceiling — the part you differentiate on
1
artifact to ship, restore and bill, instead of a repository plus the machine that runs it
0
prices printed on this page — they live on the pricing page, and this one refuses to keep a second copy

01 · The unit

What a customer is actually buying.

Software is normally sold as a repository plus instructions for reproducing the machine it runs on. The instructions are the product's weakest part: they drift, they assume a base image, and they are the reason “works on my machine” survived thirty years of tooling meant to kill it.

A world is the machine itself, as a versioned artifact. Not a container image, which is a build output with no history; not a VM snapshot, which has history but no semantics. A world carries its own state, its own lineage, and a small semantic layer describing what is in it — and it can be restored, forked, moved between a laptop and a datacentre, and billed, as one thing.

The commercial consequence, stated plainly. When the unit of delivery is a world, the thing you hand a customer is the thing you tested. There is no reconstitution step between your evidence and their environment, so the class of support ticket that begins “it worked in staging” has nowhere to come from. Every argument on this page is downstream of that one.

02 · The comparison that matters

When a world beats a repository and a CI pipeline.

Not always. A world is a worse fit than a repo for a library, a pure function, or anything whose value does not depend on the state around it. The honest version of this argument is a boundary, not a claim of superiority — so here is the boundary.

What you are shippingRepo + CIWorldWhy
A library, an SDK, a pure functionbetterworseNo environment to carry. A world is overhead you pay for nothing.
A stateless service with a managed datastorebetterevenThe platform already reconstitutes the machine. You are buying a second copy of a solved problem.
A stateful system a customer must run themselvesworsebetterThe install guide is the product risk. A world deletes the guide.
An agent or model pipeline with tuned local stateworsebetterThe tuning is the value and it lives in the state, which a repo does not carry.
Anything you must reproduce months later for auditworsebetterRe-running a pipeline reconstructs an approximation. Restoring a world does not reconstruct anything.
Regulated work needing a defensible chain of custodyworsebetterLineage is a property of the artifact, not of a log a reviewer has to trust.

Two of those six favour the incumbent, and they are the common cases. The world argument is not that everything should be a world — it is that a specific and well-paid class of software has been paying reconstitution costs it did not have to, because the only unit on offer was a repository.

03 · The property that makes the margin

The part you differentiate on rides along nearly free.

A world is mostly operating system, runtime, packages and data — bulk that is identical across every customer you have. The part that is yours is the semantic layer: what the system knows, how it is wired, the decisions recorded against it. That layer is bounded by design, and the bound is enforced rather than hoped for.

At its enforced ceiling, the semantic layer is 0.21% of a 100 GB world. Roughly one part in five hundred. Which means the marginal cost of carrying the thing that makes your product different from the next one is, to a first approximation, zero against the cost of the world that carries it — and the bulk it rides on is exactly the part that deduplicates across customers.

That is the whole margin story, and it is worth being precise about what it does and does not say. It does not say worlds are cheap; a 10 TB world costs what 10 TB costs. It says the differentiated fraction of a world is nearly free to version, fork and ship, so the operations that a repository-shaped business treats as expensive — branch a customer, snapshot before a risky change, keep last quarter restorable — stop being priced like copies of the whole system.

04 · Run the numbers

Size a world, and price it.

Object storage at $0.015/GB-month with the first 10 GB free and zero egress. Margin is (price − COGS) / price. The published tier table is reproduced by this calculator exactly — if a value here disagrees with the table, one of them is wrong and the falsification row below says which.

world sizing · COGS and the 65%-margin price arithmetic · $0.015/GB-mo, 10 GB free
Sizes step through the published tiers: 10 GB · 100 GB · 500 GB · 1 TB · 2 TB · 10 TB.
storage COGS
per month, billable GB only
price at target margin
per month
WRL graph share
of the backup, at the 212.70 MB ceiling
The correction this calculator encodes. The published table's price column was once computed as COGS ÷ 0.65, which yields a price where COGS is 65% of revenue — a 35% margin, the inverse of what was intended. Every row was wrong. The formula here is price = COGS ÷ (1 − margin), giving $438.43 at 10 TB for a 65% margin, not the $236 that circulated for weeks. Set the slider to 10 TB and check it against the table.

05 · Running it

Where a world actually runs, and who pays for that.

A world runs on your machine or on ours, and the same artifact does both — that is the point of the unit. Locally it costs you disk and the machine you already own. Hosted, it costs a monthly figure driven by the arithmetic in §04: billable gigabytes, at a rate, at a margin.

This page prints no prices, on purpose. [World] Cloud publishes its own tiers, and a masterclass that kept a second copy of them would be a second source of truth — which is precisely the failure this site has gates against elsewhere. The calculator above is arithmetic you can check; the numbers you would be invoiced are on the pricing page, and if the two ever disagree, the pricing page is right.
[World] Cloud pricing → What the hosted tier is
Where it runsWhat you payWhat you get
Your machineDisk you already ownThe whole world, offline. No per-seat anything. This is the default and it is not a trial.
Your infrastructureYour existing hosting billSame artifact, your cloud account, your compliance boundary.
[World] CloudPublished tiers — see the pricing pageHosted worlds, restore and fork without provisioning, and the control room to drive them.

The local tier being genuinely usable is a commercial position, not an oversight. A world you cannot run without us is a hostage, and hostages are sold once. The hosted tier has to win on being better than doing it yourself — restore speed, forking without provisioning, and not having to operate the substrate — rather than on being the only way to open your own artifact.

06 · The client-side argument

Four things you can charge for once the unit is a world.

This section is for people building on the system rather than buying it. Each of these is a line item that is hard to bill for when you ship a repository, and straightforward when you ship a world.

Line itemWhy a repo makes it hardWhy a world makes it billable
A per-customer environmentEvery one is a bespoke reconstitution and the labour is invisible until it breaks.A fork. The bulk deduplicates; you carry the 0.21% that is actually theirs.
Restore-to-a-known-goodYou can restore data or redeploy code, but not the machine as it was — so you sell “best effort”.The known-good state is a stored artifact, so the promise is checkable and can carry an SLA.
Audit and reproductionReproducing last quarter means rebuilding it and arguing the rebuild is faithful.Restore the world version. Nothing is reconstructed, so nothing has to be argued.
Try-before-you-changeStaging is an approximation of production, which is why it fails to predict production.Fork the world, make the change, compare, discard. The fork is production, minus the consequences.
What none of this includes: a customer count. No revenue figure, no logo wall, no case study appears on this page, because there is not one yet that could be published honestly. These four are arguments from the properties of the unit, and they should be read as arguments — the section below is how you would go about breaking them.

07 · The ending this page has to have

Here is precisely how to prove us wrong.

Every masterclass on this site ends by handing you the thing that would break it. A commercial page has more reason to do this than a technical one, not less.

ClaimStatusHow to check itWhat kills it
The calculator reproduces the published tier arithmetic arithmetic §04 — set 10 TB / 65% margin COGS $153.45, price $438.43. $236 is the pre-correction value and implies a 35% margin, not 65%. If the slider disagrees with price = COGS / (1 − margin), the page is wrong.
The semantic layer is 0.21% of a 100 GB world at its ceiling measured The ceiling is enforced, not observed — the figure is what the bound permits, not an average of real worlds A real deployed world whose semantic layer exceeds the enforced ceiling, or a demonstration that the ceiling is not actually enforced. Either one collapses the margin argument in §03.
Two of six shipping shapes are better served by a repo judgement §02 — the boundary table Argue a third row the other way, or argue that one of the four “world better” rows is a platform problem already solved by managed services. This is the softest table on the page.
Four billable line items follow from the unit argued, not evidenced §06 No customer is running any of these yet. These are arguments from properties, and a single deployment where forking a world turned out to cost the same as rebuilding an environment would end the section.
The hosted tier wins on merit, not lock-in a position, not a measurement §05 — run a world entirely locally Find a capability that exists only in the hosted tier and is required to open, restore or fork your own world. If one exists, the local tier is a trial and this page is marketing.
No prices appear on this page structural grep -c '\$' world-economics-masterclass.html — the only dollar figures are in §04's published arithmetic and the falsification row above A tier price hard-coded anywhere in this file. It would be a second source of truth against the pricing page, which is the drift class this site builds gates against.

What this page deliberately no longer contains. It replaces an earlier masterclass which spent two of its seven sections on an internal vocabulary correction and an unresolved architecture question. Both are real and both are still published — on the factory floor, where a defect register belongs. A curriculum page is not a defect register, and the earlier version confused a reader's first encounter with the idea for a status meeting about it.