9. Economy

Currencies#

Each currency has one main source and one main sink, so the player always knows what each number is for.

Users are not a currency. They are the state that every currency depends on.

Cash#

Sources#

The Payments drawer#

Revenue first goes to the Payments drawer. The player clicks Collect to move it into Cash.

• The drawer holds 60 seconds of revenue at the current rate, at size 1, and never less than $25. Each size step doubles it.

• Revenue beyond the drawer's size is lost while the player is playing. While the player is away, the offline simulation pays revenue straight into Cash (Simulation).

• Research "Stripe Billing" (Garage, 30 ◇) collects automatically. After that, the drawer only shows the rate.

This keeps the Garage Era active and then removes the click before it becomes a chore (pillar 5).

SLA credits#

B2B SaaS Ventures from Series B run under an SLA, a promise of availability one level above the Era's target (Ventures). While the trailing calendar week sits below that promise, the Company pays credits back to its customers: 10% of the subscription revenue it is earning, charged with the Burn. A Company on a 99.95% SLA that misses a week watches Cash fall faster without a single extra Node on the board.

The credits need enterprise-sales research, which signs the contract (Progression). No other Venture signs an SLA, a Venture that has not reached Series B has not signed yet, and a week that met the target costs nothing. This is the one place where reliability is a line on the P&L rather than a figure in the HUD (Incidents).

The cloud bill#

Burn is the total Cash spent per second. The HUD shows Cash with its net change per second, income minus Burn. The Cost panel ($) splits Burn into the lines below, with Revenue and the margin beside them. A data Node's standby is a second box and doubles that Node's Burn; failing over to it makes the standby the primary and halves the Burn again until the player adds a new standby (Incidents). The table is the bill:

Egress, observability, and third-party fees are the lines real teams tend to miss until the invoice arrives. The Cost panel shows each one on purpose.

Base prices#

These are Garage prices. Every price in this document, including placement, Burn, egress, and third-party fees, is multiplied by the Era's cost index, except the Payment Provider's fee and the Auth Provider's per-user price. The index keeps Burn a real share of revenue at every size of company. Balancing holds the index and the salary scale, and Balancing holds the full table.

How prices map to real prices#

Each price follows one of two rules, so a player who knows the real price can check it:

LLM tokens use the time-billed rule because they compete with GPU Nodes for the same prompts. On the same scale, a busy H100 is cheaper per prompt than the LLM API only while it is more than half busy, as in real self-hosting (Reality check).

Request values set what a user is worth. A user sends about 0.02 requests a second, which on the time-billed rule is 0.02 an hour in real time, at about $0.25 each, so revenue is about $4 per user per real month. Meta's 2025 worldwide ARPU was $57 a year, about $4.75 a month.

The two rules disagree, and the gap is deliberate. Per-use prices are about 3,600 times cheaper, relative to compute, than in a real bill, because at the time-billed rate every page view would pay half its value in egress. Request values, the cost index, and the salary index are balance controls and do not follow a real price.

The per-use prices of a function, a Queue, and Object Storage are the exception. The list prices of Lambda, Workers, SQS, and S3 and R2 are multiplied by 1,200. For a function that puts the crossover at about 30% of an equal VM's steady load, where Components puts it. At the straight list prices a function would never cost as much as the VM it replaces.

Pricing models#

Every Node bills at one rate until the Series A Era. From Series A, each compute Node (an App Server or a Worker) has a pricing model. The player picks it on the Node's card, and the box price in the cloud bill is multiplied by it.

Sized compute Nodes take On-demand, Reserved, or Spot. Serverless is not a model on a box. It is the Serverless Function's own bill: each request and each GB-second on the Requests line, with nothing standing while it idles (Components).

The cheapest fleet mixes all three. Reserved instances cover the base load, on-demand instances cover the daily peak, and spot instances run batch work.

savings-plans (Progression) takes a further 20% off every compute Node's box, on top of whichever model it already runs. It needs no per-Node commitment, so it is a Company-wide discount and not a fifth model on the card.

Egress#

media at 5 MB per request makes egress the largest line on the bill in the Series A Era unless a CDN or R2 serves it. A player who ignores egress sees revenue rise and margin fall.

egress-optimization (Progression) moves part of a server's egress bytes onto the R2 row: $0 instead of the origin price. The part is the share that upload and media traffic reaching an R2 Object Storage Node accounts for. The research reads this from the server's own traffic mix, not a per-byte trace, and only an Object Storage Node that is R2 counts.

Third-party fees#

Funding#

When users reach an Era milestone, investors offer a round. The player can raise or decline.

A grant is about 17 sim minutes (1,000 s) of the Burn at the milestone, so it stays worth taking at every size of company.

Money buys growth. An Era that begins with a raised round runs at the full growth rate g. An Era that begins with a declined round runs at 0.7 × g, so it takes about 1.4 times as long (Simulation). The player trades Equity for speed.

Declining a round keeps Equity and gives no grant. The Era still advances, so the new kinds still arrive, and the player must pay for them from revenue. A player who declines every round and still reaches IPO earns the achievement "Default Alive", named after Paul Graham's essay. The pacing bots check that this path does not beat the raise-every-round path in Legacy per hour (Balancing).

Dilution compounds. A founder who raises every round keeps about 0.90 × 0.81 × 0.81 × 0.87 × 0.88 × 0.88 ≈ 39.8% before the IPO offering and about 33.8% after it, which is above a typical real founder stake at IPO.

Valuation#

The company has a valuation at all times. It decides the Exit payout.

The clamp applies to base only. Investor relations adds 2 to the multiple, and a public quarter adds its bonus, from −10 to +10 (Progression), so either can take the multiple past 30.

valuation_scale is 1.5 × 10⁻⁴. Request values are stylized game dollars, so Cash and revenue run far above real figures. The scale converts a year of game revenue into the real-dollar ARR that the company would report (Balancing). The Valuation, Exit payout, and Legacy use real dollars. Cash, revenue, and Burn use game dollars.

The Rule of 40 is a real heuristic for software companies: growth plus profit margin should be at least 40%. It gives the late game a reason to care about margin, not only growth.

Runway and running out of Cash#

Runway is Cash divided by net Burn.

The HUD shows runway beside Cash's net change in 30-day months while net Burn is positive, and the Dashboard's Company panel shows the same figure. What happens at zero is a sequence of events.

When Cash reaches zero:

1. Placement and upgrades stop, and so does hiring: nothing that costs Cash starts until the balance is positive again. Selling Nodes, layoffs, Research, the backlog, and the Payments drawer keep working, and Burn keeps billing.

2. Burn continues and Cash goes negative.

3. At −1 calendar day of Burn, investors offer a bridge: a grant that is 3 calendar days of Burn for 8% dilution. The player can accept, or cut costs. The terms are fixed when the offer arrives, and the Company is offered one bridge.

4. If the player declines (leaves the offer unanswered) and Cash reaches −3 calendar days of Burn, Nodes shut down, most expensive first, until net Burn is zero. A shut Node serves nothing, bills nothing, and stays on the board until the player sells it or pays to replace it.

Running out of money is a setback and the Company continues (pillar 5). The Company row of the Dashboard shows the bridge's grant, dilution, and Accept while the offer waits (Interface).

Some details of the thresholds, the grant, the dilution, and the shutdown:

• The two thresholds count the gross Burn, which this document calls Burn. The shutdown counts net Burn, which this document names.

• "Most expensive" is the Node with the highest Burn now, excluding Users, Payments, and any Node already shut.

• The grant is fixed at the offer's Burn, like a term sheet, so later cost cuts do not shrink it.

• The 8% comes off the Equity the founder holds at the time, so two bridges would compound; the build offers one.

Sources and sinks summary#

The three main currencies (Cash, Insight, Dev Hours) compete for the player's attention, not for each other. Nothing converts one into another, except that Cash hires engineers and pays for Analytics. That keeps each choice legible.