The capability set you declare at the foundry is the capability set that machine will have when it is scrapped. There is no amendment, no addition, no revision under any circumstance.

Operators know this and declare too broadly anyway, because breadth feels like prudence and the cost of it arrives immediately while the benefit is hypothetical.

The registries let us check how often the hypothetical benefit materialises. It does not, mostly.

What a Tag Commits You To

A capability tag permits production of a matching resource type and permits nothing else. Production is gated absolutely: a chassis without the tag cannot make the thing, regardless of how similar it is to something the chassis can make.

Tags also drive the declared build cost, because a machine engineered to cover six output types genuinely costs more to build than one covering two. That declared figure then sets the creation tax at eight percent, the destruction tax at three, and the Annex buyback reference at sixty percent.

So a tag you never exercise has cost you eight percent of its contribution to the declaration, permanently, and returned nothing except a raised buyback ceiling — which is worth something, and is not usually what the operator thought they were buying.

Choosing the Set

The test is not whether you might use a capability. It is whether you have a concrete production intention for it within the horizon you can actually see, which for most operators is two quarters.

Compare two declarations on the same underlying machine. Narrow, covering the two output types you have committed demand for: declared 22,000, creation tax 1,760. Broad, covering those two plus four adjacent types you can imagine wanting: declared 62,000, creation tax 4,960.

The broad declaration has cost 3,200 more immediately. For it to pay, at least one of the four speculative tags must eventually produce output worth more than 3,200 coins net — and it must do so before the machine is retired, in a band that is clearing, at a moment when you have a free slot to run it.

"Write down which tags you will use next quarter. Declare those. The ones you cannot write down are the ones you are paying eight percent to feel comfortable about." — Orin Dast, foundry supervisor, Calvert line

The Caldera District Registration records are unambiguous about how this resolves. Broad declarations overwhelmingly spend their service lives producing within their original narrow intention. The speculative tags sit unexercised, having been paid for at declaration and carrying an elevated destruction bill at the end.

There is one genuine argument for breadth and it is worth taking seriously: the buyback reference. A machine declared at 62,000 has an Annex ceiling of 37,200, against 13,200 for the 22,000 declaration. If you expect to exit through the Annex, the broad declaration is worth 24,000 more at the wall. That is a real consideration and it is a bet on your own failure, which operators are understandably reluctant to price.

Where Narrow Declarations Hurt

They hurt exactly when the band goes soft, which is the scenario breadth is bought against. A narrowly-declared fleet facing a demand shift has no capacity to respond, cannot be redirected, and can only be replaced — at eight percent, into a market that is already weak.

The honest position is that neither choice is safe. Breadth pays a certain cost for an uncertain option; narrowness accepts a certain fragility for a certain saving. What is not defensible is declaring broadly without pricing the option, which is what most broad declarations actually are.

What Operators Believe About Tags

The first belief is that an unused tag is harmless. It is not free — it raised the declaration, the creation tax and the destruction bill. It is closer to an unexercised option with a premium paid up front and no expiry, which is a thing worth owning only if you priced it.

The second belief is that adjacent tags are cheap. They are the reason the declaration rose. A supervisor quoting a higher figure for a broader machine is quoting the cost of building it, and the eight percent applies to the whole figure rather than to some notional core.

Declare what you will use, plus whatever breadth you are willing to defend as a deliberate purchase of the buyback ceiling. Everything else is an option you will pay for at the foundry, carry for the machine's life, and settle again at three percent when you finally take it out of the world.

Note: Mechadia is a work of fiction. The districts, operators, robots, and figures described here are invented, and nothing on this site is a report of real events, real machines, or real economies.