Composable Twins

This outline is signed before any prose is generated. Text up here is the preamble: it belongs to no heading, is never handed to a generator, and exists for whoever opens the file next. Audience: practising engineers who have built one twin and are now asked for five.

Part I: Foundations

What a twin is, and is not

Establish the vocabulary the rest of the book leans on, and settle the twin/shadow distinction early -- later chapters assume it. Must not drift into architecture; that is Part II's job.

The model half

What the model must reproduce, and the fidelity question stated but not answered. The answer is ch-fidelity.

The data half

Synchronisation, sampling and staleness, as vocabulary only.

What a twin costs

The honest chapter. Build cost, maintenance cost, and the cost of a twin nobody trusts. Leans on ch-what's vocabulary and nothing else.

What you pay to build one

What you pay to keep one

Part II: Building one

Choosing fidelity

Answers the question sec-model raised. This is the book's technical core and the chapter most likely to need two passes.

Fidelity against a decision

The chapter's central claim: fidelity is chosen against the decision the twin supports, never maximised in the abstract.

When more fidelity makes things worse

The deadline argument. A more faithful model that misses its update window is worse than a coarse one that does not.

Keeping it in step

Staleness, dropped links, and what an operator should see when the twin is behind. Cross-references sec-data for the vocabulary it does not re-introduce.

Staleness, and how to show it

When the link drops

Part III: Living with one

Trust and what earns it

Why an operator ignores a twin that has been wrong once, and what the literature says about rebuilding that. Leans on ch-cost's "cost of a twin nobody trusts" and pays it off.

Where this goes next

Short. Open problems, with the validation gap named as the largest. No new vocabulary.