---

1. Prototype 1 never completely disappears

Return to the car factory. Prototype 1 is designed, tested, manufactured and driven. The organisation learns that a component wears too quickly, that one control confuses drivers and that another feature works better than expected. Engineers change the design. Production changes. Prototype 2 appears.

Prototype 2 is not simply a new object sitting beside Prototype 1. It contains a history. Its dimensions, materials, software, tolerances and interfaces carry decisions made because an earlier version met reality.

In that sense, Prototype 1 survives inside Prototype 2 as selected information.

This is why the geometry is a spiral rather than a circle. The system revisits similar jobs—sense, model, decide, act, measure, repair—but it need not revisit them from the same starting state.

2. Versioning requires memory

A loop without retention can correct the present and still fail to improve the future. The repair disappears when the operator leaves, the lesson is forgotten, the log is lost, the supplier changes, the institution turns over or the next generation does not know why the old constraint existed.

Memory is therefore not a decorative archive attached to intelligence. It is part of the ratchet that stops every loop from beginning again at zero.

The Intelligence library has approached this from several directions. Working Memory holds the present construction. Source Monitoring preserves where knowledge came from. Distributed Memory lets groups remember who or what carries particular knowledge. Schema Revision alters reusable structure. Belief Revision changes the model while preserving what still survives. Transfer and Recomposition asks whether retained structure remains useful in a changed setting.

Versioning is what happens when those changes are made durable enough to influence a later state.

3. A version is not automatically better

Version numbers can create a dangerous illusion. V2 sounds superior to V1 merely because two comes after one.

But systems can regress. They can learn from noisy evidence, assign credit to the wrong cause, optimise a convenient metric, remove redundancy that later proves essential, overfit one environment, forget an old failure mode or trade a visible problem for a hidden one.

This is why the closed loop needs more than feedback. Prediction Error must be interpreted. Precision Weighting must distinguish reliable signals from noisy ones. Outcome Evaluation must avoid confusing luck with process quality. Credit Assignment must locate what actually contributed. Learning Rate Control must decide how much one observation deserves to change the model. Disconfirmation Search must look for evidence capable of breaking the favoured answer.

Versioning records change. Intelligence has to earn improvement.

4. We version ourselves

A learner makes the idea personal. A child meets fractions for the first time. The first representation is crude. A half is “one of two pieces.” Then the learner meets unequal partitions and discovers that not every pair of pieces represents halves. Number lines arrive. Equivalent fractions arrive. Ratio arrives. Algebra later changes the available structure again.

The learner does not merely accumulate pages. Their internal model is repeatedly versioned.

Good education deliberately creates these loops: attempt, evidence, explanation, correction, retrieval, changed problem, transfer. A marked answer is useful only if information from the answer can alter the next attempt. A lesson becomes durable when its change survives beyond the immediate performance.

This connects versioning to metacognition. A learner who can notice, “That method worked only because the numbers were convenient,” has begun to distinguish a local success from a transferable structure. The next version of the learner carries a more conditional map.

5. We version our tools

Human intelligence does not remain inside the skull. It enters notebooks, instruments, checklists, standards, diagrams, machines, software, buildings and procedures.

A bridge design contains accumulated structural knowledge. A laboratory protocol contains remembered ways experiments can become contaminated. An aviation checklist can preserve lessons bought at terrible cost. A mathematical notation can compress centuries of conceptual development into a representation a student can learn in an afternoon. A software library can make yesterday’s difficult operation today’s ordinary function call.

These are externalised versions of learned structure. They allow intelligence to modify the environment so later intelligence begins from a different platform.

6. We version organisations

An organisation learns only when experience can change more than one person’s memory.

A recurring manufacturing defect may lead to a revised drawing, a new inspection step, a supplier requirement and a changed training procedure. A hospital incident may alter a protocol and handoff. A school may discover that a transition repeatedly loses information and redesign the handover. A company may change its release gate after a failure escaped testing.

The organisation has been versioned when the new structure changes what happens even after the original people are absent.

This is also where organisations can fail spectacularly. They may gather feedback without routing it to authority. They may write lessons that nobody retrieves. They may preserve rules after the reason for the rules has disappeared. They may reward local optimisation that damages the larger system. Versioning therefore needs provenance, ownership, review and the ability to retire obsolete knowledge.

7. We version civilisation

At civilisation scale, the same mechanism becomes enormous.

Writing allows knowledge to outlive a speaker. Libraries preserve and organise it. Schools reproduce capability in new minds. Science creates methods for making claims answerable to evidence. Engineering turns tested knowledge into repeatable artefacts. Standards make interfaces interoperable. Law records rules and procedures. Museums and archives preserve traces that allow later generations to reconstruct what earlier generations knew, made and valued. Digital networks accelerate copying and retrieval. Artificial intelligence creates new machinery for transforming and navigating accumulated representations.

Civilisation can therefore be read as a giant, imperfect versioning system. It inherits a world, acts upon it, records some consequences, forgets others, preserves selected knowledge and hands a modified world to people who did not participate in the earlier loops.

This is the civilisational ratchet. The next intelligence does not have to begin at the bottom of every spiral.

8. The world also versions us

The relationship is reciprocal. We change the world, and the changed world changes the conditions under which we learn.

A city built around cars produces different habits, distances and planning problems from a city built around walking and transit. Search engines change how people retrieve information. Calculators change which calculations need to be carried mentally. Smartphones change attention environments. Writing changes what a culture can preserve. Schools change the baseline knowledge expected of children. AI changes which cognitive tasks can be delegated cheaply.

So the loop is not simply mind → world → mind. Once action changes the environment, the next mind encounters a different environment.

Now both sides of the relationship are versioned.

9. Path dependence: every version changes what is easy next

Versioning creates path dependence. Once infrastructure, habits, standards and representations exist, some next moves become cheap and others become expensive.

A road network makes certain journeys easy. A programming ecosystem makes certain software easy to build. A notation makes certain ideas easy to express. A school curriculum makes some later learning easier because prerequisites already exist. An institution can accumulate procedures that make familiar cases fast while making unusual cases awkward.

This means intelligence must sometimes distinguish between “best from here” and “best if we could redesign the starting point.” That is one reason creativity matters. Closed-loop optimisation can improve the current path; creative exploration can ask whether another path should exist.

10. The dots are versioned too

Our dot model changes under versioning. At V1, the intelligence can see a particular set of distinctions and relationships. After experience, some dots strengthen, some disappear, some split into finer distinctions and some compress into larger structures. Creativity may