Nothing Is Lost: How to Work with AI Without Losing Your Mind

metadata

Title:	Nothing Is Lost: How to Work with AI Without Losing Your Mind
Author:	Ralph B. Holland
Affiliation:	Arising Technology Systems Pty Ltd
Contact:	ralph.b.holland [at] gmail.com
version:	0.4.0
Publication Date:	2026-01-10T26:4Z
Provenance:	This is an authored paper maintained as a MediaWiki document; clarified MWDUMP as the authoritative, permission-granting artefact governing allowable reasoning across sessionstory reflects editorial changes, not collaborative authorship.
Status:	draft

Metadata (Normative)

The metadata table immediately preceding this section is CM-defined and constitutes the authoritative provenance record for this CM-master artefact.

All fields in that table (including title, curator/author, affiliation, contact, version, update history, publication date, and binding status) MUST be treated as normative metadata.

The assisting system MUST NOT infer, normalise, reinterpret, duplicate, or rewrite these fields. Any change to metadata MUST be made explicitly by the human and recorded via a versioned update, not inferred.

Curator Provenance and Licensing Notice

This document predates its open licensing.

As curator and author, I apply the Apache License, Version 2.0, at publication to permit reuse and implementation while preventing enclosure or patent capture. This licensing action does not revise, reinterpret, or supersede any normative content herein.

Authority remains explicitly human; no implementation, system, or platform may assert epistemic authority by virtue of this license.

Nothing Is Lost: How to Work with AI Without Losing Your Mind

Abstract

People are using AI to think, write, design, and explore ideas — but almost everyone feels the same frustration: things disappear. You explain something carefully, build up a line of thought, then a tab closes, a context window fills, or a model forgets what mattered. Bigger context windows and “AI memory” do not solve this; they make it worse by turning your ideas into something the platform owns.

This paper introduces a different way to work with AI: one where your knowledge lives outside the model, on your own machine, and can be shared or withheld on your terms. Instead of relying on chat history or hidden memory, it treats every idea as something you can save, reuse, revise, and move between AI systems without losing meaning. Nothing is kept secretly, nothing is silently changed, and nothing disappears without you knowing.

The result is simple but powerful: you can think with AI without giving up your mind. Your work becomes durable, portable, and yours — no matter which tool you use, or when you come back.

1. The Problem Everyone Feels

If you use AI to think, write, or explore ideas, you have probably felt it already.

You start a conversation. You explain something carefully. You build up a chain of reasoning. The AI begins to understand what you mean. For a while, it feels like you are working together.

Then something breaks.

The browser tab crashes. The context window fills. The model suddenly “forgets.” You find yourself explaining the same thing again, often worse than the first time. Important details are missing. Subtle distinctions are lost. You know you had it right before — but now it is gone.

This is often treated as a technical problem: “Just give the AI more memory.” “Just make the context window bigger.” “Just save the chat history.”

But these fixes miss the real issue.

The problem is not that the AI forgets. The problem is that your thinking has nowhere safe to live.

Right now, most AI systems treat your ideas as temporary messages flowing through a conversation. They are not objects you can keep, edit, or reuse. They are just text that happens to be visible for a while. When it scrolls off the end, or the session resets, the thinking behind it disappears.

This creates a strange and exhausting experience. You are not really building on your own work. You are constantly re-deriving it. Each session feels like the movie Groundhog Day, where everything resets and you have to start again from memory.

People try to cope with this by:

• copying and pasting into notes,
• keeping long documents on the side,
• or trusting whatever “memory” the platform offers.

None of these are designed for how people actually think. They are fragile, confusing, and often owned by the platform rather than by you.

Before we talk about how to fix this, it helps to understand what is really going on when you type into an AI system — and why it so often sounds like it knows what it is talking about even when it doesn’t.

2. Why Everything Sounds So Confident

One of the strangest things about modern AI is how sure of itself it sounds.

You ask a question and it answers smoothly. You give it an idea and it expands it into something that looks like a polished article. You ask for an explanation and you get something that reads like a textbook.

This creates a powerful illusion: it feels like the AI knows what it is saying.

But it doesn’t.

What it is really doing is something much simpler and much stranger: it is turning patterns into words.

Inside the system there is no memory of your conversation, no model of what you meant yesterday, and no understanding in the human sense. There are just enormous statistical structures that can take what you typed and produce text that looks like a continuation.

That output is not a memory. It is not a belief. It is not even really a thought.

It is a transcription of something happening inside a machine.

This is why everything it writes sounds oddly academic. Academic writing is designed to do exactly this: take ideas and turn them into careful, well-formed text that looks authoritative. AI systems have learned that style very well.

The danger is that we start to treat this transcript as if it were the thing itself.

When an AI writes a paragraph, it feels like knowledge has appeared. But what has really appeared is just a piece of text that projects meaning. The meaning only exists because you, the reader, bring it to life.

That is not a problem by itself. The problem begins when platforms and users start treating these transcripts as if they were durable knowledge — something that can be remembered, relied on, or built upon without care.

They are not.

They are more like spoken words in a room: vivid while you hear them, gone as soon as the sound fades.

To fix this, we need a way for our thinking to become something we can keep, not just something we briefly see.

That starts by giving ideas a place to live outside the AI.

3. A New Way to Work with AI: Keep Your Knowledge, Share What You Like

Imagine if every time you thought of something important, you could save it and come back to it later. Imagine if you could share that idea with others, but keep control over it. Imagine if your ideas could move between different tools, different platforms, and different AI systems — but stay yours the whole time.

This is what Nothing Is Lost is about: creating a way to work with AI that doesn’t force you to re-derive your thinking or lose your ideas to the machine.

At the heart of this approach are two simple principles:

1. Your thoughts are externalised. These are like durable, meaningful fragments of your mind. They don’t live in the AI or the platform. They live in your own space, and only you control them. Every idea you generate is treated as an object you can keep, edit, or share.
2. You govern how your knowledge moves. The second part of the system is the governance layer: Epistemic Attributes (EA). These describe how your ideas move between states. For example:
   • Is the idea a draft or something final?
   • Is it personal or shared with others?
   • Can it be reused, expanded, or referenced?

Every time you share something with an AI or another person, you define the rules around how that idea is used. Nothing is automatically assumed or silently changed.

This makes AI work differently. Instead of being a black box that “remembers” your thoughts or resets at random, the AI becomes:

• a tool for externalising your thoughts
• a platform for sharing your knowledge
• nothing more, unless you decide to give it more.

You get to choose whether your ideas are temporarily processed in the AI or if they are made permanent, portable, and shareable.

It’s not about AI memory. It’s about controlling the memory of your thoughts.

4. The Practical Workflow: How It All Comes Together (revised)

Imagine you’re thinking, exploring, and working with AI, but instead of feeling like your thoughts are lost or erased every time you close a tab, you can start, expand, and refine your ideas as part of a dynamic, living cognitive space. In this new way of working with AI, the focus is on preserving thoughts, organising them over time, and sharing them when and how you decide.

Here’s how the process unfolds in the new system:

1. Capturing Your Thoughts

As you start typing or interacting with the AI, the system captures thoughts. Each thought is captured as an ephemeral thought — something that’s not fully formed, not permanent, and not yet finally organised.

It does not live beyond the current session, but it is kept with you on the client side, in your knowledge cache — in the first instance as an Ephemeral Thought. It is essentially ephemeral — it can be discarded, expanded, changed, or moved around with other thoughts until you decide what finally happens.

2. Thoughts Turn Into Thought Bubbles

As you continue typing, you may expand the idea. It could turn into a Thought Bubble, which is an evolving, informal thread of thinking that you haven’t yet formalised or assigned permanence to — it can remain ephemeral and be left to grow or dwindle. A Thought Bubble is a collection of Thoughts that are still in a provisional space of your mind — all captured by the AI platform.

The Thought Bubble can grow as you continue the conversation with the AI. It’s not a finished thought, nor is it yet something you want to keep permanently. It’s still ephemeral, just a working set of ideas that could either be discarded or moved forward.

These Thought Bubbles exist as unresolved threads of thinking, which may be kept in the client's cache temporarily for use, but they are not yet governed and importantly not evicted. You can carry these thoughts around in this state — or you can share them.

3. Deciding to Promote Your Thoughts

Once you decide that a thought or a thread of thinking (a Thought Bubble) is worth keeping or formalising, you can label it, make it permanent, or export it to a friend. This is where you inform the system of your intentions for your Thought, and the system updates the status of your thought with metadata that describes your intent.

These promoted Thoughts are deemed more significant than the ephemeral Thoughts. They become even more important when you export them to a file when you run save-as.

4. Aggregating and Applying Knowledge to the Universe of Discourse (UoD)

As you continue working, you collect more thoughts and you may choose to aggregate your Thoughts into a broader collection, such as a Universe of Discourse (UoD). This is where your knowledge becomes part of a larger framework or structure — for example:

• A research project
• A domain for a collection of papers
• A design specification
• A legal argument

The UoD acts as a domain of context where your ideas are not only organised but also formalised. This is where the thought truly takes on “weight” — it’s no longer just an ephemeral thought or a casual exploration. It’s part of a larger, structured body of knowledge that has a defined place, purpose, and governance.

When you aggregate your Thoughts, you are essentially formalising them and applying them in a specific way. It’s like moving from drafting to final / operations, but with the added power that you can always go back, revise, or refine your thoughts when needed.

5. The Role of the AI

The AI system is always there, helping you to process your Thoughts, to expand your Thoughts, and to suggest new connections. However, AI never takes over the process. It never asserts meaning without your permission.

Instead, the AI helps you:

• Develop Thoughts into meaningful knowledge
• Generate new Thoughts based on your existing thinking
• Connect your Thoughts to other related Thoughts
• Suggest how to aggregate or apply your knowledge, including within one or more UoD.

Conclusion

This means the AI is always in the background, helping you expand, refine, and organise your Thoughts — but it 'does not own them. You remain the one in control of your knowledge, deciding when it becomes formal, when it’s shared, and how it’s applied.

These Thoughts are stored in the app or local client cache on the client side and only transiently used by the platform when the session work is scheduled. You get to keep your Thoughts, govern your Thoughts, and the platform never owns them. This is how you work with AI and never lose your mind. This is how you avoid Groundhog Day.

With this system, AI becomes a tool for collaboration, not a replacement for your mind. The AI doesn’t remember you, doesn’t control your thoughts, and doesn’t fill up an arbitrary context window with your ideas. Instead, it’s there to help you externalise, organise, and move your knowledge in a way that fits with how you naturally think.

You own your thoughts. You control their memory. You decide how they move, grow, and apply to the world.

Appendix A — Engineering Mapping Between 'Nothing Is Lost' and Cognitive Memoisation (CM-2)

Status: Normative Applies to: 'Nothing Is Lost: How to Work with AI Without Losing Your Mind' (v0.4.0) and 'Let’s Build a Ship — Cognitive Memoisation for Governing Knowledge in Human–AI Collaboration' (CM-2 v1.3.0)

Normative Anchor

This Appendix is normatively grounded in CM-2 as specified in 'Let’s Build a Ship — Cognitive Memoisation for Governing Knowledge in Human–AI Collaboration' (Binding: Normative) and the cognitive workflow defined in 'Nothing Is Lost: How to Work with AI Without Losing Your Mind'. All terminology, object semantics, authority boundaries, and lifecycle constraints in this Appendix MUST be interpreted using CM-2 definitions of Externalised Objects (EO), Epistemic Attributes (EA), Universes of Discourse (Thought Bubbles), and the Promotion Boundary (Save As). Where any ambiguity arises between narrative in 'Nothing Is Lost' and CM-2, CM-2 semantics take precedence.

Abstract

This Appendix establishes a binding translation layer between the user-facing cognitive workflow of 'Nothing Is Lost' and the governance architecture of CM-2. Ephemeral Thoughts, Thought Bubbles, Knowledge Cache, Promoted Thoughts, and Universes of Discourse are mapped to EO, EA, EA-governed Universes of Discourse, the Memoisation Cache, and the Save As promotion boundary, respectively. The result is an implementable, auditable, and portable realisation of cognitive sovereignty without model memory.

A. Terminology Normalisation Layer

The following mapping is normative.

'Nothing Is Lost' term	CM-2 term	Normative meaning
Ephemeral Thought	Externalised Object (EO)	Non-authoritative unit of externalised cognition
Thought Bubble	EA-governed EO set (Universe of Discourse)	Bounded semantic context
Knowledge Cache	Memoisation Cache	Client-side continuity substrate
Promoted Thought	EO after Save As	Human-authoritative durable knowledge
Epistemic Attributes	Attachment Objects (EA)	Governing and qualifying metadata on EO
Universe of Discourse (UoD)	Thought Bubble	EA-defined context boundary
Save As	Promotion Boundary	Explicit human act that grants authority

B. Object-Level Semantics

B.1 Ephemeral Thought

An Ephemeral Thought is an EO with no authority.

Property	CM-2 meaning
Object type	EO
Authority	None
Memoised	Yes
Persistence	Ephemeral or persistent
EA allowed	Yes
Promotion eligible	No (must aggregate via EA-defined UoD)

B.2 Thought Bubble (Universe of Discourse)

A Thought Bubble is not a container; it is an EA-defined aggregate of EO.

Property	CM-2 meaning
Object type	EA-governed EO set
Authority	None
Memoised	Yes
Context scope	Declared entirely by EA
EO membership	Declared by EA
Promotion eligible	Yes, via Save As acting on the EA-defined aggregate

B.3 Promoted Thought (Save As)

Save As is the promotion boundary that creates authority.

Property	CM-2 meaning
Object type	EO (promoted)
Authority	Human
Memoised	No (replaced by external artefact)
Persistence	Durable
EA allowed	Yes (binding)
Canonical form	Saved CM artefact (e.g., wiki, PDF)

B.4 Knowledge Cache

The Knowledge Cache is the Memoisation Cache.

Property	CM-2 meaning
Location	Client-side
Contains	EO + EA
Authority	None
Purpose	Continuity without commitment
Can decay	Yes

C. Authority and Lifecycle Constraints

Continuity and authority are separate; only Save As creates authority.

State	CM-2 object	Authority	Cache	External
Typed	EO	None	Yes	No
In Thought Bubble	EO + EA	None	Yes	No
Aggregated	EA-defined EO set	None	Yes	No
Save As	Promotion boundary	Human	No	Yes
Rehydrated	EO	Human	Yes	Yes

D. Aggregation and Universes of Discourse

Aggregation is performed by EA. A Universe of Discourse exists when EA declare that a set of EO are to be interpreted together under a shared scope, status, and purpose. EO remain context-free and identity-stable; EA supply membership, scope, and eligibility for promotion. Save As operates on the EA-defined aggregate, not on EO individually.

What it is related to or depends on

• How it can be shared, reused, or edited

At this point, the EO is still **ephemeral** unless explicitly **saved**. The EA defines how this idea is to be treated — whether it’s still just a working idea, something provisional, or whether it’s ready to be considered final.

4. Aggregating and Applying Knowledge to the Universe of Discourse (UoD)

As you continue working, you may choose to **aggregate** your EO into a broader **Universe of Discourse (UoD)**. This is where your knowledge becomes part of a larger framework or structure — for example:

• A research project
• A paper
• A design specification
• A legal argument

The UoD acts as a **context** where your ideas are not only **organised** but also **formalised**. This is where the thought truly takes on “weight” — it’s no longer just an ephemeral thought or a casual exploration. It’s part of a larger, structured **body of knowledge** that has a defined place, purpose, and governance.

When you **aggregate** your EO into a UoD, you are essentially **formalising** your thought and applying it in a specific way. It’s like moving from **drafting** to **finalisation**, but with the added power that you can always go back, revise, or refine your thoughts when needed.

5. The Role of the AI

The AI system is always there, processing your EO, expanding on your Thought Bubbles, and suggesting new connections. However, it never **takes over** the process. It never **asserts** meaning without your permission.

Instead, the AI helps you:

• **Develop** EO into meaningful knowledge
• **Generate** new EO based on your existing thinking
• **Connect** your EO to other related thoughts
• **Suggest** how to aggregate or apply your knowledge within a UoD

This means the AI is always in the background, helping you expand, refine, and organise your thoughts — but it **does not own them**. You remain the one in control of your knowledge, deciding when it becomes formal, when it’s shared, and how it’s applied.

Conclusion

With this system, AI becomes **a tool for collaboration**, not a replacement for your mind. The AI doesn’t remember you, doesn’t control your thoughts, and doesn’t fill up an arbitrary context window with your ideas. Instead, it’s there to help you externalise, organise, and move your knowledge in a way that fits with how you naturally think.

You own your thoughts. You control their memory. You decide how they move, grow, and apply to the world.

4. The Practical Workflow: How It All Comes Together (rejected)

THIS NEEDS TO BE REWORKED BECAUSE I WANT TO DO THIS IN TERMS OF HOW THE PLATFORM SHOULD OFFER IT>

Now that we’ve outlined the theory behind **Nothing Is Lost**, let’s walk through how this works in practice. It’s about shifting from the **old way** of relying on AI memory and context windows, to a **new way** where **you control your knowledge** and how it interacts with AI systems.

Here’s how the process looks when you’re using a CM-2 compliant system:

Step 1: Externalising Your Thoughts

Every time you have an idea or a thought that you want to keep, you externalise it into an **Epistemic Object (EO)**. This is simply a meaningful chunk of your cognition that exists outside of the AI. It could be:

• a draft of an article,
• a model for a project,
• a hypothesis you're testing,
• or even an insight you've gained.

You can create these EO’s in any format — text, diagrams, lists — and they are saved in a way that **doesn’t require memory** or context to persist.

Step 2: Governing the Knowledge (EA)

Once you have an EO, you attach **Epistemic Attributes (EA)** to it. The EA describes how the EO can be used, by whom, and under what conditions. For example:

• Is this EO a **draft** or something ready to be shared publicly?
• Does it depend on or override another EO?
• Is it **provisional**, **final**, or **superseded**?
• Can it be referenced by others, reused, or is it private?

By tagging each EO with EA, you are defining **how it can move** through systems and how others can interact with it.

Step 3: Sharing and Collaboration

When you need to share your knowledge — either with other collaborators or with the AI — you **decide how it’s shared**.

• If you share an EO with the AI, you can choose whether it’s for **temporary processing** (just for this session), or if it should be **stored** for future use.
• If you share it with a collaborator, you define what they can do with it — edit, reference, expand, or just view it.

This is where **real-time collaboration** happens. Multiple people can contribute EO’s to a shared session. But instead of the AI owning everything, you have the ability to track and govern what’s shared and what isn’t. The AI simply **processes** and **synthesizes** knowledge, without controlling it.

Step 4: Rehydrating and Using Your Knowledge

Whenever you need to come back to an idea, you can retrieve it from your EO store. This is the opposite of relying on a memory that fades or disappears — your knowledge is preserved as **governed, durable objects**.

The AI can rehydrate these EOs, process them, and generate new insights, but it never **replaces** or **overwrites** them. This allows you to **build** upon your own knowledge without fear of losing it or having it altered behind your back.

You can switch between platforms, collaborate with others, and use different tools — but your knowledge stays **stable** and **portable** across them all.

Appendix B — Thought Lifecycle Under Cognitive Memoisation

Nothing Is Lost isn’t about creating new features for AI. It’s about creating a new relationship between humans and machines: one where the AI acts as a tool for working with your knowledge — not owning it.

This workflow creates a frictionless way to work with AI without losing your thoughts in the process. You now own your knowledge, and can use it however you see fit, without worrying about it disappearing or being silently mutated.

Status: Normative Applies to: 'Nothing Is Lost: How to Work with AI Without Losing Your Mind' (v0.4.0) and 'Let’s Build a Ship — Cognitive Memoisation for Governing Knowledge in Human–AI Collaboration' (CM-2 v1.3.0)

Normative Anchor

This Appendix is normatively grounded in CM-2 as specified in 'Let’s Build a Ship — Cognitive Memoisation for Governing Knowledge in Human–AI Collaboration' (Binding: Normative) and the cognitive workflow defined in 'Nothing Is Lost: How to Work with AI Without Losing Your Mind'. All lifecycle semantics, object transitions, and authority boundaries in this Appendix MUST be interpreted using CM-2 definitions of Externalised Objects (EO), Epistemic Attributes (EA), Universes of Discourse (Thought Bubbles), and the Promotion Boundary (Save As). Where any ambiguity arises between narrative in 'Nothing Is Lost' and CM-2, CM-2 semantics take precedence.

Abstract

This Appendix specifies the lifecycle by which individual thoughts are captured, related, governed, aggregated, and promoted into durable knowledge under CM-2. It defines how Ephemeral Thoughts in 'Nothing Is Lost' become Externalised Objects (EO), how Thought Bubbles arise through EA-governed aggregation, and how Save As creates authority without relying on model memory.

1. Lifecycle Overview

A thought does not become knowledge by repetition, fluency, or accumulation. Under CM-2 it moves through a governed lifecycle of capture, aggregation, and explicit promotion.

2. Stage 1 — Capture of a First Thought

When a human enters a thought, a new EO is created. The EO contains the text or structure supplied. The EO has an identity and timestamp. The EO has no scope, status, or authority. The EO is memoised in the client cache to prevent rediscovery.

This corresponds to an Ephemeral Thought in 'Nothing Is Lost'.

3. Stage 2 — Emergence of Related Thoughts

When additional thoughts are entered, additional EO are created. At this stage the EO remain independent and context-free. No relationship exists between EO until EA are created.

4. Stage 3 — Formation of a Thought Bubble

When the human recognises that multiple EO belong to a shared working context, one or more EA are created. These EA declare:

• which EO participate
• the scope of interpretation
• provisional or draft status

The EA define a Universe of Discourse (Thought Bubble) over the selected EO. The EO remain context-free; the EA carry all aggregation and scope.

5. Stage 4 — Evolution and Pruning

EO may be added to or removed from a Thought Bubble by updating EA. Rejected or superseded EO remain memoised but are no longer governed by the Thought Bubble EA. No deletion or rewriting is required to change membership or meaning.

6. Stage 5 — Aggregation for Promotion

When the human judges a Thought Bubble to be worth preserving, the EA governing that Universe of Discourse are marked as a promotion candidate. The aggregate of EO defined by those EA becomes the promotion target.

7. Stage 6 — Save As (Promotion Boundary)

Save As is executed by the human. The EA-defined aggregate of EO is externalised as a durable artefact. The promoted EO acquire authority. Non-promoted EO and EA remain non-authoritative and may decay.

8. Stage 7 — Stand-Down and Rehydration

After promotion, the session may end. The memoisation cache may be cleared. The promoted artefact remains authoritative and portable. In a new session, EO may be rehydrated from the artefact and re-enter the memoisation cache with their authority intact.

9. Governance Invariants

Continuity is provided by EO memoisation. Context is provided by EA. Authority is created only by Save As. No memory, summarisation, or inference is permitted to replace these mechanisms.

Appendix C — Memoisation Cache, Working-Set, and Eviction Semantics

Status: Draft Scope: CM-1 compatible; CM-2 aligned Binding: Informative (subject to refinement)

Normative Anchor

This Appendix is normatively grounded in CM-2 as specified in 'Let’s Build a Ship — Cognitive Memoisation for Governing Knowledge in Human–AI Collaboration' and the cognitive workflow defined in 'Nothing Is Lost: How to Work with AI Without Losing Your Mind'. All cache, working-set, and eviction semantics defined herein MUST be interpreted using CM-2 definitions of Externalised Objects (EO), Epistemic Attributes (EA), Universes of Discourse (Thought Bubbles), and the Promotion Boundary (Save As). For the purposes of this Appendix: EOe (Enduring Store) is the CM-2 Memoisation Cache (the Nothing Is Lost Knowledge Cache), and EOc (Working-Set Cache) is the subset of EO and EA currently projected into the active interaction surface (the context window). This Appendix governs operational continuity and projection only. It MUST NOT create, infer, or substitute epistemic authority, which remains exclusively defined by Save As under CM-2. Where any ambiguity arises between cache mechanics and CM-2 governance rules, CM-2 semantics take precedence.

Portability and Non-Enclosure Clause

Compliance with this Appendix and with CM-2 does not require any specific cache architecture, eviction policy, storage backend, or runtime mechanism. An implementation is CM-2 compliant if and only if it preserves identity-stable Externalised Objects (EO) and preserves Epistemic Attributes (EA) governing those EO and can serialise and deserialise EO and EA without semantic loss. All other aspects of cache design including working-set management, eviction heuristics, persistence layer, indexing, and projection into an interaction context are implementation-defined. No vendor, platform, or system may assert epistemic or economic control by imposing a proprietary cache format, opaque memory system, or non-exportable working set. Any system that prevents EO and EA from being exported, rehydrated, or moved between implementations is not CM-2 compliant regardless of performance or convenience. This clause exists to ensure that cognitive sovereignty is preserved across tools, time, and vendors.

X.1 Purpose

This appendix defines how large numbers of Epistemic Objects (EO) are managed when projected into a finite interaction context window. It ensures that scale does not collapse meaning and that operational caching does not become semantic authority.

X.2 Two-Level Substrate Model

CM operates with two operational layers.

Layer	Name	Role
EOc	Working-set cache	The Epistemic Objects currently active in the interaction surface (context window).
EOe	Enduring store	The durable Epistemic Object repository (disk, browser cache, file store, or external database).

EO identity is preserved across both layers. Movement between layers does not create new Epistemic Objects.

X.3 EA as the Control Plane

All movement, visibility, and protection of EO is governed by Epistemic Attributes (EA).

EA may define: pinned = true or false eviction_class = protected or normal or evictable uod_binding = UoD identifier promotion_state = parked or candidate or canonical

EA rules override any cache heuristic. No EO with pinned true or eviction_class protected may be evicted from EOc without explicit human action.

X.4 Cache Heuristics

Within the set of EA-eligible EO, implementations may apply operational cache policies such as: Least Recently Used Least Frequently Used Decayed Frequency Hybrid scoring

Implementations should maintain transient fields such as: last_access_at access_count decayed_access_score

These fields are operational only and must not be treated as semantic truth.

X.5 Admission Control

An EO must not enter EOc unless: EA allows it, and it has been either referenced by the human, promoted by explicit action, or exceeded an implementation-defined access threshold. This prevents accidental flooding of the working set.

X.6 Eviction Safety

Eviction from EOc: must not remove the EO from EOe, must not violate EA pinning or protection, must be reversible by rehydration. Eviction is a performance optimisation, not a loss of knowledge.

X.7 Human Override

The human may at any time: pin or unpin EO, move EO between EOc and EOe, or override cache decisions. No automatic process may negate human intent.

X.8 Design Rationale

The context window is finite. Meaning must not be. By separating operational cache mechanics from epistemic governance, CM allows thousands of Epistemic Objects to exist without collapsing into noise or being silently forgotten.