Part III · Bottom-Up and Top-Down Knowledge Formation

The Bottom-Up Personal Composition System

People can begin with a question, fragment, or tension before they know the final task. The system lets thought develop until it has earned a form—writing, planning, inquiry, editing, or no durable result yet.

Intent engineeringSemantic productIntentional standingReintegration

1. Begin where the person is

People do not begin every consequential project with a stable ontology, complete plan, and approved workflow. They begin with:

a question;
a tension;
a source;
a fragment of language;
a decision they cannot yet frame;
an unfinished document;
a voice note;
a contradiction;
a felt direction whose object is still unclear.

The bottom-up system lets these beginnings remain cognitively appropriate. It does not force exploration into planning or thinking into a deliverable before the work has earned that form.

When purpose is incomplete, clarification can itself be the first semantic product. The system can help distinguish possible aims, consequences, and governing questions, while the person determines which purpose—if any—receives intentional standing and becomes a commitment. This is the bottom-up entry to intent engineering: purpose emerges through reasoning before it becomes top-down operating ground.

2. The sequence of formation

From attention to durable changed ground

How open attention can form into the kind of work it has earned and return as durable ground for the next attention.

flowchart LR A["Attention\nquestion · fragment · tension"] --> T["Thinking\nopen development"] T --> F["Formation\ndistinctions · claims · possible structure"] F --> C{"What has formed?"} C -->|"language"| W["Writing or artifact composition"] C -->|"route"| P["Planning"] C -->|"unknown"| E["Explore"] C -->|"bounded change"| R["Editing"] C -->|"nothing yet"| T W --> I["Integration and review"] P --> I E --> I R --> I I --> D["Durable changed ground"] D --> N["Next attention"]

Intent engineering · canonical11-the-bottom-up-personal-composition-system--m01.mmd

Text equivalent

Attention — question · fragment · tension → Thinking — open development; Thinking — open development → Formation — distinctions · claims · possible structure; Formation — distinctions · claims · possible structure → What has formed?; Writing or artifact composition → Integration and review; Planning → Integration and review; Explore → Integration and review; Editing → Integration and review; Integration and review → Durable changed ground; Durable changed ground → Next attention.

The central question preserves optionality: formation may become writing, planning, inquiry, editing, or more thinking. All productive routes reconverge at integration and review before changing durable ground, so no branch promotes itself merely by completing.

The system helps a movement find the form it has earned. It does not assume that every thought should become a task or every insight should become durable policy.

This is a developmental sequence, not an automatic promotion path. A thought can become a useful artifact without becoming reusable expertise. A recurring pattern can be named without becoming a standard. At every scale, exact source, proposal, acceptance, and later correction remain distinguishable.

3. Functional spaces for formation

Different files can carry different cognitive functions.

Purpose space

Carries why the work exists, higher-order aims, standing constraints, and the direction that judges local decisions.

Thinking space

Carries developing thought, incomplete language, tensions, questions, and distinctions that have not yet earned formal structure.

Plan space

Carries accepted branches, dependencies, commitments, checkpoints, next actions, and reopen conditions.

Work space

Carries research, analysis, designs, decisions, and artifacts in active development.

Expansion space

Carries a formed structure unfolded into a substantial deliverable.

Context and relation records

Carry purpose, state, standing, summaries, provenance, and cross-file relations.

These are functional spaces, not a mandatory folder template. A project may represent them differently while preserving the distinctions.

4. The turn as a living-file event

Each interaction can be understood as:

ATTENTION
the person focuses a file, object, or question

READ
the system interprets the input within its semantic situation

COMPOSE
the system assembles purpose-shaped context and guidance

RESPOND
one bounded reasoner performs the situated act

INTEGRATE
the result is related to the living files at its honest standing

EXTEND
authorized or candidate next lines become available

The visible response is one moment in a larger artifact cycle.

5. Thinking remains distinct from presentation

Open thinking benefits from:

provisional language;
branching;
ambiguity;
repetition;
contradiction;
discovery of the real object;
freedom from premature audience or format constraints.

Writing benefits from:

contribution;
audience;
form;
source fidelity;
sequence;
coherence;
closure conditions.

AIOS separates these movements so that a writer is not asked to discover, structure, polish, route, and evaluate simultaneously.

6. From thinking to document

An insight enters an artifact through composition rather than direct paste.

Composing formed thought into an artifact

How completed thinking is translated into a destination-aware artifact rather than pasted directly into a document.

flowchart TB T["Completed thinking movement"] --> S["Source snapshot"] S --> P["Editorial plan"] P --> Q["Which contribution belongs where?"] Q --> G["Exact source and destination ground"] G --> E["Bounded writing or editing"] E --> H["Whole-document review"] H --> A["Accepted artifact and refreshed map"]

Canonical artifact · canonical11-the-bottom-up-personal-composition-system--m02.mmd

Text equivalent

Completed thinking movement → Source snapshot; Source snapshot → Editorial plan; Editorial plan → Which contribution belongs where?; Which contribution belongs where? → Exact source and destination ground; Exact source and destination ground → Bounded writing or editing; Bounded writing or editing → Whole-document review; Whole-document review → Accepted artifact and refreshed map.

The path inserts two judgments before prose changes: what the thinking actually contributes, and where that contribution belongs. Exact source and destination ground then bound the edit; whole-document review, not local fluency, determines acceptance.

Does not establish The route does not authorize direct paste or automatic integration; source meaning, destination fit, seams, and whole-document integrity remain judgments.

This asks:

What did the thinking actually establish?
Does it belong in an existing movement, a new movement, a branch, or nowhere?
What source meaning must be protected?
What destination context changes its expression?
What seams or dependencies does the integration affect?

7. Backflow is part of the model

The movement is not one-way from thought to output.

research can change the question
planning can expose a missing distinction
writing can reveal an unsupported claim
editing can reveal a structural defect
review can reopen purpose
delegated work can challenge accepted ground
completion can expose a reusable method

Bottom-up intelligence includes the ability of later forms to teach and revise earlier ones.

8. Formation detection

The system may observe that something is becoming:

a term;
a criterion;
a principle;
a project;
a workflow;
a recurring question;
a personal reasoning pattern;
a candidate standard.

Observation is not promotion. Formation detection produces a named candidate and evidence path, not an installed rule.

The standing can develop over time:

observed occurrence
  → named pattern candidate
    → evidence-bearing proposal with scope and counterexamples
      → accepted, narrowed, deferred, or rejected knowledge

Later evidence may qualify, supersede, revoke, or reopen an accepted resource. The original occurrences and prior disposition remain available as lineage.

9. Suggestions preserve optionality

A bottom-up suggestion can offer:

continue this line;
test this distinction;
compare these files;
resolve this contradiction;
open an Explore branch;
form an outline;
compose a document;
review possible integration;
leave the thought open.

The system can prepare many internal possibilities while showing only a few—or none. Constant capability does not require constant interruption.

10. Null is a valid landing

Not every turn produces durable work, a recommendation, or a candidate pattern.

A useful after-read asks:

Did anything materially move?
Is something cohering enough to name?
Is there honestly something worth offering next?

The answer may be no. A system that always produces a next action is optimizing engagement, not intelligence.

11. Personal fitting

Over time, the system can preserve explicitly accepted local ground such as:

preferred terminology;
recurring decision criteria;
project-specific principles;
personal quality standards;
successful scaffolds;
known failure patterns;
examples that express the person's intent;
the contexts in which those preferences do not apply.

Personal fitting remains scoped and revisable. It should not become a personality profile that pre-interprets every future request.

12. The person is not the system operator

The person should not need to manage:

context-window budgets;
prompt assembly;
internal seat selection;
metadata synchronization;
thread state;
parser grammar;
source identifiers;
memory compaction.

They should be able to work in natural language and ordinary files while the system makes its important states and choices inspectable when needed.

13. Example: a vague idea becomes a research program

Attention

“Maybe organizations do not have an AI adoption problem; maybe they have a decision-memory problem.”

Thinking

The system helps distinguish adoption, trust, organizational memory, decision lineage, and authority.

Formation

Three hypotheses emerge, along with a contradiction: adoption can increase while decision quality declines.

Explore

A bounded research branch compares evidence on adoption, augmentation, and organizational learning.

Planning

The return produces an inquiry architecture, discriminating evidence, and a phased study.

Writing

One document develops the theoretical frame; another records the empirical design.

Review and integration

The project updates its governing question, preserves weak evidence as weak, and records what would falsify the hypothesis.

Candidate practice

Only after repeated use might the distinction between adoption and decision-memory become a reusable research scaffold.

The initial fragment remains traceable through every transformation without becoming a permanent constraint.

14. Bottom-up does not mean ungoverned

The bottom-up system still preserves:

source standing;
exact file identity;
bounded authority;
explicit structural effects;
person take-up;
independent review where consequence warrants it;
promotion boundaries;
supersession and correction.

Freedom applies to formation and judgment. It does not imply uncontrolled mutation.

It also does not imply that every input is safe future ground. Untrusted sources and generated interpretations remain observations or proposals until their provenance, trust domain, evidence, and scope support promotion. A candidate retrieved from memory cannot expand its own authority or permissions.

15. Failure modes

Premature formalization

Every thought is converted into a task, outline, or ontology term.

Transcript dependency

The system cannot preserve development outside the conversation that produced it.

Auto-integration

A strong response is silently inserted into an artifact without editorial judgment.

Engagement pressure

The system continually recommends further work even when the person has a coherent stopping point.

Personalization lock-in

Past preferences pre-classify new situations and suppress surprise.

Formation-as-truth

A repeated phrase becomes accepted principle without evidence or authority.

16. Research evaluation

The bottom-up model can be evaluated through:

whether users can begin with unstructured input without losing later traceability;
whether thinking-to-document separation improves source fidelity and artifact coherence;
whether formation detection produces useful candidates without excessive noise;
whether null recommendations are appropriately common;
whether backflow leads to meaningful plan or purpose correction;
whether personal fitting improves work without narrowing open inquiry;
whether users experience less pressure to manage system state.

17. Relation to the whole

The bottom-up system is not an isolated ideation tool. Memory ecology preserves the difference between an observation, a proposal, and accepted ground. Canonical files and companion metadata records give each formed artifact a durable, provider-independent landing. Document hierarchy lets a local movement find its contribution at larger scales.

Chapter 12 develops the other direction of the same cycle: accepted, scoped knowledge returning through context composition to improve later work. Human purpose and promotion authority connect the two directions; no model, registry, or workflow becomes the center of the system.

The Cognitive Agency research brief and Context Composition research brief provide the nearest evidence on meaningful cognitive acts and movement-specific scaffolding. Neither establishes the complete bottom-up sequence.