Document Hierarchy and Fractal Composition

1. Documents are nested fields of contribution

AIOS does not treat a document as a flat token sequence. A document is an artifact whose parts make different contributions to a purpose.

The relevant hierarchy is semantic, not merely typographic:

wording
  → sentence or claim
    → paragraph or local movement
      → section or movement
        → document
          → related document field
            → project
              → corpus and reusable practice

Each level can be read locally and in relation to the whole.

2. The Fractal Seed at document scales

The seed does not make the scales identical. It makes their contributions mutually intelligible.

3. Semantic hierarchy versus file hierarchy

A folder tree answers where a file is located. A semantic hierarchy answers what the file and its parts contribute.

One folder can contain:

• sources;
• thinking;
• decisions;
• plans;
• canonical artifacts;
• exports;
• workflow results;
• superseded versions.

Their location alone does not establish standing or relation.

AIOS therefore combines:

• ordinary paths for custody and navigation;
• stable identities for continuity;
• companion metadata for purpose and state;
• explicit relationships for semantic bearing;
• maps and Atlas projections for cross-scale understanding.

4. Three hierarchies must not collapse

AIOS uses hierarchy in three different senses.

Document hierarchy is primarily representational. It can guide context selection and computational decomposition, but it does not grant authority. A high-level synthesis is not truer because it sits above a source. A recursive subcall does not acquire permission to revise a document. A planning projection can propose structure without becoming the accepted Document Spine.

The Fractal Seed can recur across all three hierarchies while preserving their different functions. Purpose may travel downward and results may travel upward, but consequential authority still follows explicit grants, standing, and review.

5. The document spine

The Document Spine is the accepted compact account of:

• why the document exists;
• how its load-bearing movements develop;
• what the document establishes or leaves open.

It lives in companion metadata and can be projected as a readable Document Map.

The spine is not:

• a visible three-heading template;
• an immutable outline;
• an enforcement schema for prose;
• proof that the document is correct;
• authority to rewrite the artifact when the map disagrees.

It is a resumable account of current accepted structure.

6. Movements, not just headings

A heading states a label. A movement account states a contribution.

Weak map:

1. Background
2. Analysis
3. Recommendation

Stronger map:

1. Establishes why the apparent adoption problem is actually a trust problem.
2. Separates evidence about capability from evidence about institutional uptake.
3. Shows that governance architecture, not additional model scale, is the binding constraint.
4. Proposes a bounded pilot and names the evidence that would change the conclusion.

Contribution-oriented maps help both people and models understand how local material bears on the whole.

7. Bottom-up formation

Meaning can form from below:

The widening ladder protects formation: distinctions gather into movements before a document structure is imposed, and an artifact earns its larger role through contribution. Reaching the top makes practice a candidate; it does not retroactively formalize every fragment.

This route protects discovery:

• thinking can remain incomplete;
• repeated distinctions can become a movement;
• movements can earn a document;
• documents can reveal a project architecture;
• completed practice can produce candidate methods.

The system does not require the final taxonomy before the thought begins.

8. Top-down bearing

Structure can also guide from above:

Each downward step translates bearing for a smaller scale: project outcome shapes the document field, the spine shapes a movement, and the movement commissions a passage. The chain remains revisable from below; exact language is not dictated by altitude alone.

Top-down bearing helps the system decide:

• which document should receive a finding;
• whether a new section is warranted;
• what a passage must contribute;
• which source resolution is necessary;
• whether local polish would serve or distract from the whole.

The governing structure remains revisable when bottom-up work reveals a better form.

9. Composition is the movement between levels

Content composition is not only writing sentences. It includes:

• integrating thinking into a document;
• splitting a movement into distinct sections;
• merging duplicated contributions;
• moving a section when argument order changes;
• creating a new artifact when one document has multiple centers;
• branching a line that needs independent development;
• composing findings across source files;
• revising project relationships after an artifact changes;
• promoting a repeated method after evidence review.

Composition is therefore a structural intelligence spanning multiple resolutions.

10. Exact descent and semantic ascent

The solid path descends from outcome to source when precision is needed. The dotted path climbs through changed consequences: evidence can challenge a passage, alter its movement, reshape the document, and finally change the project route.

The system descends when precision matters and ascends when consequence matters.

An edit can begin at one sentence but require ascent if it changes a definition used across the document. A project review can begin at high altitude but require descent when a disputed claim turns on exact evidence.

11. Writing and structure are distinct altitudes

Writing altitude

Concerned with:

• language;
• argument and narrative formation;
• exact passages;
• transitions;
• local and whole-document coherence;
• accepted artifact revision.

Structure altitude

Concerned with:

• movement order;
• boundaries;
• document architecture;
• cross-file relationships;
• proposed new sections or documents;
• pending writing commissions;
• plan and project consequences.

Structure work should not silently write prose. It can record a commission for later writing or invoke an explicitly authorized structural operation.

12. Accepted and proposed structure remain separate

A planning projection can show:

• current accepted structure;
• proposed moves;
• annotations and tensions;
• pending writing;
• branches or new-document candidates.

Until authorized and applied, proposed structure must not overwrite accepted companion metadata.

observation
  ≠ proposal
    ≠ person authorization
      ≠ file effect
        ≠ accepted new spine

This protects exploratory planning from becoming unintended document mutation.

13. Single-document composition loop

The repair loop is the argument. A bounded edit returns to exact source and target ground whenever whole-document integrity fails; only a document that holds can receive an accepted spine and refreshed map.

This separates:

• idea formation;
• editorial judgment;
• exact grounding;
• prose production;
• whole-document evaluation.

The separation reduces the chance that an enthusiastic writer will reinterpret the source or that a local edit will evade whole-artifact review.

14. Cross-file composition loop

The authorization fork prevents a useful plan from becoming an effect by implication. Approved movements land in named destinations, but closure is not complete until every affected document is reviewed, lineage is written, and the project views are refreshed.

Does not establish A useful destination does not authorize source removal; a move requires explicit authorization, validated destination writing, then bounded removal.

Operation semantics remain distinct:

• integrate;
• copy;
• move;
• reorganize;
• create;
• branch;
• annotate;
• supersede.

A useful destination does not imply authority to remove the source. A move requires explicit authorization, successful destination write, validation, and only then bounded source removal.

15. Document boundaries are cognitive boundaries

A new document may be warranted when:

• one artifact contains more than one governing purpose;
• a branch requires independent evidence or method;
• audiences or completion criteria diverge;
• one line must remain open while another lands;
• the present document cannot carry the required depth without losing coherence;
• a reusable method should be separated from case-specific content.

A new document is not warranted merely because the current file is long.

16. Hierarchy and contradiction

Internal consistency does not require every document to agree.

At project scale, the system should be able to represent:

• competing hypotheses;
• sources that challenge one another;
• a decision that rejects but preserves an alternative;
• superseded versions;
• role-specific accounts with different scopes;
• a minority position that remains material;
• unresolved tension requiring a future judgment.

The hierarchy should locate contradiction and consequence, not flatten it.

17. Completion at multiple scales

Completion should be stated at the level being closed.

Activity and word count do not prove completion.

18. Document hierarchy as an interface model

The person should be able to move among:

• exact text;
• movement map;
• document synthesis;
• project relationships;
• plans and pending work;
• source and decision lineage;
• corpus and expertise views.

The interface should preserve the feeling of working with documents, not with a database ontology. Structural machinery becomes visible when it helps the person make a real judgment.

19. Failure modes

Flat-file reasoning

The model receives entire files without understanding their contributions or relationships.

Outline absolutism

The current structure becomes a rule that new thinking cannot challenge.

Local-edit myopia

A passage changes without reviewing definitions, references, transitions, or whole-document consequence.

Whole-document rewrite

A bounded revision becomes authority to regenerate the entire artifact.

Project monolith

All work is forced into one large document or one global context.

Fragmentation by length

New files are created without distinct purpose, audience, or completion conditions.

Folder-semantic confusion

Location is mistaken for standing or relation.

20. Research evaluation

The hierarchy can be tested through:

• whether models can predict movement contribution better from maps than from headings alone;
• whether exact-descent paths improve source fidelity;
• whether structure-altitude planning reduces unrelated prose changes;
• whether document-boundary recommendations improve coherence;
• whether cross-file composition preserves provenance and source standing;
• whether accepted/proposed structure separation reduces unintended mutation;
• whether the same Fractal Seed grammar supports transfer without imposing visible uniformity.

The two directions named here are developed as one circuit in Chapter 11, bottom-up personal composition, and Chapter 12, top-down operational knowledge.

21. Research connection: hierarchy extends reach but does not confer authority

AIOS uses document hierarchy to preserve contribution, exact descent, and whole-system bearing. Current research shows why relational and recursive structures can help without making them canonical. The peer-reviewed HippoRAG 2 demonstrates gains from graph-based external memory on its evaluated tasks. Two recent preprints sharpen the boundary: GraphRAG-Bench finds that graph retrieval gains are conditional and may bring substantial prompt, redundancy, and latency costs, while Recursive Language Models demonstrates processing reach across very large inputs with some trajectories requiring extensive model calls.

These findings support the use of representational and computational hierarchy when the work warrants it. They do not show that one hierarchy is universally superior, that a generated graph is complete, or that a computational decomposition has epistemic or operational authority. That separation is an AIOS architectural requirement to be tested directly.

Read deeper in the normalized context-composition brief, full context-composition memo, normalized relational-memory brief, and full relational-memory memo.