Registration, Promotion, and the Self-Evolving Knowledge System

1. Self-evolution is governed semantic change

AIOS becomes more capable through use because the durable environment around model judgment improves. Stronger artifacts survive. Decisions retain their reasons and reopening conditions. Relationships become explicit. Workflows, roles, and standards develop through evidence. Obsolete ground is retired or superseded. Future context can therefore draw on more useful local knowledge.

This is self-evolution in AIOS: experience becomes candidate knowledge, candidate knowledge undergoes review, and accepted knowledge changes the ground available to later work. The person governs each change of standing and scope.

The system is not learning by silently treating everything the person does as training data or belief. Evolution occurs through visible semantic products, explicit evidence, attributable decisions, and reconstructable lineage.

The complete promotion mechanism is part of the product's future vision. It will combine reasoning over accumulated work with metadata pattern detection: recurring structures, emerging tags, gaps, and relationships can be compared with the current knowledge system and surfaced as specific candidates for promotion. The process described here is the conceptual architecture for that development; the commands, detection methods, and enforcement machinery can continue to evolve inside the product.

2. Practice compounds through a flywheel

Accepted expertise shapes planning and situated work. That work produces artifacts, outcomes, failures, and traces from which patterns can be observed. A pattern becomes a candidate with evidence and counterevidence. Human review determines whether it is accepted, narrowed, deferred, or rejected.

The flywheel compounds because knowledge returns to practice. Its promotion boundary prevents frequency from becoming force.

This figure presents promotion from the registration and standing perspective. Chapter 12 presents the same movement from the operational-knowledge side; this chapter develops standing changes, route convergence, scope decisions, and continuing governance in greater detail.

3. Many kinds of knowledge can develop

The promotion system can develop several families of reusable knowledge:

• Reasoning practices: a scaffold, one-stage method, multi-stage workflow, operational role, or domain perspective.
• Artifact practices: a template, exemplar, evidence standard, or decision standard.
• Semantic resources: a scoped principle, controlled term or relation, or metadata convention.
• Governance practices: a project practice, organizational standard, or deprecation rule.

Each class makes a different claim. A memorable phrase does not yet define a method. A method that works in several cases does not thereby become a general principle. A useful template does not prove that its underlying process is optimal.

Promotion begins by identifying the smallest reusable contribution the evidence supports.

4. Promotion contains several distinct events

observed
  → candidate with supporting and challenging evidence
    → dismissed, deferred, or approved for development
      → developed into a defined resource
        → produced and validated
          → rejected or integrated
            → accepted and registered
              → revised, narrowed, retired, superseded, or restored

“Promote” names the overall movement, not one ambiguous button. Approving development, authoring a resource, validating its form, integrating it into the knowledge system, accepting its authority, registering its identity, and activating it in later context are different consequences.

This explicit lifecycle makes standing inspectable at every point. A well-developed candidate can still be rejected. A registered resource can remain inactive. An accepted practice can later narrow or retire.

5. A candidate carries its own case

One illustrative candidate record can take this form:

candidate:
  type: scaffold | method | workflow | role | principle | template | exemplar | term
  name: "..."
  observed_contribution: "What appears reusable and why it matters"
  current_standing: observed
  proposed_scope:
    projects: []
    domains: []
  evidence:
    instances: []
    outcomes: []
    counterexamples: []
  reusable_core: "What transfers beyond the originating cases"
  case_specific_elements: []
  limits_and_misuse: []
  proposed_future_bearing: "What later context or validation would change"
  provenance: []
  review_authority: pending
  related_candidates: []

The candidate record separates observation from proposed contribution and future effect. Repeated instances add evidence, while review still determines what those instances justify.

6. Pattern detection starts wide and proves itself by descent

Metadata can locate completed documents, stage records, decisions, annotations, revisions, source receipts, completion accounts, project relationships, emerging tags, recurring repair types, gaps, and repeated failures. These signals help the system notice where practice may be forming and compare the emerging pattern with what the current knowledge system already contains.

Pattern review then descends into the exact artifacts, passages, evidence, outcomes, and counterexamples. An index can identify recurrence; it cannot establish why the recurrence occurred, whether it improved the work, or which part deserves transfer.

This descent protects promotion from the seduction of neat metadata. Evidence remains attached to the actual work.

7. Reuse requires a recognizable occasion and boundary

A promotion case establishes the recurring need, the independence of its examples, the distinctive contribution of the proposed resource, the conditions under which it helped, meaningful variation across cases, counterexamples, misuse conditions, situational judgment that must remain open, supported scope, future effect, remaining validation, provenance, and privacy boundary.

The central test is practical: can a future person or composed context recognize both when the resource bears and when it does not? Reusability without an occasion becomes context noise. Reusability without a boundary becomes a cage.

Example: the Chapter 11 candidate earns a narrow scope

Chapter 11 begins with the observation that a team has more reports than before but fewer decisions anyone can explain. Suppose later projects produce several independent instances in which decision records preserve conclusions while losing criteria, alternatives, and authority. Those instances support a candidate diagnostic scaffold: when output volume rises while decision rationale becomes harder to recover, compare production activity with the preservation of decision ground.

The evidence also includes a counterexample. In one project, decisions remain well documented; the apparent loss of explainability comes from rapid policy change that made yesterday's rationale obsolete. Review therefore rejects the broader claim that increased reporting causes poorer decisions. It narrows the resource to occasions where a team needs to distinguish output volume from recoverable decision quality.

After person approval, the scaffold is authored with that occasion, its comparison method, the counterexample, its evidence boundary, and the conditions under which another explanation should be tested. Registration gives it identity, scope, provenance, and a revalidation path. Its first activation occurs only when a later planning or review context presents the same tension. The scaffold guides inquiry without deciding the cause in advance, and the outcome of that use returns as further evidence about its fit.

8. Evidence-derived and directly authored knowledge are co-equal

Some knowledge emerges from completed work. Other knowledge enters because a person, team, researcher, or partner already has a method or standard worth preserving.

Both routes pass through identity, provenance, scope, limitations, versioning, review, and retirement. Direct authorship does not need to imitate recurrence, and recurrence does not outrank deliberately developed expertise. Their evidence histories differ; their governance obligations converge.

9. Project truth and reusable practice remain separate

Knowledge remains project-local when it depends on one source set, temporary conditions, a particular stakeholder agreement, one artifact, local vocabulary, a unique exception, or a context unlikely to transfer.

It becomes a credible promotion candidate when its identity separates from case content, its contribution appears across relevant instances, counterexamples have been considered, its occasion and limits can be recognized, future benefit justifies maintenance cost, and the appropriate authority accepts its scope.

Changing a file's location never changes its epistemic standing. Generality is a claim that requires evidence.

10. Registration makes knowledge resolvable

Registration gives a resource a stable identity, type, version, owner, scope, provenance, evidence, limitations, status, canonical path, relationships, validation record, and acceptance record. The system can now locate it, compare versions, include it among exact candidates, and trace later use.

Registration creates availability. Activation occurs only when the current occasion matches, scope applies, version and source remain current, contraindications do not dominate, local configuration permits use, and context composition selects the resource at the right resolution.

This distinction lets the registry grow without forcing every accepted practice into every judgment.

11. Authority remains local to scope

Use at one level does not create authority at the next. Expanding scope makes a new claim about transfer and requires its own evidence, review, and acceptance.

Organizational authority can be delegated across these scopes. A project lead, team owner, organizational owner, or package owner can promote knowledge within the jurisdiction the organization has assigned to that role. This delegation is how promotion scales across an organization rather than returning every decision to one central authority.

12. Accepted knowledge remains subject to evidence

A resource can become stale when source standards change, counterexamples accumulate, operating conditions shift, a new model changes behavior, use expands beyond the evidence, an external obligation changes, or the original purpose no longer bears.

The system can reduce routine exposure, open revalidation, revise scope, retire the resource, or supersede it. The prior form remains in lineage with the reason its standing changed.

Competing practices can coexist where different occasions justify them. A rapid review may serve a low-consequence internal draft, while an independent two-pass review serves contested public claims. Internal consistency lies in preserving the distinction and selecting accordingly.

13. Evolution remains local-first

Canonical knowledge stays in inspectable files. This supports person ownership, provider change, exact provenance, selective sharing, independent review, local deletion and retirement, package distribution, and reconstruction without hidden service memory.

Remote models may still receive selected composed context. Local-first identifies custody of the durable corpus and authority over its use; it does not imply that every inference remains on-device.

Boundary: the system evolves through promotion, not ambient inference

AIOS can notice recurrence and prepare candidates. It cannot infer a person's beliefs, install a rule, expand scope, or activate a resource merely because a pattern appeared often. Self-evolution is the governed compounding of explicit work into accepted reusable ground.

14. Failure conditions and evaluation

The principal failure conditions are distinct:

• Ambient learning: ordinary behavior silently changes accepted knowledge without promotion review.
• Repetition bias: frequency is mistaken for truth, value, or transfer.
• Authority inferred from folder location: storage position substitutes for evidence, acceptance, and scope.
• Evidence-free standardization: a local success becomes general practice without a transfer case.
• Registration treated as enforcement: availability becomes automatic activation or binding force.
• Stale expertise: a resource remains active after its sources, conditions, or purpose have changed.
• Invisible provenance: later users cannot recover where a practice came from or why it was accepted.
• Candidate overload: review volume exceeds the person's capacity to judge promotion responsibly.

Evaluation measures candidate precision and usefulness, evidence quality, false and missed promotions, transfer to independent cases, rejection and inspection controls, scope discipline, revalidation and retirement, competing-practice preservation, context and instruction cost, and improvement in later work relative to governance burden.

A self-evolving knowledge system learns in public. Its growth is visible in the ground it changes, the authority that accepted the change, and the future judgments that become better because of it.