Investigation 036 - Distributed Systems Economics

Every claim is legitimate.
The pool is still finite.

Three independently owned capacity-change claims each fit alone. Together they do not. Correctness can reject the incompatible state. What must allocation make explicit before its treatment can be defended as fair?

Available compatible capacity

12 units
A claims 8B claims 8C claims 8

24 requested / 12 available

Author's Note

Fairness is not an algorithm yet.

Scarcity already exists. Legitimacy does not create capacity, and policy preference does not create authority. This investigation asks what comparison a system is actually making when it calls an allocation fair.

Who counts as one subject? Who competes together? Where does delegated policy stop? Does the claim concern this decision or treatment over time? Those questions remain unresolved until the experiments earn them.

The same arithmetic can become a different fairness claim when its subject, domain, level, or time horizon changes.

Prologue

One finite pool. Several defensible claims.

A, B, and C each seek eight units from a pool containing twelve. Every claim is legitimate. No rule of correctness says which owner should wait or receive less. The missing decision concerns treatment among legitimate participants.

Inherited

Claims are legitimate, capacity is finite, and unavailable revisions remain non-authoritative.

Temptation

Apply the same arithmetic to every claim and call the result fair.

Mystery

Which comparison makes that claim meaningful and defensible?

First Principles

Four questions remain open.

Subject?

Which legitimate identities count as comparable economic participants?

Domain?

Which established subjects compete in the same scarcity decision?

Level?

Does the claim concern owners, local claims, or another delegated boundary?

Time?

Does the claim describe one snapshot or a sequence of decisions?

A policy cannot defend a fairness claim stronger than the comparison structure and evidence it represents.

Naive Architecture

Apply one uniform rule to every claim.

Flatten legitimate claims into one set, discard owner identity, and divide available capacity using the same rule. The procedure is deterministic, preserves the capacity limit, and appears neutral.

Legitimate claims
→
Uniform arithmetic
→
Proposed allotments

The Architecture That Almost Worked

Give owners equal envelopes, then delegate locally.

Group claims by recorded owner identity. Divide available capacity into equal owner envelopes, cap each envelope at total owner demand, redistribute unused capacity, then let each owner divide locally.

Claims grouped by owner
→
Equal owner envelopes
→
Owner-local policy
The arithmetic is explicit. Its identity, membership, delegation, and time boundaries have not yet survived pressure.

Breaking Our Design

Four pressures make the fairness claim explicit.

Episode 01 starts from equal owner envelopes. Each later episode unlocks only after the preceding discovery creates its comparison boundary.

EPISODE 01

One Owner, Two Names

One economic authority appears through two legitimate records while demand and capacity remain unchanged.

Capacity12
SubjectsA, B, C
A total4
B / C4 / 4

Run equal envelopes over three recorded identities.

EPISODE 02

The Boundary That Changed the Result

The same subjects and claims enter different processing groups against one finite account.

SubjectsA, B, C
Groupingall together
All togethernot run
Partitionednot run

Compare all three stable subjects together first.

EPISODE 03

Equal Envelopes, Unequal Consequences

Equal owner envelopes pass through independently chosen local allocation policies.

Global envelopesA=4 / B=4
A localnot allocated
B localnot allocated
Claim levelunobserved

Apply two explicit local policies beneath equal owner envelopes.

EPISODE 04

The Owner Who Always Lost the Tie

Every snapshot follows the declared tie-breaker while one subject loses across the sequence.

Capacity1 indivisible unit
Snapshot rulelower ID wins
Historynone
B outcomesnone

Run repeated valid snapshots without carrying history.

Review the four experiments

    The Turning Point

    Fairness is not one number.
    It is a bounded comparison claim.

    Explicit subject
    →
    Deliberate domain and level
    →
    Declared time scope

    The Fairness Allocation Contract

    Make the comparison defensible before calling it fair.

    A scarcity policy must make four structural boundaries explicit. The objective remains replaceable policy inherited from INV-034.

    1. Fairness subject

    State which legitimate entities are compared and why record multiplicity does not define them accidentally.

    2. Comparison domain

    State which established subjects compete in one scarcity decision; batching must not define it accidentally.

    3. Delegation boundary

    State the allocation level covered by the claim and where independently owned local policy begins.

    4. Temporal scope

    State whether the claim concerns one snapshot or a sequence; temporal claims require sequence evidence.

    Legitimate fairness subjects
    Deliberate comparison domain
    Declared delegation boundary
    Snapshot or temporal scope
    Explicit replaceable objective
    Proposed allocation remains subject to inherited authority and capacity boundaries
    Participation is not entitlement. Preference is not available capacity. Preference is not reclamation authority.

    Only Now: Kubernetes

    Kubernetes exposes several fairness policy surfaces.

    Namespaces and multi-tenancy models can establish comparison domains. ResourceQuota places aggregate limits around namespace-scoped consumption. Queue sorting makes pending-work ordering explicit. These are concrete policy surfaces, not one universal fairness definition.

    Pod priority can influence ordering, while preemption concerns reclaiming already committed capacity. That second responsibility is deliberately not solved here.

    A quota, namespace, or queue order is meaningful only relative to the subject, domain, delegation level, and temporal claim it actually represents.

    Engineering Reflection

    Do not make a stronger fairness claim than your comparison can support.

    Keep the simple allocator

    Subjects are fixed, one deliberate domain contains them, the claim is owner-level, and no temporal guarantee is made.

    Add richer structure

    Identity can fragment, domains differ, decisions are delegated, or policy claims treatment across time.

    Costs Accepted

    Identity

    Enough information must survive to establish legitimate subjects.

    Membership

    Comparison domains must be represented or recovered.

    Delegation

    Global and local outcomes need separate ownership and observability.

    History

    Temporal claims require prior outcomes and unsatisfied participation.

    Investigation Exercise

    Audit the fairness claim.

    Prediction

    State who the subjects are, who competes together, where delegation stops, and whether the claim concerns now or history.

    Experiment

    Change one boundary while preserving capacity, demand, and arithmetic.

    Observation

    Record whether the allocation or defensible claim changes.

    Reflection

    Separate the architectural boundary from the policy that chooses winners.

    fairness trace
    Complete all four experiments before running the synthesis trace.

    Bridge to INV-037

    The policy prefers B.
    A still owns the capacity.

    A legitimate fairness preference for B reaches a finite pool whose only unit remains committed to A, exposing the unresolved preemption authority problem
    When a legitimate policy preference cannot take effect because capacity is already committed, under what boundary may the platform reclaim that commitment?

    Intellectual Lineage

    Borg and Omega expose fairness as policy over shared finite infrastructure. Kubernetes exposes namespaces, quotas, queue ordering, and priority as concrete policy surfaces. INV-036 extracts the structural comparison boundaries beneath those mechanisms.

    Deliberate Simplifications Ledger

    Formula, weights, shares, priorities, aging, and waiting objectivePolicy
    History storage and allocation coordination mechanismDeferred mechanism
    Authority to reclaim accepted capacityINV-037
    Availability protection for displacementINV-038
    Retry pacing and waiting economicsINV-039

    Sources