Entity Economy Empirical Research Proposal
A companion empirical program for the Entity Economy paper series
Status: Research proposal; protocol specification; no empirical results claimed.
Relationship: Non-numbered companion to The StegVerse Entity Economy Volumes I–III and informed, where relevant, by the separate Convergence / Comparative Signal Treatment. It is not Volume IV and does not alter any existing paper identity.
Purpose: Define reproducible tests of material falsifiable propositions arising from the series without making completion or success of those tests a prerequisite for publication of the numbered papers.
Owner: Publisher research/editorial custody.
Date: 2026-09-30.
Abstract
The Entity Economy series develops a governed economic architecture, examines persistent human and AI entity participation, and applies those foundations to the future economics of increasingly abundant intelligence, infrastructure cost incidence, individual data value and equitable participation. This companion proposal converts material falsifiable propositions from that series into a staged empirical program.
The program compares functionally equivalent outcomes rather than nominal model calls. It measures total user-borne and system resource cost, governance overhead, evidence reuse, quality, latency, reliability, infrastructure incidence, attributable participation, authorized data use and actually settled compensation. Negative controls, null-result reporting and replication are first-class requirements so that governance, sovereignty, compensation or network-learning mechanisms are not assumed beneficial by construction.
This proposal contains no benchmark result and does not claim that StegVerse has demonstrated lower costs, superior outcomes, lawful data markets, user income, operational rights enforcement or economic advantage.
1. Research objectives
RQ1 — Economic efficiency. For matched admissible AI workloads, does governed execution change total user-borne cost per completed equivalent outcome after governance, verification, failure and evidence-custody costs are included?
RQ2 — Infrastructure incidence. When workloads move among centralized, device-side and hybrid execution, who bears capital, energy, network, depreciation, maintenance, reliability and compliance costs? Orbital compute may be modeled prospectively only from independently supportable inputs; it is not an observed production condition by assumption.
RQ3 — Evidence reuse. Under valid scope, freshness, provenance and authority constraints, can admissible retained evidence reduce duplicate execution or verification cost without degrading pre-registered outcome quality or violating purpose limitations?
RQ4 — Data value and compensation. Under voluntary, purpose-bound and legally reviewed data-use arrangements, can attributable authorized use produce measurable actually settled compensation while preserving the declared custody and revocation model?
RQ5 — Distribution. How are benefits, costs, delays and risks distributed across entities with different usage levels, device ages, network access, electricity prices, accessibility needs, jurisdictions and data marketability?
RQ6 — Entity-conditioned utility. Does authorized entity context or governed relationship/network composition improve a pre-registered task outcome relative to an equivalent context-poor or non-conditioned baseline, and at what privacy, governance and compute cost?
2. Pre-registered hypotheses
- H1 — Net user economic benefit: for at least some matched workload classes, governed execution can reduce total user-borne cost per completed equivalent outcome after governance overhead.
- H2 — Evidence-reuse efficiency: eligible evidence reuse can reduce duplicate compute or verification expenditure without materially degrading quality, latency, reliability, safety or admissibility.
- H3 — Cost shifting: provider or inference efficiency can fail to improve user welfare when device depreciation, electricity, network, subscription, retry, verification or other costs shift to the user.
- H4 — Authorized data compensation: where a valid market transaction exists, purpose-bound attributable data use can produce measurable settled compensation without treating custody, consent, payment and lawful use as interchangeable.
- H5 — Heterogeneous incidence: economic effects differ materially across user/entity cohorts and can be obscured by aggregate averages.
- H6 — Context-conditioned utility: appropriately scoped entity context can improve selected outcome measures, but unrestricted persistence or irrelevant context can reduce quality, privacy or safety.
- H7 — Null/adverse outcomes: transaction overhead, weak demand, market power, enforcement cost, hardware shifts, governance overhead or poor context selection can eliminate or reverse predicted benefits.
Each hypothesis permits confirmation, partial support, null effect or contradiction. No result is converted into a StegVerse runtime disposition.
3. Study design
Use prospective, pre-registered matched comparisons. Before observing primary outcomes, freeze workload family and eligibility; baseline and treatment; outcome-equivalence threshold; observation window; sampling/recruitment and exclusions; primary and secondary endpoints; cost-allocation rules; statistical analysis; missing-data/failure treatment; and privacy, consent and evidence-retention plans.
Where random assignment is feasible and ethical, use randomized treatment assignment. Otherwise use matched cohorts or within-subject crossover designs and pre-register confounding controls. Do not describe an observational comparison as randomized or causal.
The primary denominator is a completed equivalent admissible outcome, not a token, prompt, model call or attempted transition. Equivalence criteria must be domain-specific and may include blinded human scoring, deterministic task accuracy, latency, reliability, safety and verification burden. All attempted outcomes remain in the cost ledger; rejected, failed and retried attempts cannot disappear merely because they did not enter the completed-outcome denominator.
4. Experimental arms
Applicable studies should define arms from: conventional baseline; governed execution; governed execution plus eligible evidence reuse; authorized entity-context treatment; context-poor control; genuinely available centralized/device/hybrid infrastructure; bounded lawful data-use/compensation cohort; and nonparticipant control where appropriate. No participant should be induced to surrender rights or data merely to populate an experimental arm.
5. Measures
Economic measures
Measure provider/model charges, subscriptions/direct fees, device depreciation, energy, network, storage/evidence custody, retry/failure cost, governance/admission/verification overhead, applicable dispute/settlement overhead, subsidies/cross-subsidies, and compensation actually paid and separately irrevocably settled.
Use Volume III's core definition: NUEB = C_baseline - C_governed + R_settled. Report avoided cost and settled compensation separately before their sum. Do not count transfers as aggregate resource savings.
Outcome measures
Pre-register task-specific quality, completion rate, latency, reliability, safety/admissibility, verification time, retry count/cost, duplicate execution avoided, evidence-reconstruction success and participant burden.
Data-use measures
Separate consent/authority, authorized purpose, usage event, attribution confidence, contractually owed, accrued, disputed, paid and settled amounts, revocation/expiration, and onward transfer. Payment does not prove authorized use; authorization does not prove payment.
Distributional measures
Where sample size permits, stratify by usage intensity, device age/capability, electricity price, network quality, accessibility requirements, jurisdiction, economic burden and data marketability. Report who pays upfront, bears risk and receives delayed value.
6. Counterfactuals and cost allocation
Every claimed benefit requires an explicit counterfactual. Prefer an observed contemporaneous baseline; publish assumptions and sensitivity ranges for modeled baselines. Allocate fixed infrastructure and device costs consistently and identify each cost bearer. Do not add mutually exclusive infrastructure scenarios into one user's cost stack. Prospective orbital scenarios require separate uncertainty and break-even analysis.
7. Confounders and sensitivity analysis
Pre-register model/version changes, provider prices, demand growth, learning effects, workload difficulty, hardware age, electricity rates, network conditions, subsidies, promotional pricing, model downgrades, participant selection, missing observations, jurisdiction, data demand and market power. Run sensitivity analyses for uncertain depreciation, fixed-cost allocation, energy attribution, baseline choice, compensation timing and outcome-equivalence thresholds. Report conclusions that change under plausible assumptions.
8. Negative and adversarial controls
Include controls for unauthorized/expired data use; duplicate/forged usage evidence; disputed/absent settlement; stale/out-of-scope evidence reuse; missing predecessor/custody chain; hidden device-upgrade cost; omitted provider subsidy; lower-quality model masquerading as savings; and rejected execution counted as successful completion. A control failure invalidates the affected research claim; it does not authorize inventing a runtime disposition.
9. Statistical analysis
Publish the analysis plan before primary outcome inspection. Report effect sizes with uncertainty intervals, denominators, exclusions and missingness—not p-values alone. Correct multiple primary comparisons when applicable. Use methods appropriate to the distribution and clustered structure. For causal claims identify assignment mechanism and estimand; label observational findings associative unless justified otherwise. Define a smallest effect of practical interest before testing.
10. Falsification and stopping criteria
A proposition is weakened or contradicted when its pre-registered primary measure fails the declared direction/effect threshold under an adequately powered applicable design, subject to uncertainty and replication. Pause for material consent/privacy failure, uncontrolled exposure, invalid assignment, corrupted measurement, inability to maintain outcome equivalence, or applicable safety/legal conditions. Null and adverse results must be retained and reported.
11. Evidence and admissibility
Preserve distinctions among research observation, source validation, simulation, authentic runtime transition, sovereign organization ledger readback, Master Records reconstruction where applicable, data-use authorization, contractual obligation, payment/settlement, and public publication/deployment evidence.
Repository CI, a processed manifest, source-only consumer or simulated receipt cannot substitute for an authentic governed transition. When a StegVerse transition is part of a study, preserve its authentic ALLOW, DENY or FAIL_CLOSED and applicable predecessor/custody evidence. If an authenticated interface cannot be reached, record UNKNOWN_NOT_AUTHENTICALLY_OBSERVED rather than manufacturing a disposition. Research datasets should use privacy-preserving identifiers and publish only evidence required for independent verification.
12. Ethics, consent and legal review
Human-participant work requires an appropriate ethics determination before recruitment. Obtain informed, purpose-specific consent and make withdrawal consequences explicit. Before live personal-data licensing or compensation research, obtain applicable jurisdiction-specific review of lawful basis, sensitive/minor data, onward transfer, retention, derivative use, revocation limits, consideration, tax/fee treatment, dispute rights and trained-model irreversibility. This proposal creates no financial instrument, token, payment processor, ownership right or legal conclusion.
13. Reproducibility and independent replication
For completed studies publish, as privacy and contracts permit: pre-registration/protocol version; analysis code; de-identified/aggregated data or controlled-access path; exact workload definitions; model/provider/version and hardware/environment metadata; cost-allocation assumptions; exclusions/missingness; aggregate result tables; sensitivity analyses; protocol deviations; negative/null results; and provenance hashes. Replication should reproduce analysis from retained evidence without author-only interpretation. Independent replication is preferred before broad economic generalization.
14. Proposed study phases
Phase 0 — Measurement validation. Validate instrumentation, outcome-equivalence scoring, cost allocation, evidence capture and privacy controls using synthetic/non-sensitive workloads.
Phase 1 — Matched execution economics. Compare baseline and governed execution. Primary endpoints: total user-borne cost per equivalent completed outcome and quality/reliability non-inferiority.
Phase 2 — Evidence reuse. Introduce only eligible retained evidence; measure duplicate execution avoided, verification cost, stale/out-of-scope rejection and quality effects.
Phase 3 — Context-conditioned utility. Compare scoped entity context with context-poor controls; measure task quality, burden, compute cost and negative effects from irrelevant persistence. This can test questions highlighted by the separate Convergence / Comparative Signal Treatment without treating external convergence as validation.
Phase 4 — Infrastructure incidence. Compare genuinely available centralized/device/hybrid implementations and model prospective alternatives separately; attribute costs to actual bearers.
Phase 5 — Authorized data-value pilot. Only after applicable ethics/legal prerequisites, test a bounded voluntary purpose-specific data-use arrangement with independently reconciled usage and settlement. Measure actual demand and settled value; zero demand or compensation is valid evidence.
Phase 6 — Replication and distribution. Repeat material findings across workloads, providers, cohorts and independent teams; examine distributional incidence before generalizing.
Phases need not all execute if earlier evidence makes later hypotheses uninteresting, infeasible or unethical.
15. Relationship to the paper collection
Volumes I–III remain the numbered Entity Economy series. Their publication does not depend on completion or favorable results from this proposal.
The separate Convergence / Comparative Signal Treatment records dated external comparison signals and research questions. It is not Volume IV, empirical proof, an origin adjudication or a substitute for the studies proposed here.
Empirical results should be separately identified research outputs, datasets, replication reports or later explicitly versioned editions where editorially appropriate. Negative findings should revise claims rather than be hidden.
16. Research-program success criterion
The program succeeds if it produces trustworthy evidence capable of changing Entity Economy claims in either direction. Findings that governance overhead outweighs savings, evidence reuse fails quality constraints, data demand is negligible, compensation is uneconomic, or context persistence harms outcomes are scientifically useful. The objective is not to prove the Entity Economy correct; it is to determine which propositions survive reproducible measurement, under what conditions, for whom and at what cost.