01
Accountability
Objective, owner, business process, legal entity, and financial-statement mapping
Use-case recordVersioned evidence
Research and methodology design proposal · Version research-2026-08-05 · Not production active
Intelligence Under Management research
Alpha is developing an independent opinion on whether AI investment is producing verified operating and financial results. Strong returns do not excuse weak governance.
The non-compensation rule
Capability does not prove adoption. Adoption does not prove value. Value does not excuse weak governance.
The Alpha AI Governance Rating remains distinct. Economic upside cannot erase a critical safety, rights, security, or control failure. Strong governance cannot convert unverified activity into proven value.
This proposal does not issue a rating, predict an actuarial loss, certify compliance, or attribute a stock-price movement to tokens or AI spend.
The rating family
One legible opinion. Four components that cannot hide one another.
The notation keeps governance, value, evidence, and direction visible. It rejects a blended score in which strong economics can compensate for weak control.
Component A: Alpha AI Governance Rating, AAA to D. Applicable missing evidence produces NR. Critical failures impose non-compensable constraints. Inspect the governance methodology.
V1Enterprise-compounding
Causally supported and financially reconciled value across material processes, with positive AI economic profit and demonstrated persistence.
E3+V2Repeatable value
Verified outcomes and positive fully loaded returns across multiple production workflows or a material segment.
CalibrateV3Operationally proven
Production quality or productivity gains are verified, while financial attribution, persistence, or scale remains incomplete.
CalibrateV4Emerging
Production adoption exists, but benefits remain localized, management-estimated, short-lived, or weakly baselined.
CalibrateV5Unproven or destructive
Evidence is primarily activity-based, or demonstrated full costs and losses exceed the verified benefit.
CalibrateE1Disclosed
Public information or unreconciled management claims.
DisclosureE2Documented
Issuer records, inventories, policies, samples, and management attestations.
EvidenceE3Verified
System-of-record extracts, telemetry, financial reconciliation, representative sampling, and selected independent validation.
EvidenceE4Continuously assured
Traceable production telemetry, controlled baselines, independent assurance, exception testing, and continuous evidence freshness.
EvidenceConstraint: a V1 opinion cannot rest on E1 or E2 evidence.
The evidentiary spine
Tokens become relevant only when the value chain remains intact.
Every transition is an evidentiary claim. The confidence of the enterprise value conclusion cannot exceed the weakest material link.
01Technical inputs
Compute, models, data, tokens, tools, agents, identities, and permissions.
01/0502Accepted work
Quality-adjusted output that meets the declared task and risk threshold.
02/0503Operating outcomes
Changed cycle time, capacity, quality, loss, service, or decision performance.
03/0504Financial reconciliation
Revenue, cost, margin, working capital, capex, expected loss, and cash effects.
04/0505Durable enterprise value
Free cash flow and economic profit tested for persistence, capital intensity, concentration, and risk.
05/05Raw token volume, seats, prompts, agents created, revenue per employee, and AI spend are diagnostic inputs. None is evidence of enterprise value without task quality, sustained adoption, a credible baseline, full costing, and financial reconciliation.
Analytical dimensions
Assess the complete production system.
These dimensions support the Value Realization Class. They are not presented as a false-precision weighted score.
D1Strategic allocation
Is intelligence deployed against material value pools and a declared risk appetite?
AssessD2Operational penetration
Has AI changed production workflows and decision rights, not merely activity volume?
AssessD3Human-agent operating system
Are accountability, skills, oversight, incentives, contestability, and job design coherent?
AssessD4Technical intelligence efficiency
Does the enterprise obtain accepted outcomes at competitive quality, cost, latency, and reliability?
AssessD5Financial value conversion
Do verified benefits exceed full costs, losses, and capital charges and reconcile to results?
AssessD6Durability and adaptation
Can value survive model change, vendor failure, attacks, regulation, and demand shifts?
AssessThe differentiating evidence asset
The Intelligence Value Ledger.
Each material use case receives one versioned record joining technical telemetry, accepted work, human oversight, financial outcomes, risks, controls, and evidence confidence.
02
System perimeter
Models, providers, data, tools, agents, autonomy, identities, and permissions
Use-case recordVersioned evidence
03
Baseline
Eligible population, pre-AI performance, workload unit, and accepted-output definition
Use-case recordVersioned evidence
04
Operating evidence
Quality, error, override, escalation, incident, and sustained-use measures
Use-case recordVersioned evidence
05
Human work
Review load, exception handling, training, redeployment, and remediation time
Use-case recordVersioned evidence
06
Technical cost
Tokens, compute, latency, energy where material, licenses, infrastructure, and routing
Use-case recordVersioned evidence
07
Value evidence
Revenue contribution, cost avoided, working capital, loss avoided, capacity, and option value
Use-case recordVersioned evidence
08
Attribution
Method, confidence interval, persistence, evidence freshness, and reconciliation
Use-case recordVersioned evidence
09
Control evidence
Risk tier, stakeholders, jurisdictions, tests, exceptions, remediation, and assurance
Use-case recordVersioned evidence
Aggregation proceeds from use case to process, segment, legal entity, and consolidated enterprise. Low-risk office copilots cannot conceal one material autonomous system.
Candidate indicators
Measure accepted outcomes, not AI theater.
The candidate indicators below require calibration. Each pairs a definition with a guardrail against obvious gaming.
01
Governed Intelligence Coverage
Material production AI activity within inventory, ownership, logging, evaluation, and policy controls divided by detected material production AI activity.
CoverageUse multiple denominators such as spend, workflows, actions, and affected decisions.
02
Workflow Penetration
Material eligible workflows with sustained production AI use divided by material eligible workflows.
AdoptionWeight by value or risk exposure, not workflow count alone.
03
Human Oversight Intensity
Human review, exception, and remediation hours per 1,000 risk-adjusted agent actions.
Human and agent workNever convert agent hours into a fictional digital full-time equivalent.
04
Accepted Outcome Yield
Accepted, quality-adjusted work units divided by fully loaded AI operating cost.
EfficiencyCompare only within declared task, quality, modality, and evaluation cohorts.
05
Risk-Adjusted AI Value
Attributed benefit adjusted for confidence and persistence, less full AI cost, realized losses, and expected residual loss.
Financial conversionUse mutually exclusive benefit categories and reconcile them to financial records.
06
Model Substitution Time
Time required to replace a critical model or provider and recover the required outcome threshold.
DurabilityTest the procedure under representative production conditions.
Activation boundary
Public research now. Effective opinions only after institutional proof.
Publication of this architecture does not activate a methodology. These gates must be completed and authorized before Alpha issues an effective Enterprise Intelligence Opinion.
- 01Ratify the construct, perimeter rules, indicator dictionary, value-class anchors, and evidence requirements.
- 02Reconcile the governance methodology version without allowing value to compensate for critical control failure.
- 03Run multi-sector shadow assessments without publishing effective opinions.
- 04Measure analyst agreement, disclosure bias, size bias, and issuer-cooperation bias.
- 05Backtest governance indicators against incidents and value indicators against persistent operating and financial outcomes.
- 06Complete independent accounting, actuarial, technical, legal, labor, and human-rights challenge.
- 07Operate committee approval, surveillance, correction, appeal, withdrawal, conflict, and change-control processes.
Research basis
Selected institutional sources.
The proposal draws on economic measurement, productivity research, management-system standards, risk governance, and the institutional mechanics of existing rating markets.
01
Federal Reserve: The AI Buildout and the Economy
Institutional source supporting the research architecture.
Primary source02
U.S. Census Bureau: Business Trends and Outlook Survey
Institutional source supporting the research architecture.
Primary source03
NBER: Generative AI at Work
Institutional source supporting the research architecture.
Primary source04
SEC: Credit Rating Agency Reforms
Institutional source supporting the research architecture.
Primary source05
NIST: AI Risk Management Framework
Institutional source supporting the research architecture.
Primary source06
ISO/IEC 42001:2023
Institutional source supporting the research architecture.
Primary sourceResearch and methodology design proposal. Reviewed . Correction path.
Methodology briefing
Help test the institutional standard for governed value.
Alpha is seeking board, investor, private-equity, insurance, accounting, actuarial, technical, labor, and human-rights challenge before any activation decision.