DOKU INFRASTRUCTURE
AI INFRASTRUCTURE TECHNICAL CREDIT METHOD APPENDIX

Method version: 0.4
Published: July 23, 2026
Prepared by: Jorg Doku

PURPOSE
Define a reviewable path from technical evidence to commercial effect,
financial exposure, lender action, and residual uncertainty.

INITIAL SCOPE
- North American AI infrastructure financings
- Neo-cloud and GPU finance
- Construction-stage data centers
- Power, cooling, network, compute, contracts, collateral, and surveillance

ROLE BOUNDARY
Doku organizes evidence, identifies contradictions, tests causal links, and
proposes credit actions. Doku does not replace licensed engineering, legal
counsel, utilities, appraisers, insurers, rating agencies, or lender approval.

CORE DATA ENTITIES
- Asset
- Capacity state
- Evidence
- Risk
- Action
- Covenant
- Owner
- Decision

CAPACITY STATES
Requested, contracted, energized, commissioned, deployable, utilized, and
billable capacity remain separate fields.

EVIDENCE HIERARCHY
1. Executed agreement or official record
2. Official letter, certification, or accepted test
3. Responsible expert model or report
4. Source-system extract or telemetry
5. Management assertion or working assumption

WORKFLOW
1. Define the decision, timing, threshold, and owner.
2. Inventory sources, authority, versions, provenance, and gaps.
3. Reconcile capacity, cost, schedule, performance, and acceptance.
4. Trace technical facts into commercial, cash-flow, collateral, and debt effects.
5. Propose conditions, reserves, covenants, reporting, and intervention.
6. Preserve review, override, change, closure, and residual uncertainty.

SOFTWARE ROLE
- Controlled manual intake first
- Structured upload, CSV, API, or source-system access when approved
- Field extraction and version comparison
- Contradiction and anomaly surfacing
- Evidence lineage and change preservation
- Accountable human review and approval
- PDF, structured register, CSV/API, dashboard, and surveillance outputs

STRESS LIBRARY
- Energization and completion delay
- Utilization shortfall
- Power-price or tariff variance
- Hardware-generation change
- Customer loss or concentration event
- Collateral removal and redeployment friction

VALIDATION PLAN
1. Timestamp evidence, assumptions, grades, reviewers, and predictions before outcomes.
2. Label completion, billable capacity, draws, covenants, restructuring, recovery, overrides, and false positives.
3. Publish sample size, inclusion rules, calibration error, reviewer agreement, and limitations before performance claims.

DATA AND MODEL CONTROLS
Client evidence is not used for general model training without written
authorization. Deployment, access, retention, AI use, export, deletion, and
incident handling are defined before restricted evidence is accepted.

VALIDATION STATUS
The method is not adopted by a rating agency, approved by a regulator, or
validated against a complete default dataset. No predictive-performance claim,
customer outcome, completed transaction, or external endorsement is made.