Quantitative Valuation, TCO & Collateral Methodology (M-2026.10-r2)

1. Strict 5-Band Valuation Taxonomy & SpotWire Component Boundary

M-2026.10-r2 enforces strict evidentiary boundaries across every hardware profile:

• New OEM / Channel Reference (0m): Published integrator or AI Hardware Index asking median for 0-month server BOM.
• Secondary Refurbished Asking (T3): Unexecuted secondary dealer offer band (explicitly non-transactional).
• Executed / Comp Secondary FMV (T1): Requires n ≥ 5 invoice-verified 8-GPU/4-GPU server-cluster transactions executed within the last 60 days. Single-card component eBay datasets (e.g., SpotWire gpu_ebay, separated by data_backed / thin / estimated coverage and quarantined when >60d old) are exposed solely as diagnostic component cross-checks and never promoted to server-cluster FMV.
• Credit Borrowing-Base Halt vs. Hypothetical T4 Model: Currently 0 of 8 hardware profiles in SQLite hold ≥5 fresh (≤60d) executed server-cluster comps; consequently, Credit-Eligible Borrowing Base OLV and Senior Loan Principal are strictly €0 / INELIGIBLE across all profiles until verified ITAD comps are ingested. Parametric T4 decay figures are isolated in a separate non-credit simulation panel.
• LIVE MARKET DATA vs. STATIC REFERENCE ASSUMPTIONS: Every response exposes data_provenance_split separating live ingested feeds (ECB FX, Eurostat X_VAT power, Fraunhofer spot, Azure/AWS retail rate cards) from static parameters in methodology_v1.json (including neocloud_ref_secure_usd_gpu_h, tagged T4_STATIC_MODEL_ASSUMPTION—never a live GPU rental index).

2. Data Center Load Physics & Eurostat tax=X_VAT Industrial Power

Enterprise AI clusters draw significant baseline power at idle and pay regulated industrial grid tariffs rather than raw wholesale spot:

P_facility(kW) = N_nodes × P_IT_max × (0.35 + 0.65 × Utilization) × PUE

• 35% Base Idle Draw: Accounts for dual-socket host CPUs, 2 TB DDR5 RAM, NVSwitch fabric, BMC/cooling, and PSU losses prior to dynamic tensor load.
• Eurostat nrg_pc_205 with tax=X_VAT: European siting calculations ingest Band IF (70,000–149,999 MWh/yr, ~8–17 MW facility) using official Eurostat tax code X_VAT (Excluding VAT and other recoverable taxes and levies, while retaining non-recoverable industrial network charges and levies). Earlier I_TAX values (which include recoverable VAT) are purged.
• Wholesale Spot Volatility Spread: Fraunhofer ISE Energy-Charts day-ahead spot is displayed strictly to quantify grid/levy spreads—never substituted as the full colocation bill.

3. Contract Existence vs. Verified Contractual Cashflows (DSCR Gate)

Credit committee metrics prevent unverified form inputs from masquerading as investment-grade contracted cashflows:

• 4-State Contract & Cashflow Separation: Unverified form inputs are labeled USER_DECLARED_CONTRACT_SCENARIO. Even when a verified_contracts row confirms contract existence (AUDITOR_VERIFIED_CONTRACT_EXISTENCE), if no audited contractual contracted_rate_usd_gpu_h is stored, DSCR is classified as SIMULATED_DSCR_UNVERIFIED_CONTRACT_RATE and blocked from credit-eligible DSCR. Only contracts with both verified existence AND an audited $/GPU-h rate achieve AUDITOR_VERIFIED_CONTRACT_AND_RATE (using the exact contractual rate rather than a static -15% assumption).
• 12–60 Month Mathematical Parity: Python JSON responses, Monthly_Schedule live-formula Excel workbooks (dynamically generating 12 to 60 monthly rows), and Executive PDF briefs share identical full-precision unrounded base inputs, guaranteeing <0.01 rounding parity across all horizons and currencies.

4. Accounting, Appraisal & GeForce EULA Exclusions

• Not USPAP / ASA / IFRS 13 Certification: Outputs do not replace physical serial-number inspection, burn-in telemetry verification, or a signed appraisal by an Accredited Senior Appraiser (ASA).
• GeForce EULA Borrowing-Base Block: Consumer GeForce SKUs (RTX 5090) are hard-coded as INELIGIBLE_EULA_RESTRICTED (€0 eligible loan and €0 commercial data-center debt simulation) due to NVIDIA GeForce Software EULA restrictions on commercial data center deployment.