# Run 03 transformer material forensics

No numerical transformer material coefficient is admitted to the fixed plant's primary balance. The investigation produces reproducible physical models, explicit validation errors, conductor alternatives, a closer PDU gross-mass sensitivity and specific unresolved inputs. UNKNOWN remains distinct from zero.

Canonical SNAP-RUN02 calculation supplies 24 campus 84 MVA 230/34.5 kV transformers in B3; 428 IT 5 MVA and 214 facility 3.5 MVA 34.5/0.48 kV transformers in B2; and 1,712 isolated-transformer 1 MVA 480-to-415 V PDUs in B2. The latter are an additional transformer-bearing class. No counts, topology or equipment selections were changed.

All four classes merit high Cu/Al/electrical-steel research priority. The large population of MV/LV and PDU transformers makes even moderate per-unit winding/core mass potentially material. This is a priority inference, not a coefficient. An exact transformer catalogue model, cooling stage and conductor are still absent. Rating alone cannot identify physical mass.

## Physical reconstruction and validation

The winding model uses rho × bare conductor area × physical turns × mean turn length × phases. A parameter interface relates MVA and voltage/connection to phase current, current density to bare area, and frequency/flux/net core section to nominal turns. It requires physical tapped turns and winding geometry as additional inputs; cooling and impedance do not map uniquely to mass.

Three independent design-team cases from the CLASP 2011 dry-transformer report provide geometry and engineering BOMs. Across six HV/LV windings, reconstructed copper differs from published design mass by at most **0.209%**. Physical maximum tap turns are necessary: using nominal turns omits the extra copper. Bare wire/foil area excludes insulation. These are **ENGINEERED design checks**, not measured teardowns or declared selected equipment.

A symmetric three-limb core approximation uses rho × net iron section × [3 × (clear-window height + yoke depth) + 4 × limb pitch]. Diagram dimensions and equal-section assumptions are explicit. Three core checks differ by at most **1.222%**. Joints, nonuniform yokes and detailed core construction remain unresolved. Net area already excludes voids, so a second stacking factor would double-remove them.

Good agreement only establishes conditional plausibility with complete geometry. It does not validate transfer to the selected 84/5/3.5/1 MVA classes. **Zero measured physical-object validation cases were retrieved.** Selected conductor, cooling/type, winding connection, flux, bare area, turns, geometry, loss specification and gross mass remain absent. Target low/base/high therefore remain null.

The tempting shortcut m = density × load loss / [electrical resistivity(T) × mean current density squared] was tested against all 18 CLASP 2013 large-transformer designs. Its mean absolute error is **21.348%**, with errors from **−18.001% to +70.746%**. Treating aggregate load loss as DC winding loss and reported mean current density as effective per-winding density fails. This shortcut is **rejected**, including for the base case. No kg/MVA regression or linear scaling was used.

## New physical evidence and source-only scenarios

The Vertiv PPC brochure and installation manual provide a close PDU family: 480 V, 60 Hz input, 415 V output, isolation transformer and Cu/Al options. However 1,000 kVA is engineer-to-order; the weight table stops at 750 kVA. Scaling that weight would invent a relationship.

Schneider's Galaxy PMM1000-CUB has the target 1,000 kVA rating, 480-to-415 V function and a copper transformer. Whole-PDU installed gross is **4,550 kg**, with **4,700 kg shipping mass excluded**. Applying canonical count yields a **7,789.6 t source-product gross sensitivity**. Manufacturer, selected conductor, options, dimensions and selected gross are unqualified. This is a separate **QUALIFIED PROXY scenario**, not a Cu/Al/steel coefficient or a finite physical bound on selected equipment. The PDU transformer is already nested inside this gross.

A Larson manufacturer catalogue offers a near-voltage-matched 5 MVA, 34.5/0.48 kV, 60 Hz, Al-wound KNAN product. It supplies a useful aluminum-conductor lead, but fluid volume conflicts (700 versus 190 gallons), weights leave a 666 lb discrepancy, and listed losses conflict with listed efficiency. Material transfer and gross bounds are rejected.

Public NYISO procurement/interconnection evidence describes an 84/112/135 MVA two-winding 230/34.5 kV transformer with ONAN/ONAF/ONAF ratings, alongside a separate three-winding product. It reinforces the need to identify cooling stage and winding topology before matching an “84 MVA” object. Neither product is selected for this plant.

## Conservation, bounds and terminology

The CLASP 2011 priced BOMs **including cabinet** exceed the reported transformer weight by 245, 79 and 85 kg. Cabinet exclusion from bare-transformer weight may explain the scope mismatch. The combined BOM is inadmissible against that gross until clarified; no source typo is asserted and no negative residual is assigned to steel. Core-plus-winding subtotals independently remain below gross and leave unresolved residuals.

Core electrical steel, structural steel, cast iron, generic iron, copper conductors, aluminum conductors, oil and insulation remain separate. Neither a whole transformer nor a residual is relabeled steel. Material-specific finite upper bounds on selected transformers remain unavailable because exact gross mass is unknown. Source-product gross masses cannot bound an unselected object. Nonnegativity supplies only a trivial zero lower bound, with high null.

Cu/Al alternatives preserve unresolved selected conductors without changing equipment topology. Their material masses are null. B2 and B3 transformer material closure remains **OPEN**. Isolation-transformer composition must replace the corresponding PDU parent residual when it becomes available; it must never be added to a complete parent mass.

## Files and checks

- evidence.json: target lineage, model admission, scenarios, null material ranges, bounds and closure.
- physical_model.py: parameterized winding/core models and rejected loss shortcut.
- validation_inputs.json and validation_results.json: every numerical input, prediction, error and source conflict.
- sources.json and claims.json: canonical-style source/claim entries with locators, dates, evidence levels and hashes.
- captures/, capture_manifest.json and fetch_manifest.json: original PDFs/HTML where accessible; explicit hashed rendered captures where original fetching returned 403.
- search_log.json: exact retained search queries and disclosed gaps in earlier query-text retention.
- build_handoff.py and build_registers.py: reproduce research outputs without canonical edits.
- test_physical_model.py: six physical/scope checks.

Run:

    python research/run03/transformers/build_handoff.py
    python research/run03/transformers/build_registers.py
    python -m unittest discover -s research/run03/transformers -p 'test_*.py' -v

Run02 source conflicts are preserved: CLASP 63 MVA case4 detailed core428,500 kg versus summary42,850 kg, with cost arithmetic supporting42,850; 25 MVA case2 load loss98 versus98.5 kW. SNAP-RUN02 remains unchanged.
