# Run 04 independent reconstruction model

`R4-RECONSTRUCTION-01` quantifies selected equipment stocks for `PLANT-GB300-1GW-2026Q4-A`. It holds the Run 03 population ledger at the base quantity state and the normalization at 1,000 MW commissioned IT. Fractional B1 racks and their equipment are conditional population equivalents. Electrical headroom and empty provisioned positions supply no additional populated compute assemblies. B1, B2 and B3 are disjoint increments. `B1-B3` adds the three once.

Each output row identifies the equipment and component, the population ledger, the estimate method, the source key, and the physical or nested mass view. Source entries link to retained captures and give a specific locator and a statement of transfer limitations. Registry `assemblies` holds the complete parameters for each row. `physical_kg_per_unit` covers the sum of modeled disjoint parts only. Gross residuals compare that physical sum to the named scope's gross where a comparable source limit exists. A residual leaves those kilograms unallocated. Some gross checks use different manufacturer, voltage class, duty, or option proxies. They test the arithmetic and constrain the working construction, while actual target masses and residuals remain open.

Low/base/high are three simultaneous engineering parameter states for the named construction. They are endpoint examples, with independent uncertain inputs intentionally moved together while keeping the same topology and population. They carry no probability, statistical coverage, or claim of physical outer bounds. The historical excerpt describes separately selected material-marginal witnesses. Corresponding endpoints therefore answer different questions. Replacement constructions change the specified parts in the full balance and keep installed populations fixed. Each alternative is separately compared to the base construction.

## Engineering assumptions and scope

- **Compute** uses the source CAD conductor/manifold shapes for two assemblies per rack and comparable frame/tray gross checks. Two 1,400 A busbar assemblies give 151.2 kW at the scenario's assumed 54 V nominal operation, before margin and transient qualification. The general copper stock density is 8,890 kg/m³. Near-pure conductor density and carbon, electrical and stainless steel density are engineering conversion assumptions. Sheet area is developed stock area, including equivalent structural channel faces. Coldplate substrate and stainless plumbing belong to the compute tray, with the supplier's module mass as a separate nested check. PSU thermal and conductor parts belong only to the separately counted PSU modules. Electronics, PCB stock, optics, data cables, nickel plating, rack coolant, semiconductor and packaging masses remain open. The first high frame geometry exceeded 180 kg by 2.512 kg. The 180 kg comparable assembly budget was retained, and the high sheet allowance was reduced from 9.3 m² at 2.5 mm to 9.0 m² at 2.45 mm, leaving about 6.9 kg for unmodeled objects in that comparison scope. Neither the area nor its revised headroom is a measured Supermicro dimension.
- **Transformers** use `4.44 f B A` sinusoidal EMF and phase voltage to select integer LV turns. Nominal primary turns then follow the voltage ratio, within 1.5%, and a 5% high-side turns allowance is applied separately to physical length. HV uses star-star and MV/LV and PDU use delta-star as construction choices. Bare area follows phase current divided by assumed current density. Both windings share their case's net iron section. The three-limb core centerline approximation uses iron density, net area, window height, yoke depth and limb pitch. The winding screen uses two adjacent winding packs in a shared window at at most 36% bare fill. It models each concentric coil at 60% bare packing over the assumed axial window height and requires 5 mm radial clearance to the equivalent circular net core, adjacent coil and half-limb pitch. Increasing the IT and facility low-side mean-turn length allowances to 1.80 m and 1.62 m at base leaves a finite modeled core gap at low, including for the aluminum alternative. These are simplified geometric clearances, with actual insulation and dielectric clearance unresolved. Integer ratios and mean-turn/perimeter checks rejected initial guessed PDU geometry. Wider pitched limbs now accommodate adjacent coils in every power class. The revised PDU uses a 0.078 m² base net limb section and a 0.57 m pitch. It clears the 4,550 kg proxy assembly gross with roughly 1,108 kg remaining at base and 569 kg at high for unmodeled hardware. Cooling performance, leakage, winding build, tap-switch geometry, detailed exterior dimensions, exact voltages under load and losses remain open. Oil assumes independent free volume, excluding solid cores/windings and tank shell. A fixed-resistance aluminum MV/LV replacement uses 1.65× the bare conductor area at fixed turns and geometry, then rechecks the window.
- **Distribution** uses 3-phase route counts with independent length allowances assigned by endpoint type. The Trinergy manual provides wire sections/parallel counts for rectifier input, bypass and output. The scenario takes one common input branch to the split and a local bypass branch after it. Other lengths and conductor sections are independent construction assumptions, with typical equipment-room routes in tens of metres and a separate MV campus allowance of 90/180/340 m. The scenario includes two massive transformer-to-main-switchboard secondary paths by equipment class, intermediate UPS and PDU feeders, an ATS-to-chiller branch, generator takeoff and MV campus feeders. Lower-rated branch screens use their stated duty class with reserve in assumed amperage. Approximate current density and single-layer phase-cable packing screen lengths/sections for this construction. Protective earth and steel ladder stock are separate components. Exact routes, lugs, sheath, neutral/earthing details, sharing topology, thermal grouping, protection, fault levels and code/ampacity qualification remain open. The aluminum phase alternative uses 1.65× the bare cross-section at fixed paths and copper earth. The aluminum 800 A busway uses the source-specific alternate section and gross, replacing the four copper rails.
- **Batteries** use 486 physical cells at 20 Ah per Li5 proxy cabinet. The declared 5 Ah IFR32650 recipe supplies normalized material fractions for a hypothetical 20 Ah cell. Active LFP follows capacity divided by specific capacity. Cases use the reversible theoretical limit, 160 Ah/kg and a 7.29% reserve on the 160 Ah/kg case. Per-part metal and chemistry scale by the LFP recipe fraction. The nested elemental Li view uses 6.94 g/mol divided by each host's conventional molar mass. It is counted once by host identity in the element view and never added to physical gross. LFP iron is outside the ferrous stock. The SDS iron recipe category has no qualified metallurgy and remains other ferrous. Cells plus metallic module trays/interconnects must fit the independently derived module-population gross. Cabinet metal fits the separate intercept envelope, and both share the exact 590 kg selected cabinet cap. Exact selected chemistry and hardware composition remain open.
- **Cooling** applies manufacturer assembly/material identity, related dimensional evidence where available, and explicitly hypothetical detail geometry. The chiller distinguishes direct copper-tube free cooling, aluminum condenser channels/fins, copper/steel evaporator, compressor castings and motor laminations. The 13,110 kg legacy dry proxy screens metal quantities. Exact source charge is one 89 + 80 kg R513A booking per chiller, outside the dry proxy. CRAH's dry fin area follows depth divided by pitch times face area, two sides, and a net area factor. Internal wet area is bore circumference times developed length. Air and water conductances are put in series using assumed 100 W/m²/K air coefficient, 80% fin efficiency and 5,000 W/m²/K water coefficient. These yield 37.55/45.75/57.71 kW/K against the retained design-day 33.66 kW/K approximate requirement. This is a screening construction with uncertain air and EG-side film coefficients, omitted wall/fouling/crossflow penalties and unqualified pressure drop, airflow and tube circuitry. The preliminary CDU uses a wet gross proxy, with its residual including unidentified fluid. Pump estimates are component geometry, with no bulk kg/kW conversion. Straight pipes preserve the selected diameters and aggregate lengths including return, with no extra supply/return multiplier. The retained pipe catalog walls also supply a heavier replacement. Separate initial water/EG stocks fill only the selected external pipes and buffer vessels. A closed 30 wt% EG mixture at 1,030/1,040/1,050 kg/m³ is an input assumption. No annual withdrawal/evaporation or internal liquid masses follow from this inventory.
- **Generator** uses explicit volumes for cast engine/alternator pieces, steel shaft and laminations, alternator copper and aluminum radiator, and equivalent sheet stock for the skid. Cylinder head casting allowance is screened against the matched-family complete 20-head proxy gross. The selected engine/alternator uses a separate 25 t dry comparison limit with substantial unallocated mass. Exact engine, alternator, frame and acoustic enclosure configurations may vary. The external fuel train, operating fluids and fuel are open.

## Relationship to the recovered report

The battery module and cabinet gross-budget endpoints are marginal constraint envelopes. Their separate low or high budgets are not jointly additive partitions of the 590 kg whole cabinet; only the actual modeled disjoint components are summed and tested against the whole-cabinet cap. Do not add the reported budget residuals to stock or to one another.

The recovered report supplies comparison points and descriptions. Its equipment coefficient package, route lengths and detailed design witnesses were unavailable and were not used to constrain the construction above. The central estimate is lower in Cu, Al and total ferrous, while electrical steel is higher. Direct comparison follows the generated reconciliation tables. Plausible causes are shorter or different effective external paths in individual B2 feeder branches, the distinct coil and sheet-frame dimensions, the concentration of more material in modeled transformer cores, a deliberately incomplete generator decomposition, and missing installation and operating fluid stock. These are differences in model construction and coverage. The excerpt alone cannot establish the original bill of materials, reconcile all scope differences or verify either construction.
