What AI Is Made Of

Run 03 cooling material forensics

The fixed cooling architecture is preserved. Counts are read from calculate_balance(..., configuration="CFG-GB300-2026Q4", snapshot="SNAP-RUN02"); snapshot_quantity_trace.json retains the resolved rows and their upstream derivations. No canonical BOM, existing source capture, or Run 02 record was edited.

ClassExact selected count / allowanceMaterial resultPriority inference
FH3145 chiller1,070R513A 180.830 t inherited; bulk metals remain OPENHigh Cu/Al/ferrous: 18-fan, roughly 13 t product; separate condenser/freecool/evaporator metals
Preliminary CDU2300856Conditional wet gross envelope 1,534.808 t; actual metal bounds unknownMedium Cu/Al, medium/high stainless: plate HX, piping, redundant pumps
Generic warm-water CRAH2,140Selected gross and metal bounds unknownHigh/medium Cu/Al: many fin-tube coils; frame steel; exact coil duty unqualified
Generic 45 kW primary pump428Selected gross and metal bounds unknownLower expected Cu/Al than chiller fleet, medium ferrous; this is a research inference, not a ceiling
External primary pipes214 × 200 m = 42,800 mValidated shell model; unselected construction alternatives onlyHigh ferrous potential if steel; cannot select material from a hydraulic DN
External secondary pipes856 × 30 m = 25,680 mValidated shell model; unselected construction alternatives onlyMedium/high stainless potential if stainless; material remains unselected
Closed buffer tanks214 × 50 m³Existing WTS alternatives retained; no actual tank metallurgical selectionMedium/high ferrous; volume alone does not identify vessel pressure, wall or alloy

Priority is an inference from equipment function, counts and gross-mass evidence quality. It is not a coefficient and does not certify that pumps or CDUs are immaterial. Especially, a 1,535 t conditional CDU envelope can remain material to an otherwise small copper subtotal.

evidence.json contains engineering inputs, equations, six validation objects, rejected reconstruction models, explicit conditional bounds, closure blockers and the inherited COMMON refrigerant bridge. proposed_forensics.json adapts the COMMON R513A bridge and six pipe scenario rows to the Run03 engine schema; all seven rows validate with the shared engine. sources.json and claims.json retain source identities and locators. capture_hashes.json hashes source PDFs, extracted text and OCR text. The source PDF, rather than OCR, is authoritative where glyphs differ. web_search_captures.json and SEARCH_LOG.md preserve search outcomes, including failures.

Chiller physical evidence

The newly captured OEM North American AFC guide specification (SL-71334GS Rev A 08-24, pp2–4) identifies the material-bearing assemblies independently of an EPD:

These are material identities, not masses. They avoid an incorrect blanket Cu-tube condenser assumption and an incorrect residual-to-steel allocation. The2024 NA guide §2.1 associates F direct free cooling with fin-and-tube and N indirect free cooling with microchannel. Generic2026 family schematics depict a microchannel freecooling assembly; applicability to the fixed FH3145 direct topology has not been verified. This is a family/vintage qualification question requiring the full selected suffix and drawing crosswalk, not evidence of a qualified microchannel substitution in the fixed FH design. No alternative that changes the frozen direct topology is selected or quantified.

The 2021 EU manufacturer product-design manual is also captured from a public distributor mirror. Its FH3 165HG0 column has an 18-fan, 12.106 m frame, 45 m² freecooling coil face area, 3 tube rows and 1.8 mm fin spacing. This is close physical family evidence for leads, but differs from the 460 V / 60 Hz FH3145 selection. The EU evaporator describes plastic baffles and cast iron heads, whereas the NA guide specifies carbon steel; retain that conflict. European table anomalies in fluid volumes and operating mass prevent treating a net-to-operating difference as a measured coil quantity.

The attempted reconstruction is parameterized in R3-MOD-C-CHILLER-COILS:

m_Cu = rho_Cu × annular tube section × developed tube length + headers + return bends

m_Al_fins = rho_Al × fin thickness × fin count × net fin area

m_Al_microchannel = rho_Al × [extrusion solid section × extrusion length + header solid volume + fins]

m_frame = rho_steel × [developed beam-profile widths × thickness × lengths + panels]

Tube OD, wall and pitches, selected coil dimensions and circuiting, fin thickness, microchannel solid sections, beam profile widths, and exact compressor identities remain absent. Photographs and the outer frame envelope do not supply these. No calibration object with known coil composition and complete geometry was located. The model is consequently REJECTED from both strong and expanded forensic copper; all proposed Cu/Al/steel points stay null. This is an unsuccessful quantified transfer with useful physical diagnosis, rather than another exact-product declaration sweep.

The exact 169 kg R513A charge is a COMMON bridge of the existing Run 02 application: 89 + 80 kg per selected unit, totaling 180,830 kg. It is counted once across inherited/current records. Chiller qualified gross mass remains unavailable, so no residual is computed by subtracting refrigerant from an incomparable gross definition.

CDUs, CRAHs and pumps separately

CDU2300's preliminary wet 1,793 kg nominal mass produces a physically valid *conditional* each-material ceiling, given that envelope, of 1,534,808 kg for 856 units. The dry mass and fluid volume are unknown. The manufacturer expressly says the product is not finalized, validated or production-approved. This is not an actual selected-plant outer bound. Carrier65LL1.3MW and XDU1350 are useful leads for plate heat exchangers, redundant pumps and stainless piping, but do not identify2300 plate count, braze alloy or pump masses; no capacity scaling is performed.

For CRAHs, new OEM CW guide/catalog captures document welded tubular steel frames, copper-tube/aluminum-fin coils, stainless condensate pans, fans, filters and controls. CW181's cold-water geometry and mass do not establish the generic selected200kW minimum *net sensible* warm-water duty. Neither its mass nor its coil can be multiplied by the target CRAH count. Thus even a gross-only upper bound remains unavailable.

For primary pumps, a manufacturer Product Centre print of a Grundfos NB150-315/29145kW is indexed, with552kg nominal net mass. Its50Hz380–415V motor/duty is not the selected450m³/h at250kPa480V-fed pump. The mirror PDF destination was denied by network policy; the search capture is retained as a lead. No unverifiable mass or copper percentage is admitted. Dominant pump assemblies are housing/impeller (cast iron if that variant), motor electrical steel and conductor, shaft, base/VFD and seals. Motor power is not a copper-mass coefficient.

Validated pipe-shell construction scenarios

The annular-cylinder model uses m = rho × pi/4 × (OD² − ID²) × L, with millimetres converted to metres. Six declared nominal physical pipe objects from SHIELD carbon-steel and Haihao stainless-steel engineering tables validate the mass arithmetic. Maximum absolute error is1.14035%, below the predeclared1.5% acceptance threshold. This validates straight-shell arithmetic, not metallurgy selection or pressure suitability. Density7850kg/m³ carbon steel and8000kg/m³ stainless steel are explicit engineering assumptions and are tested against the table masses.

Application keeps the existing approximate450mm/200mm hydraulic bore allowances and exactly42,800/25,680m selected lengths. It does not silently replace them with standard-pipe manufactured IDs. Standard-product wall values are borrowed as construction alternatives; the implied ODs are derived and the result is not represented as an exact standard SKU. The wall choices have no assertion of pressure/corrosion/structural qualification. No architecture or routing changes are made.

Scenario-only wall alternativesPrimary carbon steel tSecondary stainless steel tCombined ferrous pipe shell t
PIPE-THIN-WALL:6.35 /3.76mm3,058.687494.4723,553.159
PIPE-INTERMEDIATE-WALL:9.52 /6.35mm4,617.476845.6945,463.170
PIPE-THICK-WALL:14.27 /8.18mm6,992.9091,099.0758,091.984

Thin/intermediate/thick name construction alternatives; they are not a confidence interval, certified physical lower/upper bounds, or an admitted base case. Actual material/wall uncertainty is still null. The full line count already includes both supply and return; it must not be doubled again. The model covers external straight pipe shells only and excludes fluid, insulation, joints, valves, bends, supports, equipment internal pipework and B1 rack manifolds. Those missing objects are unresolved residual scope, not assigned to steel.

Existing WTS50m³ stainless tank alternatives of1,836.12/1,889.62t remain isolated from the plant balance. A selected pressure, geometry and alloy are still needed. Tank fluid is separate from vessel metal; the80% usable thermal fraction does not reduce the full fluid volume.

Bounds, conservation and closure

Conditional each-material maxima from the nominal CDU wet envelope (1,534.808t) and legacy chiller dry proxy envelope (14,027.700t) are retained transparently, with actual plant upper/base fields null. The joint sum must be no greater than the common gross envelope. Separate Cu/Al/steel maxima cannot be added to create a material decomposition. No unknown chiller residual is allocated to steel. Batteries and semiconductors are outside this workstream.

Cooling Cu, Al, carbon steel, stainless steel and cast iron remain OPEN. The necessary high-contribution assemblies are identified, but chiller coefficients and selected CRAH/pump valid ceilings remain absent. These unresolved contributions cannot yet be shown immaterial. Pipe/tank scenarios do not close actual stocks. No new elemental metal coefficient is admitted merely because dimensions, chemistry names or arithmetic are precise.

Run python research/run03/cooling/build_handoff.py offline to reproduce the evidence calculations. It resolves quantities from SNAP-RUN02, validates six pipe objects, regenerates sources/claims/evidence/validation and hashes local captures. All outputs stay within this workstream directory. The exact inherited refrigerant result is180.830t; new primary metal admission count is zero.

Source details

Archived Run 01-03 research. These statements retain their time-of-writing context; current material estimates use Run 04. Web prose uses current site terminology; downloadable records preserve original wording. Hashes identify the downloadable records.

Source: research/run03/cooling/HANDOFF.md
SHA-256: c6b9f8457b4a90e269fc12138e7caba86ef91962c262d355aae7efff9d0f7eca
Download research file

Optional: tell Sid you’re reading
The document is open; sharing your name/email is optional.