COLD LADDER LOOP ARCHITECTURE STUDY, REV A (new document; documentation and budget study only; no schematic, PCB, component or ICD changes)

Create a NEW project document titled "Loop Architecture Study". Do not edit the Card-Cage Layout Study; reference it where needed. ICD Rev A remains governing. Classify EVERY number: I = ICD, F = Founder ruling, S = study assumption, L = literature, V = supplier-stated, D = derived. Do not present bulk or room-temperature values as cold values. Run all arithmetic in the calculation tool and read back the finished document.

GOVERNING FOUNDER RULINGS (verbatim, Oct 15-17, 2026)
R-CL7: "The Cold Ladder is one platform across all bands. Footprint, card cage, slot pitch, backplane, enclosure shell, manifold and column positions are common and sized to the C Band envelope. A band change swaps only the boards, the cold-plate working fluid and its ratings, the insulation and atmosphere package, and the cold-source plant. Compressors and cold boxes sit at grade beside the cluster. Heat-rejection hardware, and in the C Band an optional gravity-feed head vessel, sit in the crown on the column's own frame. Nothing hangs from the existing building."
R-CL8: "Coolant circulates continuously through a flat-faced honeycomb tube, hexagonal cells running its length, that serves as the canal on each cold plate. The flat face clamps over each CPU and GPU die through a thermal interface; lower-power parts couple to the tube by conduction. One tube pass serves a Slot Class A slot; higher-power slots get one tube per die. The wetted path is one metal (copper preferred). In the G and W Bands the fluid stays sealed and never touches the silicon; direct die contact remains a C Band option only, for cold-native silicon. Tube ends carry dry-break disconnects for service."
R-CL15: "Cold Ladder operating-band benefits are presented as projections until prototype and beta testing measure them. Each projection states its basis (measured, supplier-specified, modeled or literature) and the test that will confirm or retire it. The operating band for a deployment is selected by measured useful compute per watt, reliability and total facility energy across a band sweep, not by an assumption that colder is better. TB-001 results are presented as evidence of thermal-governance capability, not as Cold Ladder performance."
R-CL16: "Each rung is a governed, sovereign thermal zone. It can be isolated from the spine without disturbing any other rung. Its failure-prone parts (accumulator, regulators and valves, sensors and rung controller) are line-replaceable cartridges with self-sealing connections, keyed mounts and common part numbers, replaceable without opening the CO2 loop beyond that cartridge and without tools where practical. Sensor arrays and Victor locate a fault to the specific cartridge, and a failed sensor or controller defaults the rung to a safe setpoint. Time to restore and zero neighbor impact are measured in prototype and beta."
Also in force: R-CL8A/8B/8C zoned ceramic plates (see Card-Cage study addenda); every enclosure holds dry nitrogen at slight positive pressure with dew-point monitoring. Working fluid is NOT yet ruled; CO2 is the leading candidate under study.

INHERITED INPUTS (classify as given)
- I: Slot Class A 1.5 kW nominal / 1.8 kW peak (one CPU + one OAM); CPU 358 W; OAM 750 W / up to 900 W; ICD Article 5 +5 C floor / +7 C release, -20 C routine, -40 C qualification mode.
- S: geometry from the Card-Cage study (case 6, 98 mm pitch, rungs and slots per unit, 0.144 m2 unit footprint, about 2.0 m height).
- L/V (bench-verified Oct 17): DOWFROST 50 percent PG by volume viscosity 3.11 cP at 40 C, 10.65 at 10 C, 73.02 at -20 C, 172.83 at -30 C; freeze points PG 50/55/60 wt% = -33.5/-41.6/-51.1 C. Lenovo GB300 NVL72 rack (two independent reports, table not yet directly read): 59 L/min at 2.3 psi (25 C) to 177 L/min at 18.4 psi (45 C); GB300 135 kW TDP, 155 kW peak, about 90 percent to liquid; QD static pressure 30 psi. Vertiv XDU1350: 1,200 L/min at 35.4 psi external, 13.7 kW max. CERN pumped two-phase CO2: detector heat exchangers to -43 C, plant to -53 C, 300-500 kW (IIR 2021).
- D (bench, to be re-derived): lift penalty = 1/COP, COP = eta x T_evap/(T_cond - T_evap), eta 0.5-0.6; warm-water baseline overhead about 3 percent. Bench results: cool band +10 to +15 C boiling about 12-25 percent; G Band -20 C about 37-52 percent; CO2 about 4-5 points lower than glycol at equal plate temperature.
- Pull CO2 saturation pressure, liquid viscosity, latent heat and density from NIST (REFPROP or WebBook) at +25, +15, +5, 0, -20, -40 C, plus triple and critical points. Label L.

TASKS
A. FLUID TRADE BY BAND. Bands: cool/W (+5 to +25 C setpoint), G (-20 C routine, -40 C qualification). Fluids: (1) water/PG25 (cool band only), (2) PG and EG 50-60 percent, (3) pumped two-phase CO2, (4) one low-viscosity single-phase synthetic (silicone or HFE class; note PFAS status for fluorinated options). For each fluid and band: properties, flow needed for one Slot Class A slot and one full unit, pressure drop through one R-CL8 honeycomb tube pass (state assumed cell size and length as S), pump power, static and standstill pressure, freeze and phase limits, leak consequence (conductivity, residue, asphyxiation), PFAS/regulatory status.
B. TOPOLOGY OPTIONS. (1) Per-unit pump on a shared trunk with decoupler. (2) Per-unit heat exchanger plus pump (split fluid: unit fluid differs from trunk fluid). (3) Per-rung zones on a spine liquid bus and vapor-return bus: per-rung metering valve, per-rung back-pressure regulator (sets that rung's boiling temperature), per-rung mini-accumulator. State the constraint that no rung can run colder than the vapor-return bus. (4) Condenser at unit base versus crown condenser with gravity return (thermosiphon). Compare parts count, charge, pressure zones, control authority, fault domain.
C. PRESSURE AND STANDSTILL. Required design pressures per option and band, including CO2 standstill when a unit warms to room temperature (supercritical above 31 C). Size accumulator and recovery or standby-cooling options; no venting into occupied space. Effect on plate and tube wall thickness, couplings and the existing 6 mm plate bound; flag any conflict with the Card-Cage stack budget.
D. CONTROL ARCHITECTURE. Sensors per rung (die/HBM temperature feeds, pressure, flow, vapor quality or superheat, leak/mass balance); layers: fast (boiling buffer, pump speed, metering), slow (accumulator/back-pressure setpoint), supervisory (band by workload, weather, price; learned feed-forward by Victor). Guard rails: dew-point margin, component minimums and Article 5 admission, pressure limits, rate and count limits on temperature swings (coating and tile cycle budgets). Fail-safe default setpoint on sensor, controller or heartbeat loss.
E. SERVICEABILITY PER R-CL16. Line-replaceable cartridge list per rung; isolation valves both sides; dry-break couplings for liquid and vapor at the rung drawer; sensors outside the pressure boundary where possible; trapped-CO2 recovery per swap; method to estimate time to restore; neighbor-impact check.
F. ENERGY AND HYBRID MODES. Lift penalty per band and fluid at condensing 30 C and 40 C; pump energy; three modes: setpoint switching (CO2 warm up to about +25 C boiling vs cool), condenser economizer staging (full free, partial, full mechanical), cold storage (night charge, day draw). 24-hour profiles for an Atlanta site: inference-like (daily peak) and training-like (flat). Report compressor energy with and without each mode. Use typical-year Atlanta weather if available (state source), else a labeled S profile.
G. CARD-CAGE INTEGRATION. Space per rung for accumulator, regulator and valves; coupling envelope at the drawer; added mass; wetted-zone coating needs under CO2 versus glycol (dry CO2 is weakly corrosive; water ingress forms carbonic acid); any change to the R-CL8A wetted-zone requirement is a Founder question, not a study decision.
H. PRIOR ART FLAGS. List known practice each option relies on (supermarket CO2 evaporator pressure regulators, CERN 2PACL accumulators, pumped-refrigerant economizers, in-rack CDUs, model predictive control) and mark which Ladder combinations appear unanticipated, for counsel review. No patentability opinion.

OUTPUTS
A results table per band and fluid; a topology comparison table; a pressure-rating table; a control block description; an R-CL16 cartridge list; an energy table per mode and profile; a list of Founder decisions the study raises (fluid per band, topology, condenser placement), each stated as a question, not decided.