CALC-056

One-lever-at-a-time sensitivity sweep

Feeds CLM-0248, CLM-0250, CLM-0407, CLM-0408, CLM-0410, CLM-0467. SHARED - six claims, one section. This is the largest single reduction in the catalogue. Answers QST-004.

Script. lcoe-model.py SECTION 7, section_sensitivity() (line 732), with clone_params() (690), rescale_for_efficiency() (703) and electric_propulsion_scenario() (719). RD1 throughout, 3 per cent discount rate, all other inputs at baseline.

Why it was needed. The source report's own sensitivity suite tests each lever in one direction only - for example it tests manufacturing learning improving slightly, never worsening. A sweep that tests both directions can find downside risk that does not appear in the published ranking at all, and it did.

Baseline for every row: RD1 at 0.6231 $/kWh.

Working, transcribed. Every row is one lever moved, everything else held.

LEVER                                        LCOE $/kWh    |delta|    note
baseline                                       0.6231         -
module learning rate 0.75 -> 0.85              1.0650      0.4420   WORSE; largest tested
solar cell efficiency  0.35 -> 0.70            0.1884      0.4347   mass rescaled inversely
electric propulsion (refuel 12 -> 0)           0.2069      0.4162   +17.2% mass penalty paid
capacity factor      0.997 -> 0.60             1.0354      0.4123   RD2's value on RD1
launch cost      $100M -> $10M                 0.2170      0.4061
hardware lifetime  10 -> 15 yr                 0.4771
hardware lifetime  10 -> 20 yr                 0.4570
hardware lifetime  10 -> 30 yr                 0.3302      0.2929   47.0% reduction
launch cost      $100M -> $50M                 0.3980

Correlated pairs, handled as pairs and never independently. Two rows above are not single levers and must not be read as such: - Solar cell efficiency. Module count, and therefore manufacturing and launch cost, rescales exactly inversely with the efficiency-chain ratio, holding delivered power at 2 GW. This matches SRC-012's own description of its efficiency sensitivity, printed p.75, "reduction in mass to deliver the same 2 GW of power". - Electric propulsion. Refuel-ratio removal and the +17.2 per cent mass penalty are two consequences of one architecture choice and are varied together. - Hardware lifetime changes both the maintenance launch count and the number of units fed into the learning curve, propagated together, not as two free parameters.

Results, against the source report's own figures.

The finding that is this model's and not the report's. SRC-012's own sensitivity suite tests learning rate only improving (its Manufacturing Sensitivity 2, a 3-4 per cent saving), never worsening. A 0.75-to-0.85 learning rate is the largest single swing in this whole sweep, larger than driving launch cost from $100M to $10M, and it does not appear in the published ranking in any form. The effect is a consequence of the very large production run - about 4.4 million modules and servicers over 30 years across all designs.

Where this fails. - A sensitivity ranking within the range tested (learning rate 0.65-0.85), not a probability-weighted risk assessment. No view is offered on how likely a 0.85 learning rate is. - Capacity factor is a consequence of a design's pointing architecture, not a free dial. The 0.997-to-0.60 row isolates its numerical effect only and is not a claim that RD1 could achieve RD1's mass at RD2's capacity factor. - The 30-year hardware lifetime row is three times the report's own 10-year baseline. SRC-012 states today's GEO hardware baselines to 15 years (printed p.74), itself optimistic relative to SBSP's harsher duty cycle. - The 70 per cent solar cell efficiency is NASA's own stated theoretical state of the art, not a flight-qualified state of practice. The same source separately gives NREL's tracked best research cell as 47.6 per cent. The row treats efficiency and cost-per-module as independent, which may not hold. - The electric-propulsion row does not model the report's own noted trade-off that electric transfer takes about four months instead of one, which affects financing-cost timing at higher discount rates.

FOUR LOCATOR REPAIRS ALREADY MADE IN THESE CLAIMS, CARRIED HERE SO THEY ARE NOT LOST. Every one was found by independent verification and every one was a citation defect, not an arithmetic one - all LCOE figures reproduced exactly on rerun in all four cases.

A citation to a page that does not exist would lead any checking reader to conclude the claim was invented. That is why these are repaired rather than left, and why the page numbers in this record are the corrected ones throughout.

Cite this