The review caught real physics errors. The GIF caught real engine physics.
The PMT count was wrong by 9×. The first build copied the reference
detector's 169 + 199 tubes verbatim. But a ×3 linear scale-up means ×9 the array area —
same 3-inch tubes, nine times as many. Fixed to 1521 + 1791. Then the window size
was wrong twice: first scaled ×3 along with the detector (windows 2.16× wider than their
hex pitch — massive overlap), then "fixed" to the casing diameter, still 2 % too wide.
The correct number was in our own geometry table all along: the photocathode is ⌀64 mm,
the casing ⌀76 mm.
The electron lifetime contradicted the drift length. τₑ = 1 ms sounds
clean until you notice the full drift is 2.6 ms — cathode events would lose 93 % of their
S2. Review flagged it; purified to τₑ = 15 ms (84 % survival). The background budget was
also ~1000× too low and quietly contradicted the "neutrino fog" story the animations were
telling — raised to literature scale, the ⁸B CEνNS floor now genuinely dominates.
The first capture recorded a 0.0 MB GIF of pure darkness. Modern
three.js uses physically-correct light falloff: point lights that read intensity 15 die
within metres, and a 30 m cavern swallowed all of them. The fix was twofold — opaque
materials got a same-hue emissive floor, and the S1/S2 flashes became cavern-scale light
pulses (intensity 110–140), which is also just better storytelling.
The raised camera teleported to the surface. To fix a too-low viewpoint
the capture camera moved to z = 13.5 m — 1.7 m from the exit-zone centre, radius 1.8 m.
The engine dutifully decided the player had walked to the exit and faded back to the
Martian surface; the "neutrino detection" GIF came back as a starfield. The shot now
keeps 5 m of clearance from the door.
A strict ×3 water tank cannot exist. The reference tank is ⌀10 m;
×3 is ⌀30 m — the cavern span. The delivered design uses a two-tier scale: inner detector
strictly ×3, outer shielding at spec targets (⌀14 m), a deliberate, documented compression.
From the TES station — the band plan broke first. The original 140
and 166 GHz bands overlapped across 141–161 GHz: fine on separate pixels, impossible to
diplex complementarily on one. Fixed at the system level with a 152.4 GHz crossover,
which also restored the saturation margin the too-wide filters had eaten.
Everything passed alone; the assembly collapsed. Antenna, diplexer
and filters each verified standalone — the first full-wave cascade returned 12–27 %
coupling. Four defects dug out over four cascade versions: never-designed 750 µm tee
arms, a 390 µm feed line missing from the circuit model, two antenna versions coexisting
in different files, and a 40 µm drawing-leftover feed stub that capped one band at
54.5 % — a Bode–Fano wall no matching network could climb. A three-point tail sweep
found the real optimum at 150 µm (≈λ/8, not the textbook λ/4; slot loading shifts it),
and v4 closed at 77/78 %.
Circuit models flatter dead antennas. A two-element matching study
promised 80.5 % on a proposed 119+195 GHz antenna-reuse pair; the converged full-wave
cascade returned 42.0 % — the twin-slot's radiation resistance collapses to 6.6 Ω above
190 GHz. The negative result stands in the ledger: high bands get their own antenna,
and feasibility verdicts are full-wave or they are nothing.
