A rocket exploded over Florida and lit the night sky orange, and the same week a physics frame quietly inverted under our feet. This sheet records both: a digital twin of the LightCell combustion-driven sodium emitter — and the unplanned full-scale experiment that Blue Origin's New Glenn handed us on Launch Complex 36. The thesis is one sentence. The light is chemiluminescent, not thermal — its brightness ceiling is a chemical bond, not a temperature — and that single correction moved every number by orders of magnitude.
Both an Opus-4.8 thread on the desktop and a parallel one on claude.ai independently built the emitter on a Planck/Boltzmann ceiling — and independently had to tear it out. In a 300-GW pumped medium the sodium 3p/3d populations are filled by chemical bond energy, not collisions. The excitation temperature decouples from the gas temperature; there is no exp(−hν/kT) tail to sit on.
The ceiling is the bond energy converted to kelvin:
| Bond / threshold | eV | T-equiv |
|---|---|---|
| NaCl formation | 4.26 | 49,400 K |
| Na 3p (589 upper) | 2.10 | 24,400 K |
| Na 3d (819 upper) | 3.62 | 42,000 K |
| CO bond | 11.16 | 129,500 K |
The device: methalox combustion + NaCl ablated off boiling-temp walls into the oxygen stream, radiating the Na D-lines through a transparent wall to a tuned photovoltaic. Twin of granted patents US12136898B2 / US12494732B2. Not a weapon — incoherent line emission into a cavity, nothing to aim.
The Instruments| Kernel | Check |
|---|---|
| Voigt H(a,x) Humlíček W4 (Zig) | ↔Rust 1.7e-11 |
| Cylindrical Holstein trapping | reviewer-OK |
| Cantera speciation, 360 pts | 0 fails |
| Partition kernel → WASM | native↔WASM |
| 1-D RTE transport → WASM | self-reversal |
Everything compiles to WebAssembly and runs in this page. The trapping math is ported verbatim from a reviewer-audited Rust kernel and cross-checked to floating-point round-off.
The double-hump is not imposed by an escape factor — it falls out of integrating the radiative-transfer equation along a chord through a hot-core/cool-skin cylinder. Thin → flat line; thick → the skin absorbs the core's line-center photons and re-emits at its own lower source function. Raise the chemiluminescent excitation above the kinetic temperature and the line fills back in. This is the HPS lamp's double-hump and the New Glenn fireball's yellow, same physics.
The EventOn 2026-05-28, ~21:00 EDT, New Glenn NG-4 deflagrated on LC-36 during a static fire. Methalox + sodium contamination = an unplanned LightCell shot. Nathan Schmidt called it "mostly yellow"; a clean methalox flame is blue-white, so yellow means the sodium D-lines dominated the visible. Frame photometry off the closest camera: R/B ≈ 1.8, persistently orange-yellow — confirmed.
Peak optical power, two independent methods agreeing:
| Method | Peak optical |
|---|---|
| Propellant energy ÷ burst | 375 GW – 2 TW |
| Skyglow saturation (lower bound) | ≳ 100s GW |
"Gigawatts and gigawatts," as the operator said — confirmed twice. Both are energy/saturation bounds, not calibrated radiance.
Clipping the fireball core doesn't matter if you measure the reflected light. Calibrate pixel→luminance off the Cape Canaveral Light (DCB-224, ~1 Mcd, in-frame, 1.16 km from the pad, same atmospheric path), read the firelit sand/water at known distance, invert the square law. Cameras at 0.5 → 80 km overdetermine it and solve the atmospheric extinction for free.
| Vantage | km | state |
|---|---|---|
| NSF @ Rusty's | 10.3 | saturated |
| SFN @ Fishlips est | 10.1 | saturated |
| Cape Canaveral Comm. Ctr | 9.3 | in-frame |
| Cocoa Beach Pier cam | 13 | archive ≤48h |
| Playalinda / Titusville | 22–28 | distant |
| East / Downtown Orlando | 60–80 | skyglow |
Three independent OSINT agents converged on the near-field geometry and each caught the same 2.6-km error in our pad coordinate. The Surfline archives expire in ~48 h.
Honest LedgerThe chemiluminescent rate matrix (which reaction pumps Na*, branching yields) is exposed as sliders, not derived. The soot knob is uncalibrated and drives the selectivity hard. The transport solve is one chord, Doppler-only, 589 alone. No photometry has been run on a real geolocated frame yet — the radiance numbers are bounds, not measurements. Nothing here has had an adversarial review pass. The map is honest about being a map.
~/cc/projects/lightcell/ · partition.wasm + transport.wasm run live above ·
measured est vs verified marked throughout.