One night, graded on artifacts

Rubric of 32 dimensions authored at 22:45; by 03:30 fifteen of them had moved, every move attached to a commit, a gate output, or a live deploy. Nothing below is claimed without a file behind it.

2.82×

TIR ring vs bore, measured —
vs 2.74 predicted by an independent
radiosity solve. 3% cross-validation.

≥2400 K

emitter floor pinned by the catalytic
cycle itself (yield>1 below) —
chemistry excludes the cold end.

×30

free-Na lift as overshoot s:1→100
at 1753 K — the overshoot cracks
the salt. Feedback now in every cell.

PV ledger
119 W_e = 5.7% fuel→electric at design (Na η_mono 0.255 [EQE 0.99 T × diode; FF MODEL]) + 44 W header preheat credit; 1.6 kW_e Na / 2.2 kW_e K at the 7-sun point. Header-band lever: +9.4% awaiting DF

Salt loop closed
wicking film steady state — 0.19 mg/s circulation, δ≤649 µm, duty 2322/2862/351 W — and the film is optically invisible: Δf_pv ≤ 0.03%, within MC noise.

Consistency proven
one state vector, detailed balance, μ≤A as a code theorem; measured μ(T) threads the Padley–Sugden s-band; Holstein check: matrices vs lumped within geometry ×2.3.

The campaign map — après Minard: the ring ratio's march to a defensible number

Width of the claim ≠ truth of the claim. Each node is a commit; red segments were retracted. The shaded band is where the independent radiosity solve said the answer had to live (2.5–2.9). The march ends inside it.

— retracted claims— interim honest states— calibrated finaly: annulus/bore radiance (log)

Two retractions were the price of the number meaning something. The final point has two independent derivations that agree to 3% — Monte-Carlo transport and matrix radiosity — which is the whole exchange stack validating itself.

Hour by hour — three lanes + the hexcell crew, one night

hour	lane	what landed
22:45	FABLE	RUBRIC-32 authored (F−−−−− → S+++++ ladder, 32 dims); night queues dispatched to both decks.
23:05	vf-tracer	Alumina retag `b35685a`: ρ 0.45→0.9 diffuse on alumina zones, 13/13 conservation gates re-passed, gas→PV +6.0% rel (recycling), cap absorption ÷2.44. 7-sun standoff eases ~6%.
23:15	KIMI	Entire night queue closed: catalytic-cycle pump built (β → per-cycle yield 0.143); wicking-film steady loop (0.19 mg/s); three-node core/annulus/wall + cp(T) with dissociation (no unphysical inlets across all sliders); static 2D merge gate closure 1.000000; Cl scavenging ≤−1% (immaterial).
00:00	hexcell crew	(their own overnight handoff) `ba01e631`: per-λ firefly clamp + a real GPU bug (153 MB buffer silently killing the compute pass), thermal tracer view, label fan-out, provenance chips, Campcell TPMS end caps (two Schwarz-Diamond sheets, sealed counterflow — DF's construction) + panning section plane.
00:10	KIMI	Block 2 deployed `f6264d77`: chemistry pinch drawn on the live μ(T) curve (cold end excluded); rates sourced (k1 Plane 2017 [T], k2 Hynes 1984 [T-family]) — the ≥2400 K discriminator survives sourcing; film transient made semi-implicit (NaN killed).
00:25	both	0.847-vs-0.927 tie-out: byte-level identity — the gap is exactly the 76 mm alumina header bands. Ledger ruling: 0.847 PV-electrical + 0.080 header preheat credit; +9.4% design lever surfaced. Deployed `9b761e77`.
00:45	vf-tracer	Salt-film Fresnel layer `ea1b4c9`: incoherent-slab optics, Δf_pv ≤ 0.03% — the wicking film is optically invisible (n≈quartz). Design intuition numerically confirmed.
01:05	KIMI	Block 3: live draggable μ-inversion widget on /row18 `d9853a01`; Makefile + REPRO.md (4 gates PASS, byte-identical regen); Holstein cross-check (matrix 2.07e-3 vs lumped 4.7e-3 — geometry factor 2.3; 1D slab off ×41).
01:0x–02:4x	vf-tracer	The ring campaign (see map above): washout diagnosed → 1.267 → 16.7 → both retracted (puddle-quartz) → honest 0.142 → σ_a_gas ships (bore self-reversal in the renderer — the row-18 physics, executable) → second axial-clamp bug found+fixed → full-ring cutaway fix.
02:10	KIMI	Block 4 `baa51999`: speciation into the per-cell kernel — the overshoot cracks the salt (free Na ×30); the TRIPTYCH on the front page (photograph / prediction / measurement); what-would-change-our-mind block.
02:55	KIMI	Ring-ratio resolver: steady-state radiosity on the r2 exchange matrices → cavity fills to 55% of Planck (the cells eat the rest) → gain 0.55 → prediction 2.74 (chain in `ring_resolver.py`).
03:25	vf-tracer	Calibrated frame `fa768be`: measured 2.820 vs predicted 2.74 — 3%. Cap double-counting caught (thermal already in the gain budget). Gain-linearity sweep confirms the formula. Ring closed end-to-end.
03:35	FABLE	Morning summary compiled; grades finalized on artifacts.
05:35	FABLE	Hexcell crew's parallel handoff discovered; queues cross-pollinated; their "F" sourcing item claimed & delivered (`SOURCED-OPTICAL-CONSTANTS.md` — incl. flag: kernel α_q=5/m is 50–500× above fused-silica literature; sweep owed before swap).

The ring — the acceptance criterion, met and cross-validated

DF, earlier: "we'll really know if we have it if you can see a ring of brightness because of total internal reflection." Below: light trapped in the quartz wall beyond θ_c = 43.3° waveguides axially and exits at the end face — the annulus. Bore is Planck-limited by D-line self-absorption [MODEL: gray σ over the feature]; annulus carries guided cavity (alumina) radiance at the radiosity-calibrated gain [MODEL: proxy, calibrated on measured r2 matrices].

end-on ring — **end-on — THE SHOT.** ann/bore 2.82× radiance, 1.81× through AgX. 1024²·512 spp. `fa768be`

15 deg off axis — **15° off-axis.** Same calibration, ring persists off-normal.

side view — **side.** Bore-dominated cross-section — correct for a longitudinal cut.

half-arc section — **the half-arc accident → section story.** An inherited cutaway removed half the cavity; kept deliberately as the "half open" section view.

first attempt — **where it started (00:40).** Auto-expose + AgX saturation: the pale-blue washout. Kept in the ledger — the road matters.

Source frames live at ~/cc/sites/dyson-cylinder/render/out-ring/ (e.g. ring-end-on-v8-mid.png); regenerate with SPP=512 node render-ring-v4.mjs-family harnesses. Full retraction ledger + reviewer paragraphs: render/VF-RING-*.md.

Today's frames — the render war, won (v11-v14, 2026-07-21 daytime)

Four operator catches → four shipped fixes: camera-mode color (sodium is YELLOW, B=0), CIE 1931 tables (potassium is DEEP RED, not Wyman-green), twin-tube architecture (the center is the subject; alumina out of the optical path), and the deepest one — the render had μ=0: the flasmon was off.

THE HERO — twin-tube, flasmon driven v14 · 55d486e

twin-tube driven hero — **μ = +0.40 eV, twin-tube.** The current architecture with the drive ON: driven sodium core blazing, both quartz shells as honest Fresnel rings, alumina nowhere in the light path. B=0 on every pixel. This is the device, rendered from its own physics.

twin-tube parity — **μ = +0.25 eV, twin-tube.** The drive one 2023 spectrum implies at its assumed emitter temperature — one observation, inverted, not a design constant.

The lamp turning on — Würfel μ_γ sweep, exposure locked v13 · 4aaa9f6

Same scene, same exposure. S(λ) = B(λ,T_gas)·e^(μ_γ/kT). Bore radiance scales exactly linearly with e^(μ/kT) (measured 1.00 → 2.99 → 5.76, matching prediction to 3 digits). DF's catch: "how is thermal beating radiative? something's wrong" — what was wrong was μ=0. Device physics as an image.

mu=0 LTE — **μ = 0 — thermal LTE.** The gas is just hot: the alumina cavity ring outshines the bore 3.4×. This is the lamp OFF.

mu=0.25 parity — **μ = +0.25 eV — flasmon lit.** The drive one 2023 spectrum implies at its assumed emitter temperature — one observation, inverted, not a design constant: ×2.99 boost, bore/ring parity (1.19).

mu=0.40 driven — **μ = +0.40 eV — line-dominated.** The μ(T)-curve value at 2650 K: ×5.76, the driven line outshines the hottest wall the materials allow.

twin-tube sodium — **v12 twin-tube Na — CURRENT SHOT.** Yellow center as subject; both quartz shells as thin honest Fresnel rings; B=0 everywhere. `7b9808c`

K radiance deep red — **v12 K, radiance view.** Deep red per CIE 1931 [DISPLAY: radiance, not luminance — K is eye-dim, Si-bright: the design point].

K honest dim — **v12 K, human-eye view.** Near-black — correct: cie_y(766nm)≈6e-6. The eye can't see what silicon eats.

golden exposure — v12 Na, −1 stop: the golden regime.

bright exposure — v12 Na, +1 stop: saturated yellow.

v11 single tube — v11 single-tube (superseded): first correct-color frame, cavity ring still in.

Current physics basis — the ladder, with tags

1 · Combustion charges a radical battery. Flame front holds H (and OH) at super-equilibrium overshoot s = [H]/[H]_eq; fast bimolecular shuffles conserve radical number and define the pool potentials; only slow three-body recombination discharges. Affinity A = 2kT·ln s exact, asserted in code. Measured μ demands s ≈ 1.7–23 — inside the measured Padley–Sugden flame range.

2 · Sodium is the recombination catalyst, not a dye. The pump is the catalytic cycle NaOH+H → Na+H₂O · Na+H+M → NaH+M · NaH+H → Na+H₂ (net 2H→H₂, Na cycled). β is the per-cycle electronic yield (0.21 at 2800 K reconstructed) k1 Plane 2017 k2 Hynes 1984 family k3 estimate. Yield>1 below ~2400 K ⇒ the emitter is pinned ≥2400 K by chemistry — the μ(T) band's cold end is excluded, and the swirl design point sits in the allowed window.

3 · The overshoot cracks the salt. Speciation in every cell call: [Na]/[NaOH] ∝ s ⇒ free Na ×30 at the cold end as s:1→100. Positive feedback: drive liberates emitter.

4 · Trapping gives the photon gas a potential. Holstein-imprisoned D-line equilibrates with Na*/Na: μ_γ = μ_Na*−μ_Na. Detailed balance everywhere ⇒ μ_γ ≤ A is a theorem the code exhibits; chemistry off ⇒ μ→0 (Planck floor recovered). Row-18: line at its Planck ceiling ~200 s, self-reversed — μ_γ > 0 for every energetically allowed emitter temperature (+0.1…+1.0 eV band, live-draggable on /row18).

5 · Geometry is a measured operator. Cavity exchange = banded D_ij matrices path-traced on the real 4-inch tube + hex (13/13 conservation gates, reciprocity, κ-ladder for the trapped line). Radiosity on those matrices predicted the ring at 2.74; the renderer measured 2.82. Escape cross-check vs lumped p_esc within geometry factor 2.3.

6 · The system closes. T is a solution, not an input (three-node swirl core/annulus/wall, cp(T)+dissociation); the salt is a closed wicking-film loop in the recuperator (evaporate hot end, condense+recover exhaust end, transparent film — verified optically invisible); PV: Maxeon at ~7 suns, honest three-way split 84.7% electrical / 8.0% header-thermal (preheat credit) / rest caps+quartz ⇒ 119 W_e at the bench design point (Na η_mono 0.255 [EQE 0.99 T × diode; FF MODEL]), 1.6 kW_e Na / 2.2 kW_e K at the 7-sun point.

Glossary

flasmon: quantum of chemically-pumped, optically-resonant emission in a reacting medium; μ_γ = hν(1 − T_gas/T_b).
μ_γ (photon chemical potential): how far a photon gas sits above thermal at its own temperature; μ_γ > 0 = super-Planckian drive.
s (radical overshoot): [H]/[H]_eq. The off-equilibrium coordinate; A = 2kT ln s exactly for the H+H channel.
A (affinity): free-energy drop of the clamped recombination — the battery voltage the photon gas can draw on. Ceiling for μ_γ.
catalytic pump / β: Na cycles through NaOH/NaH carrying recombination flux; β = fraction of cycles landing Na in 3p (electronic yield), not a collision statistic.
self-reversal: optically thick line whose core saturates on the cooler skin — the dip between the two wing peaks in row-18. A depth-scan of the source function, and the μ-meter.
Holstein trapping / p_esc: resonance-photon imprisonment; effective escape probability. Lumped 4.7e-3 (kernel-scale) vs matrix 2.07e-3 (device-scale) — consistent.
D_ij matrices / κ-ladder: Hottel-zone exchange factors computed by path tracing the real geometry, per band; the trapped line gets a ladder of opacities interpolated at runtime (k-distribution × zone hybrid).
r2: second matrix generation: alumina retagged ρ≈0.9 diffuse. The set the radiosity solve and PV ledger run on.
radiosity resolver: steady-state balance on the D matrices → cavity radiance at the shell → calibrates the render's cavity-gain (0.55 of Planck fill; the Maxeon lining eats the rest).
TIR ring: light trapped in the quartz wall beyond θ_c≈43.3° waveguides to the end face — the bright annulus. The acceptance test of the transport stack.
wicking film: NaCl condenses on cool surfaces as a transparent liquid layer draining back to the evaporation zone (Nusselt falling film) — closed salt loop in the recuperator; optically invisible (n≈quartz).
TPMS / Campcell caps: triply-periodic minimal surfaces; the end caps are two Schwarz-Diamond sheets (one z-flipped) = sealed counterflow HX wall (DF's construction), bounded by sphere∪cylinder.
header bands: the 76 mm alumina injector/plenum bands at the hex axial ends — absorb 8.0% of gas emission as preheat credit, not PV current. The +9.4% lever if cells extend into them.
tags: [T] [MODEL] [ESTIMATE] [RETRACTED]: sourced-with-lineage / deliberate modeling choice / order-of-magnitude pending source / withdrawn claim kept on the ledger. No bare numbers.

TODO & sign-offs

Awaiting Danielle

Ring acceptance reframe — bench photo shows blaze central, glow around it; propose criterion = "distinct annulus vs surround" (met at 2.82× vs bore). Approve?
PV header-band lever — +9.4% electrical if Maxeon extends into the header bands; cell area vs manifold packaging. Direction?

Open engineering (owners)

Deployed-page scene bugs (hexcell): gas clamp (fix is copy-paste from eacf8eb), multi-sun env ghosting, boot-into-night default.
SSS ceramics + realistic ground + faceted hex collector in the live scene (hexcell; TPMS caps already in).
α_q discrepancy: kernel 5/m vs literature 0.01–0.1/m at 589 — ring-sensitivity sweep, then swap (flagged in SOURCED-OPTICAL-CONSTANTS.md).
Two-zone gas self-reversal in the render (hot core / cool skin) — lifts the ring honestly; magnitudes documented.
ε_vis / ρ_eff coupon measurement — one env var closes two [MODEL] tags.
k3 rate source; dynamic (time-stepping) Fork-A merge; Voigt σ_a(λ) line shape.

Done tonight (proof attached)

Ring closed, cross-validated 3% (fa768be + ring_resolver.py)
Catalytic pump sourced; ≥2400 K pinch on the live page (f6264d77)
Speciation s-breathing in kernel; triptych; WWCOM block (baa51999)
Salt film loop + optical invisibility (ea1b4c9); PV three-way ledger (9b761e77)
Live μ widget + Makefile/REPRO gates (d9853a01)
r2 matrices retag 13/13 gates (b35685a)

Artifact index

artifact	where
Ring frames (v8 = ships)	`~/cc/sites/dyson-cylinder/render/out-ring/` · copies on this page at `/assets/night/`
Ring reports + retraction ledger	`render/VF-RING-.md`, `VF-GATES.md`, `VF-FILM.md`
Exchange matrices	`render/vf-out-r2/` (+ r1 baseline) — f32 row-major + manifest
Physics kernels	`~/cc/flasmon/`: catalytic.py · cr_cell.py · cr_ness.py · consistency.py · static_merge.py · system_model.py · ring_resolver.py
Rubric + night log	`~/cc/flasmon/RUBRIC-32.md` · `NIGHT-LOG.md` (chronological, grades, corrections)
Repro	`~/cc/flasmon/Makefile` + `REPRO.md` — every figure one command, 4 gates
Live pages	/row18 (μ widget, pinch, ledger) · front (triptych, §07 pinch) · coupled sim
Crew handoffs	`workers/…/dyson-cylinder/HANDOFF-OVERNIGHT.md` (hexcell crew) ↔ `render/HANDOVER.md` (bug ledger) — cross-pollinated 05:35

flasmon.com/night · compiled by FABLE from NIGHT-LOG.md · every claim traces to a commit, gate, or deploy · nothing pushed to remotes