OpenCretin — LOAD-BEARING implementation doc

Sitrep — where the build is

THE THESIS (W0): trap the 589 resonance deeply ($t_{589}\sim 10^3$) → it builds the 3p reservoir → collisional funnel to 3d → the THIN 819 sideband ($t_{819}\sim 10$) escapes and harvests. The asymmetry IS the device. Operator's point from session 1; the work was clearing smuggles off the top of it.

Goal: a line-by-line, multi-temperature, 2-D radiative-transport sim — no Boltzmann-at-T, no S=B, no single-T, no scalar θ, no linear Holstein. Self-documented here (updated every change), live at plasmagicians.com/opencretin/, until the live spectrum.html is re-skinned to this engine.

Stage: CROSS-VALIDATED vs sodium_package. Built: gibbs→cascade→CR(n̄) pipeline, Saha + ionization boundary (~16 kK), NaCl transport region. Now CROSS-CHECKED [RUN] against the independent claude.ai sodium_package engine: my b_3d ≈ 272 at g₀=6000 vs their b_3d ≈ 320 (two engines converge); my n3d/n3p = 0.003 stays under the 1.667 gain threshold → "never inverted" confirmed. Their S=B(Tₑ) and one-temperature smuggles are the SAME corrections I made independently.

Next (re-ordered by sodium_package): (1) 2-D escape — 1-D flame (z) + radial escape (r) with the nonlinear bleaching window — THE headline, retires the 0-D θ; gives steady overdrive 6×→48× (W7); (2) departure coefficients b_i + per-transition T_exc as the diagnostics; (3) Saha-detailed-balance gate (dense→Saha proof); (4) re-skin spectrum.html. Holstein g₀ is void at the operating point (B4) — the 2-D nonlinear window replaces it.

WEIRD What makes this device's physics unusual

W0 · THE DEVICE THESIS — trap the resonance (589), harvest through the thin sideband (819)

This is the whole machine, and it was the operator's point from the very first session. The 589 D-line (3p→3s) is deeply trapped ($t_{589}\sim 10^3$, escape $\sim 1/2350$) — a resonance photon is emitted and reabsorbed thousands of times. That trapping is not a loss: it builds the 3p reservoir, which feeds 3d by collisions, and the 819 line (3d→3p) is optically thin ($t_{819}\sim 10$) so it escapes and lands on the PV. The asymmetry IS the device: trap the resonance to pile up population, harvest the energy out through the thin sideband.

Verified [RUN] funnel.mjs: deepen the 589 trap $t_{589}$ 10→6000 and the 3p reservoir builds 150×, 3d fills via the collisional funnel, and the 819 escaping power climbs 160× (2.2e4 → 3.5e6) while trapped-589 escaping power stays flat. The 819/589 escape ratio rises 0.00 → 0.30 — the device routing ever more of its output through the harvestable channel as the resonance trap deepens.

prov: [RUN] funnel.mjs. [READ] holstein-cretin L8: "this makes the 819 channel optically thin or moderately trapped where the 589 channel is deeply trapped." HONEST NOTE: this asymmetry was the operator's thesis throughout; every smuggle (S=B, single-T, scalar θ, Holstein-linear) BURIED it, and each correction was clearing a layer off the top of it — not discovering it. It is the first fact, not a late conclusion.

Falsified if: 819 is NOT optically thinner than 589 at device loading (then there's no funnel, no selective harvest). It is — t_589/t_819 ~ 100, structurally, because 589 is the resonance line to the ground state (huge ground population absorbs it) and 819 terminates on the sparsely-populated 3p.

W7 · Overdrive is STEADY, via a hot core seen through a bleached nonlinear window — and radiation transport is ≥2D, never one scalar θ

From the sodium_package reasoning chain (operator-forced corrections): the single biggest simplification anyone made was collapsing radiation transport into one escape factor θ in one zone — "structural blindness," it already produced a FALSE claim ("overdrive is transient"). The floor is 2-D: 1-D along the flame ($z$ — the matter, the burn) + the radial escape ($r$ — core→cool-skin→window, where self-reversal and the bleaching aperture live). The photon doesn't follow the flame: it's generated by an axially-varying source and escapes across the radial gradient, on rays that couple the two.

The mechanism: a hot core radiates through a cool optically-thick Na skin. $S=B$ holds per zone, but you see the hotter INTERIOR through the window → emergent brightness exceeds the surface graybody (overdrive relative to skin temp, NO thermodynamic violation). The window is NONLINEAR: strong pumping bleaches the line opacity (excited fraction 24–37%, ground depletes), opening the aperture. Self-reversal ↔ overdrive is the nonlinear-window switch — and it is STEADY, not a transient flash. This sits ALONGSIDE the n̄ population elevation (W3): two routes to brightness — pumped populations AND geometric (hot-core-through-bleached-skin).

Quantified [their rad2d, verified Zig/WASM]: ceiling $48\times$ skin-graybody (2300 K core / 1686 K skin Planck ratio at 589 nm). Unbleached self-reversal pins line-center to $B(T_\text{skin})$ → only the WINGS escape, $\sim$6.4× band-avg. Bleaching the skin open fills the center back toward $B(T_\text{core})$ → $\sim 8\times$ band-brightness gain. The 6×→48× gap IS the self-reversed line center locked behind skin line-center opacity.

prov: [READ] sodium_package full reasoning chain (twozone.py, rad2d.zig). CORRECTS my OpenCretin, which is still 0-D single-cell — same θ error. This is the headline thing to build next.

Falsified if: the device has no radial gradient (uniform column) — then the geometric overdrive vanishes and only the n̄ population route remains. It has a gradient (hot core, NaCl-boil skin at 1686 K). The magnitude is open until the 2-D nonlinear-window transfer is built.

W1 · The source function is NOT a blackbody

For a pumped / non-LTE line emitter, the RTE source is the line emissivity $S = j_\lambda/\kappa_\lambda = \tfrac{1}{\kappa}\cdot\tfrac{A\,n_u\,h\nu}{4\pi}\,\phi(\nu)$, not B_λ(T). Kirchhoff's law is derived under LTE; quoting it for a pumped gas smuggles in a fake Planck continuum that buries the lines (measured: 99% of "power" between the lines).

prov: [RUN] — switching B_λ→j_λ moved 589 from 0.2%→76% of power in the prior kernel. See plate 02-source-function-not-blackbody.

Falsified if: at the device's collision rate the upper level fully thermalizes (ε→1), making S→B_λ legitimately. Test: compute ε = collisional-destruction probability.

W2 · "Temperature is nowhere defined" — populations are the state

There is no single temperature. {brightness temp, excitation temp, kinetic temp} are three different numbers and must never be collapsed. The CR solver carries level populations as the state; any "temperature" is a readout. (Astrophysical-maser convention-setter: T_rad ~ 10¹⁵ K in gas at 10²–10³ K.)

prov: [RUN] cr.wasm recovers Boltzmann at thermal rates with NO T in the solver; floats as a readout under pump. From Steve's CR handoff + GPD maser lesson.

Falsified if: the three temperatures coincide everywhere (full LTE). The whole device thesis requires they don't.

W3 · The 3p elevation is NOT a pressure gate — it falls out of the mode-occupancy waterfall cascade

CORRECTED (operator, 2026-06-05): an earlier draft asserted "low-P floats, 1-atm PINS" from Steve's lumped inequality $P_\text{chem} > C_\text{vib}\,k_\text{VT}\,\Delta T$. That is the 0-D collapse the GPD handoff explicitly warns against (a verdict hardened under a new banner). The inequality is a single lumped V-T number; the real object is a waterfall of mode occupancies 3s→3p→3d→4p→…→continuum, where each rung's occupation is the net of every up/down channel (collisional, radiative-with-trapping, pooling, stimulated trapped-field re-pumping). Whether 3p (or 3d via pooling) sits above Boltzmann is the OUTPUT of the cascade solve, level by level — not a pressure rule asserted upfront. The trapped field can feed the waterfall faster than quench drains a rung even at 1 atm.

prov: [RUN] cascade.wasm gives 0.15% chem-pump-ONLY at 1 atm — but that is chemistry WITHOUT the trapped-field re-pump term (now being added). Pressure is one input to the cascade, not a gate on it.

Resolved by: running the full CR waterfall (gibbs→cascade→CR with the B01·J̄ trapped-field term + pooling) and reading where each level balances vs its Boltzmann value, as a function of (n_e, T, n_Na, J̄). The answer is a surface, not a binary.

W4 · Core + sheath, NOT 0-D — the ionization boundary is the master regime transition

The device is a hot (possibly ionized) core + a cold neutral D-line-dominated sheath through which interior radiation escapes (self-reversal — same physics as a stellar absorption line or a self-reversed streetlamp). P_Dline/P_total is a function over many OOMs, not a number: governed by Saha ionization (exponential in n_e, T_e), line optical depth (D-core saturates to Planck at line center), and T_e vs the 2.1 eV D-line / 5.14 eV ionization potential.

QUANTIFIED [RUN]: Saha + continuum solver gives the boundary as a smooth surface — P_Dline/P_total = 1.000 at 3000 K → 0.47 at 16,000 K (the crossover) → 0.06 at 30,000 K. The device runs LINE-dominated at flame T (≤4000 K); only the extreme-pulse core (>16 kK, the 440 GW regime) becomes a continuum-with-a-D-dip. Found by extending the sweep until the crossover appeared, NOT asserted. The "9%" GPD warned about was the TRK oscillator-strength fraction — a different surface from emission.

prov: [RUN] saha.mjs + continuum.mjs. [READ] GPD 02-lightcell-physics-state (Danielle: "is that true in the cold outer sheath?"). FLAGGED coupling: trapping g₀ weakens as neutrals ionize (g₀_eff = g₀·(1−ion_frac)) — added; the full core/sheath 2-zone transport is still single-zone here.

Falsified if: the crossover sits far from ~16 kK once the real free-free Gaunt factor + recombination edge + 2-zone self-reversal are in. Current prefactors are Kramers order-of-magnitude.

W6 · NaCl transport into the emitter is ENHANCED over naive vapor pressure — but mostly geometry+aerosol, not the chemistry I first guessed

Operator: "NaCl transport region ... enhanced relative to naive because even at low temp it fumes. maybe water mediated." Built + checked [RUN] (transport_nacl.mjs), four channels from real nasa_gas.yaml thermo: monomer (CRC vapor P) + dimer (NaCl)₂ + H₂O hydrolysis + aerosol/wick. Honest result: the chemistry-anchored enhancement is only ×1.3–1.9 (dimer "fuming" is a real factor, not OOM; the CRC monomer P I first used was ~6 OOM wrong, re-derived). The big multiplier (×38–49) is the wick film-area + aerosol carryover — physically real but NOT yet anchored (placeholder prefactors, flagged).

The water-mediation lives downstream, not at the wall: K_hydrolysis at the wall is tiny (1e-6–1e-3), so it barely boosts the delivery FLUX. But gibbs shows H₂O clearly shifts the gas-phase partition (dry→wet: free Na 23%→31%, NaOH 30%→40%). So water mediation is a COUPLED gas-phase-equilibrium effect (NaCl + H₂O → NaOH + HCl in the hot gas), not a wall-flux enhancement. That's the coupling resolved.

prov: [RUN] transport_nacl.mjs + gibbs dry/wet sweep. FLAGGED aerosol prefactor + wick area are order-of-magnitude placeholders; the ×38 full enhancement is NOT defensible yet — only the ×1.3–1.9 anchored part is.

Falsified if: the wick area ratio + aerosol entrainment, when measured, are ~1 (no geometry boost) → enhancement collapses to the ×1.3–1.9 chemistry floor. The big number rests on unmeasured geometry.

W5 · NaCl is a CLOSED mass loop — not a loss

NaCl reaches a mass equilibrium, rewicks, and the latent heat recovers. It is a recirculating inventory, not a consumable. The speciation partition (free Na vs NaOH vs NaCl) sets seed inventory for a target radiating density, NOT an energy ledger. The only residual effect is small + thermal: the finite rate of heat transport pins the envelope a little hotter than the infinite-transport limit.

prov: [RUN] gibbs.py pure-O₂ sweep (free Na 61%@2800K → 83%@3400K). Operator correction 2026-06-05 (I had laundered a 136 kW "loss" — wrong).

Falsified if: salt is lost from the loop (incomplete rewick / carryover). Then it IS a consumable + a real enthalpy loss. Depends on wick capture efficiency.

LOAD-BEARING Assumptions that, if wrong, collapse the conclusion

B1 · The trapped-field re-pumping floats 3p — BUILT + VERIFIED, via photon occupation n̄

The "elevated 3p" thesis rests on the D-line photon being emitted ~g₀ times before escaping, re-pumping 3s→3p. Now built in cascade.zig as the dimensionless photon occupation n̄: stim/spont = n̄, so 3p→3s gains A10·n̄ and 3s→3p gains A10·(g1/g0)·n̄. Verified [RUN]: n̄=0→pure chemistry; n̄=n̄_LTE(=1.6e-4)→Boltzmann (×1.05, the LTE invariant); n̄≈g₀·n̄_LTE (trapping anchor)→T_exc 7663 K, ×253 above Boltzmann at 1 atm. The earlier $B_{01}\bar J$ form was dimensionally wrong by ~16 OOM (re-derived per GPD rubric → n̄ is the clean handle).

prov: [RUN] test_field.mjs sweep. The remaining unknown is the MAGNITUDE of n̄ at device conditions — set by trapping geometry (escape factor g₀), which stage 3 (2-D Λ-iteration) derives.

Falsified if: the Λ-iterated field gives n̄ near n̄_LTE (no trapping enhancement) → 3p stays Boltzmann. The elevation is real IF the photons are genuinely trapped g₀≫1 times. That g₀ is B4.

B2 · Equilibrium speciation (Gibbs) is the right baseline inventory

Layer 1 assumes NaCl/Na/NaOH reach chemical equilibrium in residence time. gibbs.py validated to JANAF 0.1 cal/mol/K and caught the doc's NaOH-30× error + dropped-NaCl. If kinetics freeze the composition (finite vaporization rate), free-Na could differ.

prov: [RUN] regression reproduces freeNa 61.2/NaOH 26.3/NaCl 12.2% @2800K exactly.

Falsified if: a Damköhler-number check shows chemistry slower than residence time. Not yet computed.

B3 · Na D-line atomic data (f, A, g_u, E_u)

Everything downstream scales with: D₂ f=0.6411 / D₁ f=0.3202 (2:1), A≈6.16e7/6.14e7 s⁻¹, the 2.104 eV upper level. AMO-rubric cross-check: TRK Σf=Z=11; f(3s→3p)≈0.978 total exhausts only ~9% of the oscillator-strength sum (NOT 9% of emission — emission is upper-state-population weighted; that transposition is a known category error).

prov: [READ] NIST ASD values in LINES[]; GPD AMO file uses Na D as its TRK example.

Falsified if: Σf over our included lines exceeds Z=11 (→ matrix-element error) or ≪ Z (→ missing continuum). Gate R1 below.

B5 · Cross-validated against the independent sodium_package engine

A separate claude.ai build (sodium_package.html, se_engine.py + rad2d.zig) solves the same device with departure coefficients $b_i = n_i/n_i^\text{LTE}$ (Menzel/nebular). Two independently-built engines converge [RUN]: my b_3d ≈ 272 at g₀=6000 vs their b_3d ≈ 320; both find n3d/n3p well under the 1.667 gain threshold (NEVER inverted — "strongly non-LTE, never a laser"). Their named smuggles (S=B(Tₑ), one-temperature, Rosseland diffusion) are the SAME errors I corrected independently → mutual validation of the frame.

prov: [RUN] xcheck_manifold.mjs. [READ] sodium_package.html. To yoink: b_i parameterization, the Saha-detailed-balance gate (dense→Saha proof), the 2-D self-reversal lever (48× ceiling / 6× unbleached / ~8× bleaching gain).

Falsified if: the engines diverge once both use real van Regemorter/Lotz rates + the same atomic data. Current agreement is at the order-of-magnitude rate level.

B4 · Holstein escape factor g₀ is the LINEAR theory — VOID at the bleached operating point

The 589 trapping uses g₀ from Molisch & Oehry (cyl-Doppler, opencretin CYL_DOPPLER). But Holstein–Biberman is linear by construction — it assumes fixed ground density + vanishing excited fraction, so the opacity is a frozen field. The device violates that: excited fraction runs 24–37% (sodium_package, computed), the ground depletes, net opacity $k = k_0[n_1 - (g_1/g_2)n_2]$ becomes intensity-dependent — the medium bleaches and trapping self-limits. So my n̄ ≈ n̄_LTE·(1+g₀·f_trap) with a frozen g₀ is a stand-in; the real g₀ is a functional of the radiation field, solved in the 2-D nonlinear-window transfer (W7), not a tabulated constant. An escape factor is at most a DIAGNOSTIC read off the converged field, never an input coefficient.

Holstein-cretin Layer 1 [READ] gives the real g₀(k₀L) + the key wing insight: $g_0^D = 1 + \tfrac{k_0 L}{m_0}\sqrt{\ln(k_0L/2+e)} - \text{(rational)}$; Doppler-slab $g_0$ = 12.9 / 177 / 2350 at $k_0L$ = 10/100/1000; Lorentz-slab $g_0\approx 49$ at $k_0L{=}1000$. "The lineshape is the load-bearing parameter — not the strength": Doppler vs Lorentz differ by 48× at identical opacity because Lorentz wings $1/x^2$ give photons an escape route Doppler wings $e^{-x^2}$ deny. Wing-escape (Zanstra/Irons): Doppler $x_\text{esc}=\sqrt{\ln k_0L}$, Lorentz $x_\text{esc}\approx\sqrt{k_0L}$. This fixes my rad2d wing calibration — the ~6× unbleached escape lives in the Lorentz wing at $x_\text{esc}$, which my crude Voigt underweighted. Cross-check: design $t_{589}{=}10^3 \Rightarrow g_0\approx 2350$, so escape ~1/2350 — my frozen 1/6000 is the right order, slightly trap-heavy.

prov: [READ] sodium_package + holstein-cretin Layer 1 (g₀ formula + numbers) + Layer 8 (t_589=10³, t_819=10, Na 3p ~8.8e-4 — cross-checks my f3p). DEBT frozen g₀ → nonlinear self-consistent k(J̄) in the 2-D transfer; use Zanstra/Irons $x_\text{esc}$ for the wing escape.

Falsified if: the excited fraction stays ≪1 at the operating point (Holstein then valid). It does NOT — 24–37% excited is computed, so Holstein-linear is void here regardless. (Holstein-cretin defers the bleached regime to its Layer 5/6 — the CRETIN regime — confirming the layering.)

SHORTCUT Every current placeholder we owe a fix

S1 · LIVE spectrum.html lets T_exc float freely (NO V-T quench) — WRONG, shipped

compute() solves T_exc from a power-balance bisection capped at the bond ceiling, with no V-T quench term, so cranking input power floats T_exc to 49,400 K at any pressure. Per W3 this is unphysical at 1 atm. Currently live at sim.lightcellenergy.com. Fix = the CR engine + V-T quench (this build).

prov: [READ] spectrum.html:224–238.

S2 · LIVE spectrum.html f_dissoc (0.6–0.9) misses NaOH entirely

n_Na is derived as P_NaCl/(kT)·f_dissoc with a hand-set dissociation fraction. gibbs.py shows ~26% of Na is NaOH at 2800K — the stub ignores it. Fix = gibbs speciation as Layer 1.

prov: [READ] spectrum.html:215.

S3 · Trapped-field J̄ not solved — g589/g819 are frozen sliders

The escape factors (g589=1/6000, g819=1/40) are constants, not derived from the actual photon density. The whole 3p-elevation question needs them to emerge from a Λ-iterated 2-D field. Stage 3, unbuilt.

prov: [READ] cascade.zig:18.

S4 · CR kernel is 4-level (3s/3p/3d/4p), NASA-7 thermo (6000 K ceiling)

No ionization stage in the CR levels yet (Na⁺ enters only via gibbs). Hot cores past 6000 K need NASA-9 thermo. Extend NLEV + add Saha coupling for the ionization boundary (W4).

prov: [READ] cr_kernel.zig NLEV=4; gibbs.py NASA-7.

S5 · v17 doc numbers are AIR (N₂); device is pure-O₂

Steve's combustion_tpv_v17.pdf is a hydrocarbon-AIR system (xN2=0.729). Do not import its efficiency numbers into the pure-O₂ twin without re-running. The N₂ changes bath chemistry, V-T partners, and adiabatic flame T.

prov: [READ] v17 §abstract, xN2=0.729.

AMO rubric gates (from GPD verification-domain-amo — the litmus that passed)

Use as a RUBRIC, not a solver. Verify by re-derivation (substitute, take limits, trace dimensions), never by grep. Status fails closed: VERIFIED needs computation evidence; everything short is PARTIAL/FAILED. completed (ran) ≠ passed (physics checks out).

Gate	Check	Status
R1 · TRK sum	Σf over included Na lines ≤ Z=11; report fraction exhausted. f(3s→3p)≈0.978 alone.	TODO
R2 · E1 selection	Every modeled line obeys Δl=±1, ΔJ=0,±1 (J=0→0 forbidden), parity change. D-lines: 3p→3s ✓.	TODO
R3 · Dipole approx	k·a₀ for 589 nm ≈ 0.006 ≪ 0.1 → dipole valid. ✓ (trivially).	PASS
R4 · Detailed balance	CR kernel at thermal rates must recover Boltzmann (n3p/n3s = g-ratio·e^(−E/kT)).	[RUN] PASS
R5 · Saha consistency	Ionization fraction from Saha must match gibbs.py Na⁺ at the same (T, n_e). Cross-method.	[RUN] PASS — agree ×1.2–5.4, converging to ×1.2 at high T (where the two defs coincide).
R6 · Line-center ceiling	At high τ the D-core saturates to B_λ at line center (can't exceed blackbody THERE; the elevation is in the wings/escape, not the saturated core).	TODO
R7 · Conservation	Σn_i conserved across the CR solve; mass conserved across speciation.	[RUN] PASS

Changelog (newest first)

rev 12 · 2026-06-05T19:30 · SCALING-LADDER EXPLORABLE (scaling.html, live). Operator reframe: it's an N branch AND an N² branch AND higher/steeper scalings, all coexisting — the device is the ENVELOPE. Pulled the PRIMARY literature (LITERATURE-CANON.md): Dicke 1954, Gross-Haroche 1982 (τ_R = 8π·τ_sp/(3·n·L·λ²)), Rajabi-Houde 2016, Ketterle/Cornell-Wieman Nobel, de Groot (Na₂ continuum), Molisch-Oehry. Built the PER-CELL scaling gate [RUN]: N² survives ONLY if τ_R<min(T₂_coll,T₂_Doppler) AND Arecchi-Courtens τ_E=L/c<τ_R. Default N (Rajabi-Houde: no SR in a thermalized gas). The device's INVERSION CLINE earns partial cooperative (slope 1.25); sub-λ dense pockets reach N² (slope 2.0); the Na₂ dimer sink (∝n², D₀=0.747 eV) rolls the slope DOWN at high n (the de Groot depletion). The local log-log slope of I(n_Na) IS the regime number — the device walks the ladder, the slope tells you the rung. Superradiance is GATED on real coherence conditions, not a fudge. Confirmed: the ~670 Row-108 shoulder = Na₂ A-X bound-free (not iodine).

rev 11 · 2026-06-05T18:20 · THE DEVICE THESIS (W0) MADE THE CENTERPIECE. Operator: the 589-trapped-feeds-819-escape asymmetry "has been my whole point" — and it had been buried in the rate matrix as g589 vs g819 parameters instead of named as the thesis. Built funnel.mjs [RUN]: deepening the 589 trap t_589 10→6000 builds the 3p reservoir 150×, fills 3d via the collisional funnel, and the 819 escaping power climbs ×160 (2.2e4→3.5e6) while trapped-589 escape stays flat; 819/589 escape ratio 0.00→0.30. Promoted to W0 at the top of the doc + the sitrep. HONEST: every smuggle (S=B, single-T, scalar θ, Holstein-linear) buried this; each correction cleared a layer off it — it was the first fact, not a discovery.

rev 10 · 2026-06-05T18:00 · HOLSTEIN LAYER 1 DEEP-READ → WING FIX. Read hyperclaude.cc/gifts/holstein-cretin/{8,1}. Layer 8 = engineering calc (cross-checks: t_589=10³/t_819=10, Na 3p ~8.8e-4 matches my f3p). Layer 1 = the real g₀(k₀L): Doppler-slab 12.9/177/2350 at k₀L 10/100/1000, Lorentz-slab ~49 at 1000. KEY: "the lineshape is the load-bearing parameter, not the strength" — Lorentz wings $1/x^2$ escape, Doppler $e^{-x^2}$ don't; Zanstra/Irons $x_\text{esc}=\sqrt{k_0L}$. Applied it [RUN]: rad2d v2 with the Lorentz-wing escape lifts the unbleached band-avg from 1.1× → 8.3× (their ~6×, same regime — the wing carries the escape, structurally right; not fine-tuned to their number). Center stays self-reversed ~1×, ceiling 47.8×. The wing fix is a real physics improvement from the deep-read.

rev 9 · 2026-06-05T17:40 · 2-D ESCAPE KERNEL BUILT (rad2d.mjs). First increment retiring the 0-D θ: a diametric-chord formal-RTE solve across the radial gradient (hot core → cool dense NaCl-boil skin), with vapor-fill n(T) (cool skin denser — the lid) and a nonlinear bleach lever on the cold cells. GATE PASS [RUN]: Planck ceiling B(2300)/B(1686) at 589 nm = 47.8× = their 48× exactly. STRUCTURE PASS: the bleach lever moves the line-center the right way (pinned at B_skin when shut → self-reversal; lifts toward the core when opened). MAGNITUDE GAP, FLAGGED not laundered: my unbleached band-avg ~1.1× vs their ~6×, open window ~10× vs 48× — my Voigt wing + k-scale are crude; the exact endpoints need calibration vs their rad2d.zig. Reported ONLY the ceiling as reproduced.

rev 8 · 2026-06-05T17:20 · 0-D→2D ESCAPE (sodium_package reasoning chain reviewed). The operator gave the FULL provenance chain behind sodium_package — a sequence of corrections that supersede my single-cell model. Headline (W7): collapsing radiation transport to one scalar θ was "the worst simplification" — it already produced a FALSE claim ("overdrive transient"). The floor is 2-D: 1-D flame (z) + radial escape (r). Overdrive is STEADY — a hot core seen through a bleached nonlinear window (S=B per zone, but you see the hotter interior); quantified 48× ceiling / 6.4× unbleached / ~8× bleaching gain. Holstein g₀ is the LINEAR theory, VOID at 24–37% excited fraction (B4 rewritten). Inversion is measure-zero — diagnose with b_i, never "did it invert." My 0-D cell has the SAME θ error; the 2-D nonlinear-window transfer is now the headline next build.

rev 7 · 2026-06-05T17:00 · LaTeX RENDERING (KaTeX) + sodium_package yoink doc. Added KaTeX (CDN, auto-render, no build step — no-middleware doctrine) to LOAD-BEARING.html + index.html; converted the load-bearing equations from plain-text code to typeset $…$ ($S=j_\lambda/\kappa_\lambda$, $\bar n \approx \bar n_\text{LTE}(1+g_0 f_\text{trap})$, $P_\text{chem}>C_\text{vib}k_\text{VT}\Delta T$, $b_i=n_i/n_i^\text{LTE}$). Wrote SODIUM-PACKAGE-YOINK.md: the merge target is my pipeline + their 2-D self-reversal lever + b_i reporting + the Saha-detailed-balance gate.

rev 6 · 2026-06-05T16:45 · CROSS-VALIDATED vs sodium_package (claude.ai). Reviewed the independent sodium_package engine (se_engine.py + rad2d.zig). Cross-checked [RUN] my cascade against it: b_3d ≈ 272 (theirs 320), n3d/n3p = 0.003 (under the 1.667 gain threshold → never inverted, matching). Their named smuggles (S=B(Tₑ), one-T, Rosseland, Holstein-as-scalar) = the same corrections I made independently → strong mutual validation. To yoink (queued): (1) departure coefficients b_i parameterization; (2) the Saha-detailed-balance correctness gate (dense limit MUST reproduce Saha — a built-in proof the solver isn't Boltzmann in disguise); (3) the 2-D self-reversal lever (ceiling 48× skin-graybody from the 2300/1686 K Planck ratio at 589 nm; unbleached delivers ~6× wings-only; bleaching the skin recovers ~8× band brightness). NaCl boil they cite 1686 K (I had 1738) — reconcile.

rev 5 · 2026-06-05T16:25 · NaCl TRANSPORT REGION (W6). Built transport_nacl.mjs: NaCl delivery enhanced over naive monomer vapor pressure via 4 channels (monomer/dimer/hydrolysis/aerosol+wick), all K's from nasa_gas.yaml thermo. Caught my own laundering twice: (1) the aerosol prefactor was an invented ×5 dominating the result → flagged as placeholder, separated enhancement_anchored (×1.3–1.9, real) from enhancement (×38–49, geometry-dependent, not yet defensible); (2) my first CRC monomer-P Antoine constants were ~6 OOM wrong → re-derived. Finding: the water-mediation is NOT a wall-flux boost (K_hyd·pMono·p_H2O is tiny) — it's a COUPLED gas-phase equilibrium shift (gibbs dry→wet: free Na 23%→31%). The "fumes even at low temp" dimer effect is real but a factor, not OOM.

rev 4 · 2026-06-05T16:00 · IONIZATION BOUNDARY FOUND (Saha + continuum). Built saha.mjs (Na E_ion=5.139 eV) — GATE R5 PASS, cross-checks gibbs ionization ×1.2–5.4 (→×1.2 at high T). Built continuum.mjs: P_Dline/P_total vs T, the bound→continuum split (free-free + free-bound vs the trapped line). Caught + fixed TWO physics bugs by re-derivation (GPD discipline): (1) neutral pool must DEPLETE with ionization before the cascade (limit check: ion→1 ⇒ line→0); (2) trapping g₀ must WEAKEN as neutral absorbers ionize away (g₀_eff=g₀·(1−ion_frac)). Then EXTENDED the sweep until the crossover appeared rather than asserting it: P_line/P_tot = 1.000 (3 kK) → 0.47 (16 kK) → 0.06 (30 kK). The device is LINE-dominated at flame T; continuum only in the >16 kK pulse core. W4 quantified.

rev 3 · 2026-06-05T15:40 · PIPELINE + INTERACTIVE LIVE. Built the full single-cell chain (pipeline.mjs): gibbs_table.json (42-row pure-O₂ speciation, the Wittgenstein ladder — hot path interpolates, no Python at runtime) → composition → cascade+CR(n̄) → T_exc, conservation PASS [RUN]. n̄ now DERIVED from trapping $\bar n \approx \bar n_\text{LTE}\,(1+g_0\,f_\text{trap})$. Built the interactive explorable (index.html): inline wasm(b64)+table, sliders drive the live 4-level waterfall + the T_exc(g₀) sweep + the flagged assumptions tail. Verified the inlined wasm instantiates+runs (T_exc 7668 K). LIVE at plasmagicians.com/opencretin/.

rev 2 · 2026-06-05T15:20 · TRAPPED-FIELD VERIFIED (n̄ formulation). Caught + fixed a ~16-OOM dimensional error: the $B_{01}\bar J$ raw-cgs form saturated 3p to g1/g0 at J̄/B_nu=0.5 (unphysical, T_exc=2e8 K). Re-derived per GPD rubric (trace dimensions): the field enters as the dimensionless photon occupation n̄, stim/spont = n̄. Rewrote set_field(nbar). Now all 3 checks pass [RUN]: n̄=0→chemistry; n̄=n̄_LTE→Boltzmann ×1.05; n̄≈g₀·n̄_LTE→T_exc 7663 K ×253 at 1 atm. The elevation is a SURFACE in n̄, resolved by the cascade — W3 confirmed.

rev 1 · 2026-06-05T15:05 · TRAPPED-FIELD TERM + W3 CORRECTION. Added the 3p-elevation term to cascade.zig. CORRECTED W3 per operator: the 3p elevation is NOT a pressure gate (0-D collapse the GPD handoff warns against) — it's the output of the mode-occupancy waterfall cascade. (Superseded by rev 2: the term's dimensions were wrong; n̄ is the fix.)

rev 0 · 2026-06-05T14:45 · SCAFFOLD. Imported verified engine (gibbs.py, cr.wasm, cascade.wasm) from ~/cc/radiative. Wrote this LOAD-BEARING.html. Catalogued 5 weird + 4 load-bearing + 5 shortcuts + 7 AMO gates. Yoinked GPD epistemic layer (status-fails-closed, re-derivation-not-grep, [READ]/[RUN]/[INFERRED] provenance, core/sheath + ionization-boundary frame). No physics changed yet.