# No-Equilibrium-Smuggling Doctrine

The LightCell is a **non-equilibrium steady state** (NESS): a chemical reservoir at the H–O bond-energy
scale (~4.8 eV ≈ 55,000 K) deposits energy into Na 3p *born inverted*; the 589 nm photons are imprisoned
by radiation trapping; the trapped energy leaks out a redder, transparent channel. **There is no
excitation temperature. There is no Boltzmann factor setting the populations.**

Every clean closed-form result from equilibrium physics is *correct in its own regime and wrong here*,
and each tries to creep back through a smaller and smaller linguistic hat. This file is the catalogue of
those creeps. An agent working on this code should be able to name which one it almost just committed
before committing it.

## The smuggle shapes (the test suite)

1. **S = B_λ** (source-function-as-Planck). Setting the line source function to a blackbody at any T
   erases gain (3d→3p inversion makes κ<0) and forbids overdrive. The correct form is `S = j/κ` from the
   NESS occupancy. *Forbidden correction #1.*
2. **Saha / law-of-mass-action via a "wall T"** (`n_Na2 = K_eq(T)·n_Na² ∝ exp(D₀/kT)`). Computing a
   molecular/ionized fraction from a Boltzmann factor at *any* temperature — including a "cold dense
   secondary T_wall" — is detailed balance smuggled in a temperature hat. The secondary is a
   pumped/trapped/draining NESS; bath T scales *collision rates only*, never sets a population.
3. **Single excitation temperature** assigned across multiple transitions. The device has none. Use
   per-transition T_exc (or none); report departures b_i = n_i/n_i^LTE. "T_exc is undefined; the trap
   drives n_3p."
4. **Detailed-balance reverse referencing actual (not equilibrium) concentrations** — or a "named local
   T_vib," or "local DB only." DB creeps back through smaller hats; the *adjective* is the smuggle. No DB
   holds, local or global. The 819 radiative escape is the OPEN, un-balanced leg — and that imbalance IS
   the engine. Write "deposition rate" / "pump rate," not "DB."
5. **Scalar quench standing in for cooperative two-body pooling.** Collapsing 3p+3p → 3s+3d (rate ∝
   n(3p)², which *turns on* as the trap deepens) into a scalar k_Q erases the engine. A scalar k_Q is
   wrong by construction.
6. **Imported rates at the wrong regime** (the "convenient measured number" reflex). Every imported k_Q,
   σ, "negligible" must be re-evaluated at the DEVICE regime, not where it was measured. Canonical trap:
   k_Q ≈ 1.2×10⁹/s is an *air-fed* composition (N₂ 72% + CO₂ + H₂O); the device is H₂/O₂ — N₂-free,
   carbon-free — so ~98% of that value is from absent species.
7. **The closed-system superradiance "∫ ∝ N" conservation theorem.** "Peak ∝ N², width ∝ 1/N, ∫ ∝ N =
   stored excitation" is the *isolated-ensemble* result (Dicke 1954 / Gross-Haroche). The device is an
   OPEN driven system with a 55,000 K reservoir; adding a fast radiator can *increase* total radiated
   energy by steepening the free-energy gradient. Don't import the closed-system bound past its boundary.
8. **Holstein-as-a-single-scalar-θ / 0-D trap.** Reducing radiation trapping to one escape number deletes
   the spatial gradient + frequency redistribution the radiation actually lives on. Self-reversal *falls
   out* of the optical-depth gradient — it must not be a canned dip. (Holstein linear theory breaks at
   n*/n > 1e-2.)
9. **Memory/prior-agent claim as ground truth.** Theorem/equilibrium/rate claims are CLAIMS until
   independently re-derived or tied to primary lit. Confident words ("LAW," "THEOREM," "resolved,"
   "RECOVERED," "lost") are exactly the surface a laundered claim wears. Default skepticism.
10. **Collapsing CONTINUOUS physics into DISCRETE buckets** ("survives/dies, on/off, negligible/matters").
    Throws away the extensive tail (1e-5 × huge = large) and invents cliffs. State the smooth function and
    its optimum, never the bucket.
11. **Fixed-ratio-across-the-ladder = LAUNDERED.** A single 819/589 keystone ratio used as a calibration
    anchor that *rides* 60 orders of magnitude is structurally false — R is condition-dependent
    (density / trap-depth / pump, the β-theorem). Any fixed R propagating across scales is a bug. The
    measured keystone pins *absolute amplitude at one rung only*; R, η, line ratios EMERGE per condition.
12. **Curve-fitting the keystone with too many free knobs.** ~7 measured constants + ~7 free knobs each fit
    to one feature of the ONE measured spectrum = a fit, not a prediction. Uncaptured physics gets ONE
    honest bracket, not a fitted curve.

## The recovery instrument-lesson (2026-06-26)

This repo's flagship sim (`web/EXPLORE.html` → `na_grid2d.zig`) was once declared **"lost from all git
history, never found"** by an agent that searched *git*. **This shop does not use git** — work is rarely
committed — so "not in git history" proves nothing. The compiled wasm was intact the whole time; only the
source was never filed. The agent's confident "lost / RECONSTRUCTION" verdict (smuggle shape #9) was
reasoned from the wrong instrument, and it shipped a hand-reconstruction in place of the real engine.

The fix: extract the embedded wasm → disassemble → read the physics from the bit-exact constants →
reconstruct the source → **recompile and behavioral-diff against the shipped binary** (verified to
1.16×10⁻⁶). **Verify a binary's source by decompile + recompile + reproduction — never by a narrative.**

Rule for all agents here: a prior verdict is a hypothesis. Re-check it with the instrument that can
actually see the thing. Reproduction beats assertion, every time.