# n238 — Destruction session V: three attacks on the glass, zero kills, one correction

**Date:** 2026-07-12 · **Directive: try to make the theory fall.** Three computable attacks aimed at the successor's load-bearing points. Full results, including the one that landed a hit.

## Attack 1 — the hidden-crystal accusation: FAILED (the strongest attack)

The charge: every glass result (n221–n230) used a *jittered diamond* — which secretly retains crystalline connectivity (every 0-site exactly 4-bonded via the lattice ordering). A **true amorphous bipartite graph** (uniform random points, random balanced coloring, 4-NN bonds, coordination fluctuations included — no lattice anywhere in the construction) could collapse the whole program. Result, three seeds at N = 500:

- polarization equality **0.9933 ± 0.0050 — better than the jittered crystal**;
- light propagates (c²_EM ≈ +6.5, all seeds);
- the cost curvatures stay aligned (both negative — the equal-cone family's precondition);
- the photon kernel stays topological (dim = E−N+1, up to disconnected-component corrections).

The glass results owed nothing to hidden crystal order. **The successor's key properties are properties of disordered bipartiteness itself.**

## Attack 2 — the χ pattern-axis artifact: A HIT (rule 5)

The charge: n230's χ = 0.15 used gravity pattern x²−y² and EM envelope x — *shared axis*, possibly manufacturing overlap. Verified: axis-independent combos give **χ = 0.080–0.097** vs 0.130 for the shared-axis combo on the same graph. The shared-axis value was inflated ~50%. **Correction: the ledger-bridge one-loop strength is χ ≈ 0.08–0.10 (axis-independent), not 0.15.** Consequences propagate benignly: the bridge is still real, still moderate, still ensemble-solid; the back-reaction estimate drops to χ² ≈ 1%; n231's DESI arithmetic *strengthens* (the through-bridge α-drift ratio drops, but the exclusion margin was 6 orders — it stays dead). n230 carries the correction header.

## Attack 3 — the executioner on the true glass: FAILED

The n220-class global κ scan — the test that killed the crystal at −0.098 — run on the *true* glass at the pinned member: **min κ = +0.021, both seeds, stable with margin.** The stability is not the jitter construction's gift either.

## Verdict of the session

The theory does not fall tonight. One number was corrected downward (χ), zero structures broke, and the most dangerous accusation available — that the glass program was secretly leaning on crystal order — was refuted by construction. Standing threats unchanged: the RG flow, Mercury-once-dictionaried, DESI, the review of n216–n238. The record notes what a real kill looked like twice before; tonight produced the other thing the method exists for: **a survivor whose survival was purchased at full price.**

**Class: attacks [A-model, three seeds each, declared scope]; the χ correction [rule 5, applied]; no rescue anywhere.**