JUSTIFIED BOND MODEL (2026-03-27 FINAL)
Mean 8.1%, 20/24 under 10%, ZERO free parameters

6 DERIVED rules (all from d=3 geometry):
  1. C_BOND = pi/d^2                 [A1g of T1u x T1u]
  2. W_PI = cos(pi/d) per pi bond    [Born rule cos^2(45 deg)]
  3. LP -= (d^2+1)/d^3*(2/n)^2       [Hessian mode well filling]
     Period-3 boost: d/(d-1)          [transverse fraction]
  4. s-block *= (2d-1)/d^2 = 5/9     [Li2+Cl2 both = 0.5556]
  5. radical *= (2d-1)/(2d) = 5/6    [Hessian decay ratio]
  6. ionic += 1/(2d+1)*dIE           [exchange path counting]
     Threshold 10% asymmetry         [empirical, flagged]

HONEST OUTLIERS (unjustified fixes removed):
  NaCl: -39.8% (needs SU(2) ionic tier, not yet derived)
  CO: -19.4% (needs sp hybridization or het-triple, not yet derived)
  PH: +13.1% (radical+ionic interaction, not yet understood)

UNJUSTIFIED rules REMOVED:
  - s+p ionic tier (C vs C/4): hand-waving, not derived
  - sp hybridization boost: constants are d=3 but trigger is chemical
  These improved NaCl and CO but we cannot defend them mathematically.
