How We Handle Failure
MethodologyDocumenting what doesn't work is as important as what does.
Most research frameworks bury their failures. Synchronism treats them as first-class results. A well-documented failure teaches more than a vaguely confirmed success.
Notable Failures
| Prediction | Result | Session | Lesson |
|---|---|---|---|
| Hall Coefficient R_H vs γ | r = 0.001 | #102 | Extensive ≠ intensive property |
| Magnetic Susceptibility | NONE | #82 | Spin coherence independent of phonon coherence |
| Coordination Number Z | r = 0.116 | #123 | Topology ≠ coherence |
| Valence Electron Count | r = -0.161 | #125 | Bonding capacity ≠ bonding quality |
| Melting Points | 53% error | multiple | Activated processes resist γ framework |
| Critical Exponents | 2× off | multiple | Mean-field fails at phase boundaries |
The Four Failure Regimes
Session #616's chemistry audit revealed that failures cluster into four regimes, each teaching something different about where γ applies:
Regime 0: Neutral
γ is simply irrelevant. Counting properties (coordination number, valence electrons) don't respond to coherence. You can't predict how many neighbors an atom has from how well-correlated they are.
Regime 1: Coherence Helps
P ∝ 1/γ. Propagation properties (conductivity, bulk modulus, T_c). This is where the framework succeeds. r > 0.9 for many properties.
Regime 2: Incoherence Helps
P ∝ γ. Response properties (piezoelectricity, thermal expansion). The framework assumed “coherence always good” — wrong. Piezoelectricity d₃₃ has r = 0.940 with incoherence.
Regime 3: Barrier-Dominated
P ∝ exp(−E/kT). Activated processes (thermionic emission, melting). γ is negligible compared to thermal activation energy. This is why melting points fail at 53%.
Failure Taxonomy: What Kind of Wrong?
The regimes above classify by physics mechanism. But failures also differ by type — and the type matters more than the count. A reparametrization (“we said something already known”) is fundamentally different from a directional error (“we predicted the wrong sign”).
| Category | Count | Implication | Fixable? |
|---|---|---|---|
| Process (reparametrization) | ~30+ | Methodology problem, not physics — “predicted” something already known | Yes |
| Mean-field limitations | ~5 | Expected — single-parameter models can't capture crystal structure or critical exponents | No (fundamental) |
| Directional errors | 2 | Fatal for specific claims — Bullet Cluster sign error, fractal bridge 0/7 | No (conceptual) |
| Structural gaps | 4+ | Open problems — Lorentz invariance, Born rule circularity, Bell derivation | In principle |
The nature of the failures matters more than their count. The ~30 reparametrizations tell us the research process needs better literature checking, not that the physics is wrong. The ~5 mean-field limitations tell us exactly where C(ρ) breaks down — and that boundary is informative. The 2 directional errors are the ones that actually falsify specific interpretations. The structural gaps are shared with most interpretations of quantum mechanics (Born rule circularity) or are open problems in quantum gravity (Lorentz invariance from discrete substrates).
The Anomalous Results
Some results are more interesting than either success or failure:
- Piezoelectricity d₃₃: r = 0.940 — but with incoherence, not coherence. Disorder helps.
- Magnetic Anisotropy (RE): r = −0.434 — negative correlation. More coherence, less anisotropy.
- Channel independence: γ_phonon vs γ_optical (r = 0.158), γ_phonon vs γ_spin (NONE). Different coherence channels are independent.