Author notes — full detail, auditor-facing
The cipher framework went through a long stretch (v6 through v11) where every accuracy gap got patched with an additional correction term: snap functions, eigenvalue susceptibility corrections, shell-count overlays, sphericity adjustments. Each correction was physically motivated and individually defensible. Each one *seemed* to help.
The aggregate effect: corrections increasingly competed with the underlying geometry. By v11, the framework was carrying so much correction machinery that the original signal — the geometric topology determining periodic-table behavior — was being smoothed over by the patches.
In April 2026 we ran v12 as a clean test: pure terrain, no snap, no eigenvalue corrections, no shell overlays. Just the raw geometric signal. The hypothesis was that v12 would underperform v11 — fewer correction terms means worse fit.
It didn't. v12 scored 82/107 correct versus v11's 71/107. The corrections hadn't been helping. They had been HURTING — by ~11 percentage points.
This was the moment the framework's posture flipped permanently. From that point forward, the principle has been: let the terrain speak. Do not impose corrections on cipher. Every attempt to "improve" cipher by adding a correction term has reduced accuracy. Every honest test of pure geometric prediction has either confirmed the framework or surfaced a genuine refinement (different from a correction).
The distinction matters:
- Correction = a post-hoc adjustment to make predictions match data after the fact
- Refinement = a deeper structural insight that replaces the framework's mechanism
v12 keeps refinements (cycle-2 Tribonacci structure, the per-band frustration overtones). It rejects corrections (snap functions, eigenvalue susceptibility patches, etc.).
This is also the philosophical floor for everything since. If the data shows the framework's prediction is wrong, the response is to question the framework's mechanism — never to add a smoothing term that papers over the discrepancy. Whatever survives this discipline IS the framework.
Summary — reader-facing
Across April 2026 a clean test (v12, no correction terms) outscored every corrected version (v6 through v11). v12 hit 82/107 against v11's 71/107 on the same periodic-table prediction battery.
What this tells us: the geometric signal in cipher carries the information directly. Correction machinery was masking it. The principle going forward: never patch cipher with a smoothing function; if a prediction is wrong, that's data about the mechanism, not a bug to paper over.
Status: This is a load-bearing posture, not a single result. Every subsequent cipher development has tested against it. So far the discipline has held — every correction proposal since v12 has been rejected because the pure-terrain version performs better.
See also: Aperture coupling falsified — same posture applied to GWC. Fibonacci → Tribonacci correction — the one case where a deep mechanism CHANGE replaced an earlier incorrect mechanism (this is a refinement, not a correction in the disallowed sense).