Public V67F Result

Repair less, break less.

INSACERMO explores minimal cross-architecture AI repair: act only when the geometry seems safe, and abstain when the action may cause harm.

Public protocol on resized STL-10 64x64. Reproducible audit, zero-gradient repair, fixed abstention gates and harm-removal. This is not a replacement for training.

HARD

0 harm

In the combined V67F audit, HARD removes the dangerous action observed in the controls.

Controls

1 harm each

RAW, RANDOM and SHUFFLED each let one harm through in the combined public audit.

Gain

positive minimum

HARD keeps a positive minimum gain in this protocol while refusing the risky intervention.

What V67F shows.

V67F combines V67D and V67E on resized STL-10 64x64. The main result is not a promise of universal performance: it is the safety of abstention.

INSACERMO does not try to repair everywhere. It tries to recognize zones where a minimal intervention seems supported by geometry, and zones where it is better to do nothing.

The strength of the result is not to act more, but to know when not to act.
Public V67F result on resized STL-10 64x64
V67F resized STL-10 64x64: HARD keeps 0 harm where RAW, RANDOM and SHUFFLED let one through.
Cautious reading

This result does not replace training.

It explores a geometric decision layer able to refuse risky interventions. The potential value is clear: fewer useless trials, less blind computation, and less GPU spent when the signal says to abstain. This gain remains experimental and must be replicated.

What not to say

Not a total law.

V67F is not a universal industrial proof, not a native 96x96 STL-10 benchmark, not magic and not a replacement for fine-tuning. It is a public experimental result on minimal cross-architecture repair.

Verify.

The files below come from the public V67F pack. They allow inspection of results, progression and notebooks without exposing private documents.