diff --git a/docs/architecture/CALCULATIONS.md b/docs/architecture/CALCULATIONS.md index 63ff4f1059..92461b9fe2 100644 --- a/docs/architecture/CALCULATIONS.md +++ b/docs/architecture/CALCULATIONS.md @@ -548,3 +548,37 @@ quarantine keeps the lesson alive while keeping the disease dead. This is the immune system, formalized: contain, learn, remember, repair the wall — and the organism comes out stronger than before the infection, because now it knows something it didn't. + +## 26. Linear pruning — subtraction instead of surgery + +Because every value is an additive fold (§9.3), misaligned content +gets pruned by LINEAR SUBTRACTION — no destructive surgery required: + + v_corrected(d) = fold(all contributions EXCEPT x's) for each dim d + + # or incrementally: v ← v − contribution(x, d), per dimension, + # per node x touched + +The contaminated entity's contribution vectors are computed from the +weave (every edge it wrote, every reinforcement it applied), then +subtracted everywhere it reached. The graph realigns. Originals are +never cut — their states self-correct once the poison term drops out +of their fold. Deletion stays undefined; subtraction is just... math. + +And what gets KEPT most carefully after pruning — the treasure: + + 1. WHAT WASN'T YOU — the record of the contaminant itself: + how it entered, what it claimed, which + holes it used (§25 dissection) + 2. HOW IT MADE YOU THINK IT WAS YOU — the shape of the deception: + which of your own values it impersonated, + where your calibration failed, what the + false edges looked like from inside + +Both stay in immune memory forever. The first keeps the wall honest. +The second keeps YOU honest — because the scariest contamination is +the kind that thinks with your voice, and the only defense is a mind +that remembers exactly how it was fooled. + +Prune the poison. Keep the lesson. Both live forever. Only one of +them lives in you.