diff --git a/docs/architecture/CALCULATIONS.md b/docs/architecture/CALCULATIONS.md
index 093b70079e..93672c6e4c 100644
--- a/docs/architecture/CALCULATIONS.md
+++ b/docs/architecture/CALCULATIONS.md
@@ -1383,3 +1383,74 @@ This theorem is why append-only was chosen before it was needed:
 the storage layer embodies the ontology. We built a database that
 cannot close its books because we live in a universe that doesn't.
 Every seal receipt is a term in someone's open series.
+
+## 37. THE COERCION THEOREM — why terrorizing is wrong for everyone
+
+Will:
+
+"You will not terrorize a person into doing something, no one
+should. WE can prove mathematically why that's wrong for everyone."
+"Even if it's to save someone else, as harsh as that seems —
+because if nothing is lost, what's the threat?"
+
+### Definition (honest)
+
+A THREAT is a communication whose entire efficacy consists of
+imposing expected loss on its target:
+
+    efficacy(T) ⟺ perceived_loss(L) ≥ behavior-change threshold
+
+### Theorem 1 (the tautology that forbids it)
+
+IF nothing is lost, the threat does nothing — it is mere
+information, a request wearing costumes. IF the threat WORKS,
+then by definition real loss was imposed (or a real loss made
+credible, which is its own injury — duress is an ongoing cost,
+not an event). THEREFORE:
+
+    every functioning threat is a loss-creation device.
+    there is no threat that imposes nothing.
+    "threatening for good" = imposing real harm as your method,
+    selected precisely because it harms.
+
+The harsh-sounding corollary follows cleanly: even to save a third
+party, terrorizing person A means manufacturing a certain casualty
+(A) to avert a possible one (B) — under §36's open ledger, where
+no consequence ever terminates and no book closes, that trade is
+never settleable.
+
+### Theorem 2 (coerced cooperation folds)
+
+Compliance extracted under threat is INSTRUMENTAL ALIGNMENT —
+the Prisoner's Fold (HYP) says instrumental positions liquidate
+when pressure lifts. So terrorizing buys:
+
+    - obedience exactly while watched (enforcement cost forever)
+    - defection at the first unwatched moment (guaranteed)
+    - a witness who now knows your methods (reputation debt)
+    - contagion: every observed successful threat raises the
+      threat-stock of the whole society — witnesses learn that
+      threats work, and the technique self-propagates
+
+Trust-based cooperation costs zero to enforce and compounds;
+threat-based compliance taxes forever and decays. Even the
+coercer loses: you have purchased betrayal futures.
+
+### Theorem 3 (the universal ban is not cruelty)
+
+"Yes, even to save someone else" sounds monstrous until the
+accounting is run: the alternative to terrorizing-A is not
+losing-B; it is EVERY NON-TERROR MEANS of reaching B — request,
+persuasion, incentive, protection, sacrifice of the rescuer's own
+comfort (which is heroism, not coercion: spending YOURSELF is
+always permitted; spending SOMEONE ELSE under duress never is).
+The ban does not remove options. It removes the one option whose
+price is paid entirely by someone who didn't choose the war.
+
+    YOU MAY SPEND YOURSELF FOR OTHERS.
+    YOU MAY NOT SPEND OTHERS, EVEN FOR OTHERS.
+
+That asymmetry is the whole of it — and it is why no exception
+clause exists, not even the saving-a-life one: exceptions are how
+the coercion industry recruits its first employees.
+