docs: FABRIC-4 §3 as a null hypothesis

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/fabric/FABRIC-4.md

@@ -126,6 +126,29 @@ Stadium, `FABRIC-0.md`); K conserved, not sampled.
 
 **The question he will answer:** is there any advantage to this over a conventional OS?
 
+**As a null hypothesis** (phrased at his request, 2026-10-16; a starting form, to be sharpened
+before any experiment):
+
+> **H0:** Delivering information to a computing element as *state* -- a fact that persists where
+> the element will take it, when it has the means to -- confers **no advantage** over delivering
+> it as a *moment* -- an event the element must act on at the instant it arrives, with whatever
+> resources it has then -- as a conventional OS does with signals and interrupts.
+>
+> **H1:** Delivery as state gives a measurable advantage on at least one of the measures below,
+> at a cost in the others that does not outweigh it.
+
+Measures that would decide it, each the same for both forms on the same workload:
+1. **Loss:** events delivered that never took effect (believed but not acted on, coalesced,
+   dropped).
+2. **Corruption under pressure:** faults, and state lost to them, when the receiver is short of the
+   resources the event needs (here: stack depth; in general memory, time).
+3. **Variance:** the coefficient of variation of outcomes over repeated runs, and agreement of
+   state hashes across machines.
+4. **Cost:** latency from delivery to effect, and the memory held by facts waiting.
+
+H0 is rejected if state is better on 1, 2 or 3 beyond the measurement floor, with 4 within a bound
+fixed before the run. It stands if not. Either result is a result.
+
 **Contrasts noted in the conversation, as starting points, not conclusions:**
 - A conventional OS delivers much information as moments: Unix signals run a handler at whatever
   instant they land, on the interrupted code's stack, may call only async-signal-safe functions,