diff --git a/docs/training/corpus/Layer_0--Primitive_Facts/CORPUS-0007-information-arrives-unevenly.md b/docs/training/corpus/Layer_0--Primitive_Facts/CORPUS-0007-information-arrives-unevenly.md
new file mode 100644
index 0000000..413e13a
--- /dev/null
+++ b/docs/training/corpus/Layer_0--Primitive_Facts/CORPUS-0007-information-arrives-unevenly.md
@@ -0,0 +1,106 @@
+# CORPUS-0007
+## Information Arrives Unevenly
+### Status: Training Corpus Seed
+### Layer: Layer_0--Primitive_Facts
+### Purpose: Teach that actors do not receive the same facts at the same time
+### Repository Path: docs/training/corpus/Layer_0--Primitive_Facts/CORPUS-0007-information-arrives-unevenly.md
+
+---
+
+## 0. Principle
+
+Information does not arrive everywhere at once.
+
+A trader, buyer, porter, clerk, and official may all know different parts of the same event.
+
+The actor who hears useful information earlier may act before prices, queues, or expectations adjust.
+
+---
+
+## 1. Roman-Visible Example
+
+A timber barge is delayed on the river.
+
+The towmen know first.
+
+A dock worker hears next.
+
+A trader with a riverfront contact hears before the market.
+
+A carpenter across town hears later.
+
+A buyer in Capua may hear much later.
+
+The event is one event.  
+Knowledge of the event spreads unevenly.
+
+---
+
+## 2. Minimal Information Structure
+
+Information timing depends on at least:
+
+1. where the event occurred
+2. who saw it
+3. who can carry the report
+4. who trusts the source
+5. how far the information must travel
+6. whether anyone benefits from delay or concealment
+7. whether visible signals confirm the report
+
+Information has a route just as goods do.
+
+---
+
+## 3. Incorrect Modern Assumption
+
+Incorrect assumption:
+
+> Once something happens, everyone relevant knows it.
+
+Correction:
+
+> Events occur before they are widely known. Different actors learn at different times and with different confidence.
+
+---
+
+## 4. Simulation Use
+
+This principle supports:
+
+- rumor systems
+- delayed price reactions
+- asymmetric opportunity
+- messenger value
+- local knowledge advantage
+- stale reports
+- hidden scenario states
+- actor-specific perception
+
+The model should not give every actor perfect information when an event occurs.
+
+---
+
+## 5. Canonical Test
+
+If the model sees:
+
+> A fire damaged a workshop in Ostia.
+
+It must ask:
+
+- who saw the fire?
+- who has confirmed the damage?
+- who has only heard rumor?
+- when does Capua learn?
+- who benefits before the news spreads?
+- who still acts on old prices?
+- who may conceal or distort the report?
+
+Only then can the information effect be understood.
+
+---
+
+## 6. Success Condition
+
+If the model stops treating facts as instantly shared and starts tracking who knows what, when, and with what confidence, this file is functioning correctly.