diff --git a/migrations/031_create_network_metrics_timeseries.sql b/migrations/031_create_network_metrics_timeseries.sql
new file mode 100644
index 0000000..3d041af
--- /dev/null
+++ b/migrations/031_create_network_metrics_timeseries.sql
@@ -0,0 +1,27 @@
+-- Time-series schema for the Network Dashboard (issue #45).
+--
+-- One wide row per UTC calendar day, rather than narrow per-metric rows,
+-- since #15-#19's endpoints each want "give me the last N days of X" and a
+-- wide table lets a single row backfill/upsert cover every metric for that
+-- day without a multi-table join. `date` is the UTC calendar day the metrics
+-- were computed for (see #2/#3's day-boundary convention).
+CREATE TABLE IF NOT EXISTS network_daily_metrics (
+    date TEXT PRIMARY KEY, -- YYYY-MM-DD, UTC calendar day
+    daily_active_accounts INTEGER NOT NULL DEFAULT 0,
+    new_accounts INTEGER NOT NULL DEFAULT 0,
+    transaction_count INTEGER NOT NULL DEFAULT 0,
+    payment_volume_usd REAL NOT NULL DEFAULT 0,
+    avg_fee_stroops REAL NOT NULL DEFAULT 0,
+    median_fee_stroops REAL NOT NULL DEFAULT 0,
+    active_soroban_contracts INTEGER NOT NULL DEFAULT 0,
+    -- distinguishes "job ran and computed a real value" from "no row yet" for
+    -- #15's gap-vs-zero requirement; a present row with is_complete=0 means a
+    -- job run was interrupted partway through (see #2's partial-run note).
+    is_complete INTEGER NOT NULL DEFAULT 1,
+    created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP
+);
+
+-- date is already the primary key (and therefore indexed), but the explicit
+-- index documents intent and survives if the PK strategy ever changes.
+CREATE INDEX IF NOT EXISTS idx_network_daily_metrics_date ON network_daily_metrics(date DESC);