Add missing substrate_cids column to experiments table

database/migrations/003_substrate_cids_bootstrap_safety.sql

@@ -0,0 +1,41 @@
+-- Bootstrap-order safety net for substrate_cids.
+--
+-- Problem this records: migration 002_substrate_cids.sql adds the
+-- substrate_cids column + GIN index, but the migration runner is
+-- invoked AFTER ``PgExperimentStore.init_schema()`` in
+-- ``database/neuron.py``. ``init_schema`` runs SCHEMA_INDEX_DDL from
+-- ``shared/pg_schema.py``, which contains
+--     CREATE INDEX IF NOT EXISTS idx_substrate_cids
+--         ON experiments USING GIN(substrate_cids);
+-- On legacy deployments that booted before TEN-240 the column does not
+-- yet exist when SCHEMA_INDEX_DDL runs, so init_schema crashes with
+-- ``UndefinedColumnError: column "substrate_cids" does not exist``
+-- before migration 002 ever gets a chance to apply.
+--
+-- The crash is fixed in ``shared/pg_schema.py`` by an idempotent
+-- ``ALTER TABLE ... ADD COLUMN IF NOT EXISTS substrate_cids`` placed in
+-- SCHEMA_TABLE_DDL (executed before SCHEMA_INDEX_DDL within the same
+-- ``init_schema`` call). This file is the migration counterpart of that
+-- fix: an idempotent re-assertion that the column + index exist, so any
+-- replica restored from a snapshot taken between TEN-240 and this fix
+-- catches up without manual intervention.
+--
+-- Same table-existence guard as the other migrations: on a fresh deploy
+-- the migration runner runs against an empty schema (CI's "Migrations
+-- apply to empty DB" job), so the body must be a no-op when the
+-- experiments table doesn't exist yet.
+
+DO $$
+BEGIN
+    IF EXISTS (
+        SELECT 1
+        FROM information_schema.tables
+        WHERE table_schema = 'CURRENT_SCHEMA'
+          AND table_name = 'experiments'
+    ) THEN
+        ALTER TABLE CURRENT_SCHEMA.experiments
+            ADD COLUMN IF NOT EXISTS substrate_cids JSONB NOT NULL DEFAULT '[]';
+        CREATE INDEX IF NOT EXISTS idx_substrate_cids
+            ON CURRENT_SCHEMA.experiments USING GIN (substrate_cids);
+    END IF;
+END $$;

shared/pg_schema.py

@@ -41,6 +41,11 @@
 -- ``shared.challenge.generate_challenge`` and can exceed INT32's 2.1B max.
 -- Widen existing deployments to BIGINT; no-op if already BIGINT.
 ALTER TABLE experiments ALTER COLUMN round_id TYPE BIGINT;
+-- Migration (TEN-240): substrate_cids was added to the CREATE TABLE above
+-- after deployments already had an experiments table, so the IF NOT EXISTS
+-- table-create is a no-op and the column is missing. Add it explicitly.
+ALTER TABLE experiments
+    ADD COLUMN IF NOT EXISTS substrate_cids JSONB NOT NULL DEFAULT '[]';
 """
 
 SCHEMA_INDEX_DDL = """