Consolidate migrations 001-080 into a checkpoint baseline

.claude/ways/kg/schema/way.md

@@ -11,16 +11,21 @@ scope: agent, subagent
 
 ```
 schema/
-├── 00_baseline.sql      # Core schema (AGE, auth, API tables)
-└── migrations/          # Sequential migrations (NNN_name.sql)
+├── 00_baseline.sql      # Checkpoint baseline (migrations 001-080 consolidated)
+├── migrations/          # Post-checkpoint migrations (081+)
+│   └── archived/        # 001-080, superseded by baseline — never edit
+├── migrations-warm/     # Migrations needing a running graph engine
+└── scripts/checkpoint.sh  # Regenerate the checkpoint (replay + dump + verify)
 ```
 
 ## Creating Migrations
 
-1. Create `schema/migrations/NNN_descriptive_name.sql`
+1. Create `schema/migrations/NNN_descriptive_name.sql` (next number after 080/latest)
 2. Use next sequential number
 3. Migrations auto-apply on `./operator.sh restart postgres`
 
+`00_baseline.sql` is generated by `checkpoint.sh` — regenerate it, don't hand-edit.
+
 ## Query Language
 
 **Apache AGE uses openCypher** (not Neo4j Cypher):

docs/architecture/database-schema/ADR-210-database-schema-migrations.md

@@ -275,6 +275,19 @@ COMMIT;
 - Merge all migrations into new baseline: `schema/migrations/stable_v2_baseline.sql`
 - Archive old migrations in `schema/migrations/archived/`
 
+> **Executed 2026-07-01 (checkpoint v080):** migrations 001–080 were
+> consolidated into `schema/00_baseline.sql` and moved to
+> `schema/migrations/archived/`. The consolidation is generated, not
+> hand-merged: `schema/scripts/checkpoint.sh` replays the old baseline plus
+> every migration into a throwaway Apache AGE container via raw psql, dumps
+> the resting state (relational DDL + seed data via pg_dump; graph-seeding
+> cypher migrations carried verbatim, since AGE graphs cannot round-trip
+> through pg_dump), and verifies the candidate by building a second container
+> from it alone and diffing schema, AGE labels, row counts, and normalized
+> data against the replayed original. The baseline records versions 1–80 in
+> `schema_migrations`; the runner scans `archived/` before `migrations/`, so
+> pre-checkpoint databases keep their incremental upgrade path.
+
 ---
 
 ## Consequences

operator/database/migrate-db.sh

@@ -177,8 +177,12 @@ if [ ! -d "$MIGRATIONS_DIR" ]; then
     exit 0
 fi
 
-# Scan for migration files
-for migration_file in "$MIGRATIONS_DIR"/*.sql; do
+# Scan for migration files. archived/ holds migrations consolidated into the
+# checkpoint baseline (schema/00_baseline.sql): fresh databases have all of
+# them recorded by the baseline and skip them; databases created before the
+# checkpoint still find their pending versions there. Scanned first so
+# archived versions sort ahead of post-checkpoint migrations.
+for migration_file in "$MIGRATIONS_DIR"/archived/*.sql "$MIGRATIONS_DIR"/*.sql; do
     [ -f "$migration_file" ] || continue
 
     # Extract version from filename (001_baseline.sql → 001)
@@ -292,9 +296,11 @@ for pending in "${PENDING_MIGRATIONS[@]}"; do
         echo -e "${GRAY}----------------------------------------${NC}"
     fi
 
-    # Apply migration (capture output to detect errors)
-    MIGRATION_OUTPUT=$(run_psql_file "$FILE" 2>&1)
-    MIGRATION_EXIT_CODE=$?
+    # Apply migration (capture output to detect errors). The || arm is
+    # required: under set -e a bare command-substitution assignment aborts
+    # the whole script on psql failure, making the skip-and-retry branch
+    # below unreachable.
+    MIGRATION_OUTPUT=$(run_psql_file "$FILE" 2>&1) && MIGRATION_EXIT_CODE=0 || MIGRATION_EXIT_CODE=$?
 
     # Show output in verbose mode
     if [ "$VERBOSE" = true ]; then