fix: make the migration chain runnable from an empty database

0001 called Base.metadata.create_all(), building the schema from the
current models, so on a fresh PostgreSQL 0002 collided on
datasets.raw_sha256 and `alembic upgrade head` could never run from zero.

0001 is now an explicit, frozen baseline: the seven base tables exactly as
the real history created them, before 0002/0006 extended them. Databases
already past 0001 never re-run it. Proof: a fresh `upgrade head` produces a
schema identical (pg_dump --schema-only) to the long-lived dev database.

0019 repairs a latent bug found by that comparison: 0006 used
server_default="now()" as a string, which PostgreSQL froze at the instant
the column was created, so every Agent and Job got the same created_at.
It also adds three indexes the models declare but no migration created.

central/tests/test_migration_baseline.py guards (via AST) against
migrations importing the models or using string timestamp defaults.

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

central/alembic/versions/0001_initial.py

@@ -1,11 +1,99 @@
 """initial hub persistence
 
 Revision ID: 0001
+
+Frozen, explicit baseline. This revision used to run
+``Base.metadata.create_all()``, which builds the schema from the *current*
+models: on an empty database it created tables/columns that 0002..0018 then
+tried to create again (first collision: ``datasets.raw_sha256`` in 0002), so
+``alembic upgrade head`` could never run from zero.
+
+This body reproduces exactly what 0001 created on the databases that went
+through the real history: the seven base tables, *before* 0002 added
+``datasets.raw_sha256`` and 0006 added the agent status/timestamp columns.
+Constraint and index names match the ones those databases already have
+(PostgreSQL defaults for PK/FK/UNIQUE, ``ix_<table>_<column>`` for indexes).
+
+Databases already at 0001 or later never re-run this revision, so changing
+its body only affects new installations. It must never import the ORM
+models again: a migration describes one point in history, not the present.
 """
 from alembic import op
-from central.app.models import Base
+import sqlalchemy as sa
+
 revision = "0001"
 down_revision = None
 branch_labels = depends_on = None
-def upgrade(): Base.metadata.create_all(op.get_bind())
-def downgrade(): Base.metadata.drop_all(op.get_bind())
+
+
+def upgrade():
+    op.create_table(
+        "tenants",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("name", sa.String(128), nullable=False, unique=True),
+    )
+    op.create_table(
+        "datasets",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("kind", sa.String(32), nullable=False),
+        sa.Column("provider", sa.String(128), nullable=False),
+        sa.Column("sha256", sa.String(64), nullable=False),
+        sa.Column("version_label", sa.String(256), nullable=False),
+        sa.Column("act", sa.Text(), nullable=False),
+        sa.Column("source_url", sa.Text(), nullable=False),
+        sa.Column("downloaded_at", sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False),
+        sa.Column("status", sa.String(32), nullable=False),
+        sa.Column("record_count", sa.Integer(), nullable=False),
+        sa.UniqueConstraint("provider", "sha256", name="uq_dataset_provider_hash"),
+    )
+    op.create_index("ix_datasets_kind", "datasets", ["kind"])
+    op.create_index("ix_datasets_status", "datasets", ["status"])
+    op.create_table(
+        "ncm_entries",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("dataset_id", sa.Integer(), sa.ForeignKey("datasets.id"), nullable=False),
+        sa.Column("code", sa.String(16), nullable=False),
+        sa.Column("description", sa.Text(), nullable=False),
+        sa.Column("valid_from", sa.String(32), nullable=False),
+        sa.Column("valid_to", sa.String(32), nullable=False),
+        sa.UniqueConstraint("dataset_id", "code", "valid_from", "valid_to", name="uq_ncm_versioned_entry"),
+    )
+    op.create_index("ix_ncm_entries_dataset_id", "ncm_entries", ["dataset_id"])
+    op.create_index("ix_ncm_entries_code", "ncm_entries", ["code"])
+    op.create_table(
+        "agents",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("tenant_id", sa.Integer(), sa.ForeignKey("tenants.id"), nullable=False),
+        sa.Column("name", sa.String(128), nullable=False),
+        sa.Column("token_hash", sa.String(64), nullable=False, unique=True),
+    )
+    op.create_index("ix_agents_tenant_id", "agents", ["tenant_id"])
+    op.create_table(
+        "classification_cache",
+        sa.Column("fingerprint", sa.String(64), primary_key=True),
+        sa.Column("tenant_id", sa.Integer(), sa.ForeignKey("tenants.id"), nullable=False),
+        sa.Column("result", sa.JSON(), nullable=False),
+    )
+    op.create_index("ix_classification_cache_tenant_id", "classification_cache", ["tenant_id"])
+    op.create_table(
+        "classification_requests",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("tenant_id", sa.Integer(), sa.ForeignKey("tenants.id"), nullable=False),
+        sa.Column("agent_id", sa.Integer(), sa.ForeignKey("agents.id"), nullable=False),
+        sa.Column("fingerprint", sa.String(64), nullable=False, unique=True),
+        sa.Column("payload", sa.JSON(), nullable=False),
+        sa.Column("created_at", sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False),
+    )
+    op.create_index("ix_classification_requests_tenant_id", "classification_requests", ["tenant_id"])
+    op.create_index("ix_classification_requests_agent_id", "classification_requests", ["agent_id"])
+    op.create_table(
+        "classification_results",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("request_id", sa.Integer(), sa.ForeignKey("classification_requests.id"), nullable=False, unique=True),
+        sa.Column("result", sa.JSON(), nullable=False),
+    )
+
+
+def downgrade():
+    for table in ("classification_results", "classification_requests", "classification_cache", "agents", "ncm_entries", "datasets", "tenants"):
+        op.drop_table(table)

central/alembic/versions/0019_fix_timestamp_defaults.py

@@ -0,0 +1,43 @@
+"""Fix frozen timestamp defaults from 0006 and add model indexes never migrated.
+
+0006 declared ``server_default="now()"`` as a plain string. SQLAlchemy renders
+that as the literal ``'now()'``, which PostgreSQL casts to a timestamp once,
+when the column is created — every later row got that same fixed instant
+instead of the insert time (agents.created_at/updated_at, jobs.created_at).
+The models rely on the server default, so Jobs and Agents registered after
+0006 all carried the migration's timestamp. Existing rows are left as they
+are: their real creation time is not recoverable.
+
+The models also declare ``index=True`` on three columns that no migration
+ever indexed.
+
+Revision ID: 0019_fix_timestamp_defaults
+Revises: 0018_blocker_installation
+"""
+from alembic import op
+import sqlalchemy as sa
+
+revision = "0019_fix_timestamp_defaults"
+down_revision = "0018_blocker_installation"
+branch_labels = None
+depends_on = None
+
+_TIMESTAMP_COLUMNS = (("agents", "created_at"), ("agents", "updated_at"), ("jobs", "created_at"))
+_MISSING_INDEXES = (
+    ("ix_kb_observations_tenant_id", "kb_observations", "tenant_id"),
+    ("ix_kb_observations_company_key", "kb_observations", "company_key"),
+    ("ix_tenant_product_context_company_key", "tenant_product_context", "company_key"),
+)
+
+
+def upgrade():
+    for table, column in _TIMESTAMP_COLUMNS:
+        op.alter_column(table, column, server_default=sa.func.now(), existing_type=sa.DateTime(timezone=True), existing_nullable=False)
+    for name, table, column in _MISSING_INDEXES:
+        op.create_index(name, table, [column])
+
+
+def downgrade():
+    # The frozen-timestamp defaults are a bug and are deliberately not restored.
+    for name, table, _column in reversed(_MISSING_INDEXES):
+        op.drop_index(name, table_name=table)

central/tests/test_migration_baseline.py

@@ -0,0 +1,64 @@
+"""Guards for the migration chain being runnable from an empty database.
+
+0001 used to build the schema from the *current* ORM models, so every later
+migration collided with it on a fresh database. These checks run offline (no
+PostgreSQL) and inspect code, not comments, so they fail if that pattern
+comes back.
+"""
+import ast
+import re
+from pathlib import Path
+
+from alembic.config import Config
+from alembic.script import ScriptDirectory
+
+ROOT = Path(__file__).parents[2]
+VERSIONS = ROOT / "central" / "alembic" / "versions"
+
+
+def _migrations():
+    return {path.name: ast.parse(path.read_text(encoding="utf-8")) for path in sorted(VERSIONS.glob("*.py"))}
+
+
+def _script_directory():
+    config = Config(str(ROOT / "central" / "alembic.ini"))
+    config.set_main_option("script_location", str(ROOT / "central" / "alembic"))
+    return ScriptDirectory.from_config(config)
+
+
+def test_no_migration_builds_schema_from_current_models():
+    offenders = []
+    for name, tree in _migrations().items():
+        for node in ast.walk(tree):
+            if isinstance(node, ast.Attribute) and node.attr in ("create_all", "drop_all"):
+                offenders.append((name, node.attr))
+            if isinstance(node, ast.ImportFrom) and (node.module or "").startswith("central.app"):
+                offenders.append((name, node.module))
+    assert offenders == []
+
+
+def test_chain_is_linear_with_a_single_head():
+    script = _script_directory()
+    assert len(script.get_heads()) == 1
+    revisions = list(script.walk_revisions())
+    assert revisions[-1].revision == "0001"
+    assert revisions[-1].down_revision is None
+
+
+def test_timestamp_defaults_are_sql_functions_not_string_literals():
+    # A plain string such as "now()" is rendered as a quoted literal that
+    # PostgreSQL freezes at column creation time. 0006 did that and 0019
+    # repairs it; no other migration may repeat it.
+    offenders = []
+    for name, tree in _migrations().items():
+        if name.startswith("0006_"):
+            continue
+        for node in ast.walk(tree):
+            if (
+                isinstance(node, ast.keyword)
+                and node.arg == "server_default"
+                and isinstance(node.value, ast.Constant)
+                and re.fullmatch(r"\s*(now\s*\(\s*\)|now|current_timestamp|localtimestamp)\s*", str(node.value.value), re.IGNORECASE)
+            ):
+                offenders.append(name)
+    assert offenders == []