seed.sql

-- Seed: a small fictional music scene.
--
-- Everything here is synthetic. Names are invented; the graph shape is chosen
-- to exercise the patterns: a known shortest path, a decoy longer path, a few
-- near-duplicate names for entity resolution, and six months of activity for
-- momentum windows.
--
-- Node ids are fixed so pattern files can assert exact results.

-- ---------------------------------------------------------------------------
-- Deterministic pseudo-embeddings.
--
-- Real systems store model-produced vectors. For a self-contained demo we
-- derive a stable vector from a seed string: same seed → same vector, and a
-- `jitter` parameter produces a *near*-identical vector — which is exactly how
-- two spellings of the same real-world entity behave under a real model.
-- ---------------------------------------------------------------------------
CREATE OR REPLACE FUNCTION pseudo_embedding(seed text, jitter real DEFAULT 0)
RETURNS vector
LANGUAGE sql IMMUTABLE AS $$
    SELECT ('[' || string_agg(
                ((hashtext(seed || ':' || i) % 1000)::real / 1000.0
                 + jitter * ((hashtext('j:' || seed || ':' || i) % 100)::real / 100.0)
               )::text, ',' ORDER BY i)
            || ']')::vector(384)
    FROM generate_series(1, 384) AS i;
$$;

-- ---------------------------------------------------------------------------
-- Named cast (fixed ids used by the pattern files)
-- ---------------------------------------------------------------------------
INSERT INTO nodes (id, kind, name, name_embedding) VALUES
    -- artists
    ('00000000-0000-0000-0000-000000000001', 'person', 'Nova Reyes',       pseudo_embedding('Nova Reyes')),
    ('00000000-0000-0000-0000-000000000002', 'person', 'Cassette Ghost',   pseudo_embedding('Cassette Ghost')),
    ('00000000-0000-0000-0000-000000000003', 'person', 'Mira Solano',      pseudo_embedding('Mira Solano')),
    ('00000000-0000-0000-0000-000000000004', 'person', 'DJ Marrow',        pseudo_embedding('DJ Marrow')),
    ('00000000-0000-0000-0000-000000000005', 'person', 'Low Tide Collective', pseudo_embedding('Low Tide Collective')),
    ('00000000-0000-0000-0000-000000000006', 'person', 'Iris Vann',        pseudo_embedding('Iris Vann')),
    -- near-duplicates of 2 and 4: same underlying entity, different spelling.
    -- Embeddings are jittered copies of the original — near-identical, as a
    -- real model would produce.
    ('00000000-0000-0000-0000-000000000012', 'person', 'Casette Ghost',    pseudo_embedding('Cassette Ghost', 0.02)),
    ('00000000-0000-0000-0000-000000000014', 'person', 'Marrow',           pseudo_embedding('DJ Marrow', 0.03)),
    -- venues / orgs
    ('00000000-0000-0000-0000-000000000021', 'venue',  'The Velvet Room',  pseudo_embedding('The Velvet Room')),
    ('00000000-0000-0000-0000-000000000022', 'venue',  'Warehouse North',  pseudo_embedding('Warehouse North')),
    ('00000000-0000-0000-0000-000000000023', 'org',    'Night Signal Records', pseudo_embedding('Night Signal Records')),
    -- works
    ('00000000-0000-0000-0000-000000000031', 'work',   'Glasshouse EP',    pseudo_embedding('Glasshouse EP')),
    ('00000000-0000-0000-0000-000000000032', 'work',   'Midnight Protocol', pseudo_embedding('Midnight Protocol'));

-- Filler population: 40 session artists, deterministic names.
INSERT INTO nodes (id, kind, name, name_embedding)
SELECT ('00000000-0000-0000-0001-' || lpad(i::text, 12, '0'))::uuid,
       'person',
       'Session Artist ' || lpad(i::text, 2, '0'),
       pseudo_embedding('Session Artist ' || i)
FROM generate_series(1, 40) AS i;

-- ---------------------------------------------------------------------------
-- Edges.
--
-- The traversal story (patterns 01/02):
--   Nova Reyes ↔ Cassette Ghost ↔ The Velvet Room ↔ Mira Solano   (3 hops)
-- plus a decoy 5-hop route Nova → DJ Marrow → Warehouse North →
-- Low Tide Collective → Iris Vann → Mira Solano, so shortest-path has to
-- actually pick the short one.
-- ---------------------------------------------------------------------------
INSERT INTO edges (src, tgt, kind, direction, confidence) VALUES
    -- short route
    ('00000000-0000-0000-0000-000000000001', '00000000-0000-0000-0000-000000000002', 'collaborated_with', 'bidirectional', 0.95),
    ('00000000-0000-0000-0000-000000000002', '00000000-0000-0000-0000-000000000021', 'performed_at',      'directed',      0.90),
    ('00000000-0000-0000-0000-000000000003', '00000000-0000-0000-0000-000000000021', 'performed_at',      'directed',      0.90),
    -- decoy long route
    ('00000000-0000-0000-0000-000000000001', '00000000-0000-0000-0000-000000000004', 'collaborated_with', 'bidirectional', 0.80),
    ('00000000-0000-0000-0000-000000000004', '00000000-0000-0000-0000-000000000022', 'performed_at',      'directed',      0.85),
    ('00000000-0000-0000-0000-000000000005', '00000000-0000-0000-0000-000000000022', 'performed_at',      'directed',      0.70),
    ('00000000-0000-0000-0000-000000000005', '00000000-0000-0000-0000-000000000006', 'collaborated_with', 'bidirectional', 0.75),
    ('00000000-0000-0000-0000-000000000006', '00000000-0000-0000-0000-000000000003', 'collaborated_with', 'bidirectional', 0.85),
    -- label / works context
    ('00000000-0000-0000-0000-000000000023', '00000000-0000-0000-0000-000000000002', 'manages',   'directed', 0.90),
    ('00000000-0000-0000-0000-000000000002', '00000000-0000-0000-0000-000000000031', 'created',   'directed', 1.00),
    ('00000000-0000-0000-0000-000000000001', '00000000-0000-0000-0000-000000000031', 'created',   'directed', 1.00),
    ('00000000-0000-0000-0000-000000000004', '00000000-0000-0000-0000-000000000032', 'created',   'directed', 1.00);

-- Filler edges: each session artist collaborates with the next two (a chain
-- with local clustering), and every fifth one performed at Warehouse North.
INSERT INTO edges (src, tgt, kind, direction, confidence)
SELECT ('00000000-0000-0000-0001-' || lpad(i::text, 12, '0'))::uuid,
       ('00000000-0000-0000-0001-' || lpad((i + o)::text, 12, '0'))::uuid,
       'collaborated_with', 'bidirectional', 0.5 + (i % 5) / 10.0
FROM generate_series(1, 38) AS i, (VALUES (1), (2)) AS hop(o)
WHERE i + o <= 40;

INSERT INTO edges (src, tgt, kind, direction, confidence)
SELECT ('00000000-0000-0000-0001-' || lpad(i::text, 12, '0'))::uuid,
       '00000000-0000-0000-0000-000000000022', 'performed_at', 'directed', 0.8
FROM generate_series(5, 40, 5) AS i;

-- Bridge the filler chain into the named cast so the graph is connected.
INSERT INTO edges (src, tgt, kind, direction, confidence) VALUES
    ('00000000-0000-0000-0001-000000000001', '00000000-0000-0000-0000-000000000005', 'collaborated_with', 'bidirectional', 0.6);

-- ---------------------------------------------------------------------------
-- Events: six months of activity, denser in the last 30 days for some nodes
-- so momentum trends differ (pattern 06). Timestamps are relative to now(),
-- so windows behave the same whenever you run the demo.
-- ---------------------------------------------------------------------------

-- Cassette Ghost: accelerating (2/mo historically, ~2/wk in the last month).
INSERT INTO events (node_id, kind, weight, created_at)
SELECT '00000000-0000-0000-0000-000000000002', 'release', 1.0, now() - (i || ' days')::interval
FROM unnest(ARRAY[2, 6, 11, 17, 24, 29, 44, 75, 106, 137, 168]) AS i;

-- Mira Solano: steady (one event a month).
INSERT INTO events (node_id, kind, weight, created_at)
SELECT '00000000-0000-0000-0000-000000000003', 'show', 1.0, now() - (i * 30 || ' days')::interval
FROM generate_series(0, 5) AS i;

-- DJ Marrow: cooling off (active months 2-6, silent since).
INSERT INTO events (node_id, kind, weight, created_at)
SELECT '00000000-0000-0000-0000-000000000004', 'show', 1.0, now() - (i || ' days')::interval
FROM unnest(ARRAY[45, 60, 74, 90, 110, 130, 155, 170]) AS i;

-- Background noise: every session artist has a smattering of events.
INSERT INTO events (node_id, kind, weight, created_at)
SELECT ('00000000-0000-0000-0001-' || lpad(i::text, 12, '0'))::uuid,
       'show', 1.0,
       now() - ((i * 13 + s * 41) % 180 || ' days')::interval
FROM generate_series(1, 40) AS i, generate_series(1, 3) AS s;

ANALYZE nodes; ANALYZE edges; ANALYZE events;