DATA_MODEL — analytics-service

Sibling: DOMAIN_MODEL · APPLICATION_LOGIC · SECURITY_MODEL · platform anchor: docs/06 Data Models, docs/standards/NAMING

analytics-service owns two storage surfaces:

Cloud SQL Postgres (analytics schema) — metadata: projection definitions, metric definitions, dashboards, widgets, queries, ETL jobs/runs, DQ checks/results pointers, outbox/inbox.
BigQuery (events_raw.*, analytics_curated.*, dq_results.*) — the warehouse layer.

1. Identity strategy

ULIDs with prefixes per DOMAIN_MODEL §1. Prefixes (prj_, met_, dsh_, wid_, qry_, etl_, etr_, dqc_, dqr_) are added to the canonical registry in docs/standards/NAMING §6.

Curated table rows are keyed by domain natural keys (reservation_id, payment_id, etc.) plus tenant_id and _event_id for idempotent MERGE.

2. Storage placement

Store	Purpose
Cloud SQL Postgres `analytics`	Projection/metric/dashboard metadata, ETL state, DQ checks, outbox, inbox
BigQuery `events_raw.*`	Raw event landing (one table per subject, partitioned by ingestion ts, clustered by `tenant_id`)
BigQuery `analytics_curated.*`	Curated fact + dim tables (versioned `_v<n>`)
BigQuery `dq_results.*`	Historical DQ check results (immutable append)
BigQuery `tenant_views.*`	Per-tenant authorized views over curated tables
Memorystore Redis	Widget data cache, byte-budget counters, hot metric cache
GCS staging buckets	Backfill exports + Looker Studio embed signing material (key refs only)

3. Cloud SQL DDL (metadata)

3.1 `analytics.projections`

CREATE TABLE analytics.projections (
  id              text PRIMARY KEY,                  -- prj_<ulid>
  key             text NOT NULL,                     -- 'fact_reservation'
  schema_version  int  NOT NULL CHECK (schema_version >= 1),
  target_dataset  text NOT NULL,
  target_table    text NOT NULL,                     -- 'fact_reservation_v1'
  partitioning    jsonb NOT NULL,
  clustering      text[] NOT NULL,
  source_query    text NOT NULL,                     -- MERGE … USING (…) ON …
  merge_key       text[] NOT NULL,
  refresh_policy  text NOT NULL CHECK (refresh_policy IN ('incremental','full')),
  window_minutes  int  NOT NULL CHECK (window_minutes >= 5),
  freshness       jsonb NOT NULL,
  archived        boolean NOT NULL DEFAULT false,
  created_at      timestamptz NOT NULL DEFAULT now(),
  updated_at      timestamptz NOT NULL DEFAULT now(),
  version         int NOT NULL DEFAULT 0,            -- OCC

  UNIQUE (key, schema_version)
);

3.2 `analytics.metric_definitions`

CREATE TABLE analytics.metric_definitions (
  id               text PRIMARY KEY,                 -- met_<ulid>
  tenant_id        text,                              -- NULL = platform-shared
  key              text NOT NULL,                     -- 'reservation.occupancy_pct'
  metric_version   int  NOT NULL CHECK (metric_version >= 1),
  display          jsonb NOT NULL,
  unit             jsonb NOT NULL,
  grain            text NOT NULL CHECK (grain IN ('minute','hour','day','week','month','quarter','year')),
  dimensions       jsonb NOT NULL,
  source_tables    text[] NOT NULL,
  schema_version   int  NOT NULL,
  params           jsonb NOT NULL,
  sql_template     text NOT NULL,
  byte_cap         jsonb NOT NULL,
  freshness        jsonb NOT NULL,
  archived         boolean NOT NULL DEFAULT false,
  created_at       timestamptz NOT NULL DEFAULT now(),
  updated_at       timestamptz NOT NULL DEFAULT now(),
  version          int NOT NULL DEFAULT 0,
  UNIQUE (tenant_id, key, metric_version)
);

ALTER TABLE analytics.metric_definitions ENABLE ROW LEVEL SECURITY;
CREATE POLICY metric_definitions_tenant_isolation ON analytics.metric_definitions
  USING (tenant_id IS NULL OR tenant_id = current_setting('app.tenant_id', true));

3.3 `analytics.dashboards`, `analytics.widgets`, `analytics.queries`

CREATE TABLE analytics.dashboards (
  id              text PRIMARY KEY,                  -- dsh_<ulid>
  tenant_id       text NOT NULL,
  owner_user_id   text NOT NULL,
  scope           text NOT NULL CHECK (scope IN ('tenant','property','private')),
  property_id     text,
  name_i18n       jsonb NOT NULL,
  description_i18n jsonb,
  widget_ids      text[] NOT NULL DEFAULT '{}',
  shared_with     jsonb NOT NULL DEFAULT '[]'::jsonb,
  created_at      timestamptz NOT NULL DEFAULT now(),
  updated_at      timestamptz NOT NULL DEFAULT now(),
  version         int NOT NULL DEFAULT 0,
  CHECK (scope <> 'property' OR property_id IS NOT NULL)
);

ALTER TABLE analytics.dashboards ENABLE ROW LEVEL SECURITY;
CREATE POLICY dashboards_tenant_isolation ON analytics.dashboards
  USING (tenant_id = current_setting('app.tenant_id', true));

CREATE INDEX ix_dashboards_tenant_owner ON analytics.dashboards (tenant_id, owner_user_id);

CREATE TABLE analytics.widgets (
  id              text PRIMARY KEY,                  -- wid_<ulid>
  dashboard_id    text NOT NULL REFERENCES analytics.dashboards(id),
  tenant_id       text NOT NULL,
  spec            jsonb NOT NULL,
  position        jsonb NOT NULL,
  created_at      timestamptz NOT NULL DEFAULT now(),
  updated_at      timestamptz NOT NULL DEFAULT now(),
  version         int NOT NULL DEFAULT 0
);
CREATE INDEX ix_widgets_dashboard ON analytics.widgets (dashboard_id);

ALTER TABLE analytics.widgets ENABLE ROW LEVEL SECURITY;
CREATE POLICY widgets_tenant_isolation ON analytics.widgets
  USING (tenant_id = current_setting('app.tenant_id', true));

CREATE TABLE analytics.queries (
  id              text PRIMARY KEY,                  -- qry_<ulid>
  tenant_id       text NOT NULL,
  owner_user_id   text NOT NULL,
  name_i18n       jsonb NOT NULL,
  source_tables   text[] NOT NULL,
  schema_version  int  NOT NULL,
  sql_template    text NOT NULL,
  params          jsonb NOT NULL,
  byte_cap        jsonb NOT NULL,
  archived        boolean NOT NULL DEFAULT false,
  created_at      timestamptz NOT NULL DEFAULT now(),
  updated_at      timestamptz NOT NULL DEFAULT now(),
  version         int NOT NULL DEFAULT 0
);

ALTER TABLE analytics.queries ENABLE ROW LEVEL SECURITY;
CREATE POLICY queries_tenant_isolation ON analytics.queries
  USING (tenant_id = current_setting('app.tenant_id', true));

3.4 `analytics.etl_jobs`, `analytics.etl_runs`

CREATE TABLE analytics.etl_jobs (
  id              text PRIMARY KEY,                  -- etl_<ulid>
  projection_id   text NOT NULL REFERENCES analytics.projections(id),
  trigger         text NOT NULL CHECK (trigger IN ('cron','on_demand','backfill','event')),
  window_from     timestamptz NOT NULL,
  window_to       timestamptz NOT NULL,
  scheduled_at    timestamptz NOT NULL DEFAULT now(),
  version         int NOT NULL DEFAULT 0,
  CHECK (window_from < window_to)
);
CREATE INDEX ix_etl_jobs_projection ON analytics.etl_jobs (projection_id, scheduled_at DESC);

CREATE TABLE analytics.etl_runs (
  id               text PRIMARY KEY,                 -- etr_<ulid>
  job_id           text NOT NULL REFERENCES analytics.etl_jobs(id),
  status           text NOT NULL CHECK (status IN ('queued','running','succeeded','failed','cancelled')),
  bytes_scanned    bigint NOT NULL DEFAULT 0,
  bytes_written    bigint NOT NULL DEFAULT 0,
  rows_affected    bigint NOT NULL DEFAULT 0,
  slot_ms          bigint NOT NULL DEFAULT 0,
  started_at       timestamptz,
  succeeded_at     timestamptz,
  failed_at        timestamptz,
  error_code       text,
  error_detail     text,
  retry_of_run_id  text,
  attempt          int NOT NULL DEFAULT 1,
  warehouse_job_id text,
  version          int NOT NULL DEFAULT 0
);
CREATE INDEX ix_etl_runs_job_status ON analytics.etl_runs (job_id, status);
CREATE INDEX ix_etl_runs_status_started ON analytics.etl_runs (status, started_at DESC);

3.5 `analytics.dq_checks`, `analytics.dq_latest_results`

CREATE TABLE analytics.dq_checks (
  id          text PRIMARY KEY,                       -- dqc_<ulid>
  key         text NOT NULL UNIQUE,
  rule        jsonb NOT NULL,
  severity    text NOT NULL CHECK (severity IN ('info','warn','critical')),
  enabled     boolean NOT NULL DEFAULT true,
  version     int NOT NULL DEFAULT 0
);

CREATE TABLE analytics.dq_latest_results (
  check_id        text PRIMARY KEY REFERENCES analytics.dq_checks(id),
  observed_value  numeric NOT NULL,
  expected_value  numeric,
  status          text NOT NULL CHECK (status IN ('ok','warn','critical','error')),
  observed_at     timestamptz NOT NULL,
  bigquery_job_id text
);

History rows are written to the dq_results BigQuery dataset (§ 5.3) instead of bloating Postgres.

3.6 `analytics.tenant_query_budget`

CREATE TABLE analytics.tenant_query_budget (
  tenant_id        text PRIMARY KEY,
  daily_limit_bytes bigint NOT NULL,
  current_window_start date NOT NULL,
  current_window_bytes bigint NOT NULL DEFAULT 0,
  updated_at       timestamptz NOT NULL DEFAULT now()
);

3.7 `analytics.outbox` and `analytics.inbox_processed`

Standard platform shapes.

CREATE TABLE analytics.outbox (
  id            bigserial PRIMARY KEY,
  envelope_id   text NOT NULL UNIQUE,
  tenant_id     text NOT NULL,
  subject       text NOT NULL,
  ordering_key  text NOT NULL,
  payload       jsonb NOT NULL,
  created_at    timestamptz NOT NULL DEFAULT now(),
  published_at  timestamptz,
  published_attempts int NOT NULL DEFAULT 0
);
CREATE INDEX ix_outbox_unpublished ON analytics.outbox (created_at) WHERE published_at IS NULL;

CREATE TABLE analytics.inbox_processed (
  subject     text NOT NULL,
  message_id  text NOT NULL,
  processed_at timestamptz NOT NULL DEFAULT now(),
  PRIMARY KEY (subject, message_id)
);

4. BigQuery datasets

Dataset	Purpose	Retention
`events_raw`	Raw event landing per subject	730 d for operational, 7 y for operational_7y, 10 y object-locked policy mirrored from GCS for regulated
`analytics_curated`	Curated fact + dim tables (versioned)	rolling 5 y default, configurable per table
`dq_results`	DQ result history	2 y
`tenant_views`	Per-tenant authorized views	n/a (views)
`analytics_staging`	Backfill staging tables (TTL 7 d)	7 d

5. Curated table BigQuery DDL (selected)

5.1 `analytics_curated.fact_reservation_v1`

CREATE TABLE `proj.analytics_curated.fact_reservation_v1`
(
  tenant_id        STRING NOT NULL,
  reservation_id   STRING NOT NULL,
  property_id      STRING NOT NULL,
  room_type_id     STRING NOT NULL,
  rate_plan_id     STRING NOT NULL,
  channel          STRING NOT NULL,             -- 'direct'|'ota_booking'|'ota_expedia'|'meta'|'walk_in'
  guest_country    STRING,
  arrival_date     DATE NOT NULL,
  departure_date   DATE NOT NULL,
  business_date    DATE NOT NULL,                -- often arrival_date; used for partitioning
  nights           INT64 NOT NULL,
  status           STRING NOT NULL,             -- 'confirmed'|'cancelled'|'no_show'|'checked_in'|'checked_out'
  adults           INT64 NOT NULL,
  children         INT64 NOT NULL,
  rooms            INT64 NOT NULL,
  gross_amount     NUMERIC,
  taxes_amount     NUMERIC,
  currency         STRING,
  lead_time_days   INT64,
  confirmed_at     TIMESTAMP,
  cancelled_at     TIMESTAMP,
  no_show_at       TIMESTAMP,
  checked_in_at    TIMESTAMP,
  checked_out_at   TIMESTAMP,
  _event_id        STRING NOT NULL,             -- last event applied
  _ingested_at     TIMESTAMP NOT NULL,
  _schema_version  INT64 NOT NULL DEFAULT 1
)
PARTITION BY business_date
CLUSTER BY tenant_id, property_id, room_type_id
OPTIONS(
  description = "One row per reservation (latest state). Idempotent MERGE keyed on (tenant_id, reservation_id).",
  partition_expiration_days = 1825,
  require_partition_filter = true
);

5.2 `analytics_curated.fact_payment_v1`

CREATE TABLE `proj.analytics_curated.fact_payment_v1`
(
  tenant_id        STRING NOT NULL,
  payment_id       STRING NOT NULL,
  reservation_id   STRING,
  property_id      STRING NOT NULL,
  channel          STRING,
  method           STRING NOT NULL,             -- 'cash'|'card'|'transfer'|'wallet'|'ota_settlement'
  amount           NUMERIC NOT NULL,
  currency         STRING NOT NULL,
  fx_to_tenant_ccy NUMERIC,
  business_date    DATE NOT NULL,
  status           STRING NOT NULL,             -- 'authorized'|'captured'|'refunded'|'failed'|'voided'
  refund_of_id     STRING,
  posted_at        TIMESTAMP,
  _event_id        STRING NOT NULL,
  _ingested_at     TIMESTAMP NOT NULL,
  _schema_version  INT64 NOT NULL DEFAULT 1
)
PARTITION BY business_date
CLUSTER BY tenant_id, property_id, method
OPTIONS(require_partition_filter = true);

5.3 `analytics_curated.fact_housekeeping_task_v1`

CREATE TABLE `proj.analytics_curated.fact_housekeeping_task_v1`
(
  tenant_id        STRING NOT NULL,
  task_id          STRING NOT NULL,
  property_id      STRING NOT NULL,
  room_id          STRING NOT NULL,
  task_type        STRING NOT NULL,             -- 'turn_down'|'check_out_clean'|'maintenance'|'…'
  assigned_to_user_id STRING,
  status           STRING NOT NULL,             -- 'queued'|'in_progress'|'done'|'failed'
  business_date    DATE NOT NULL,
  duration_seconds INT64,
  started_at       TIMESTAMP,
  completed_at     TIMESTAMP,
  _event_id        STRING NOT NULL,
  _ingested_at     TIMESTAMP NOT NULL,
  _schema_version  INT64 NOT NULL DEFAULT 1
)
PARTITION BY business_date
CLUSTER BY tenant_id, property_id, status
OPTIONS(require_partition_filter = true);

5.4 `analytics_curated.fact_lock_action_v1`

CREATE TABLE `proj.analytics_curated.fact_lock_action_v1`
(
  tenant_id     STRING NOT NULL,
  action_id     STRING NOT NULL,
  property_id   STRING NOT NULL,
  room_id       STRING,
  lock_id       STRING NOT NULL,
  actor_kind    STRING NOT NULL,                -- 'guest'|'staff'|'system'
  actor_id      STRING,
  action        STRING NOT NULL,                -- 'unlock'|'lock'|'denied'|'low_battery'|'tamper'
  result        STRING NOT NULL,                -- 'success'|'failure'
  business_date DATE NOT NULL,
  occurred_at   TIMESTAMP NOT NULL,
  latency_ms    INT64,
  _event_id     STRING NOT NULL,
  _ingested_at  TIMESTAMP NOT NULL,
  _schema_version INT64 NOT NULL DEFAULT 1
)
PARTITION BY business_date
CLUSTER BY tenant_id, property_id, action;

5.5 Dimensions

CREATE TABLE `proj.analytics_curated.dim_property_v1` (
  tenant_id    STRING NOT NULL,
  property_id  STRING NOT NULL,
  name         STRING,
  city         STRING,
  country      STRING,
  timezone     STRING,
  rooms_count  INT64,
  open_date    DATE,
  closed_date  DATE,
  _ingested_at TIMESTAMP NOT NULL,
  _schema_version INT64 NOT NULL DEFAULT 1
) CLUSTER BY tenant_id, property_id;

CREATE TABLE `proj.analytics_curated.dim_room_type_v1` (
  tenant_id    STRING NOT NULL,
  room_type_id STRING NOT NULL,
  property_id  STRING NOT NULL,
  code         STRING,
  capacity     INT64,
  base_rate    NUMERIC,
  _ingested_at TIMESTAMP NOT NULL,
  _schema_version INT64 NOT NULL DEFAULT 1
) CLUSTER BY tenant_id, property_id, room_type_id;

CREATE TABLE `proj.analytics_curated.dim_tenant_v1` (
  tenant_id     STRING NOT NULL,
  display_name  STRING,
  data_residency STRING,
  plan          STRING,
  created_at    TIMESTAMP,
  status        STRING,
  _ingested_at  TIMESTAMP NOT NULL,
  _schema_version INT64 NOT NULL DEFAULT 1
) CLUSTER BY tenant_id;

CREATE TABLE `proj.analytics_curated.dim_calendar_v1` (
  date         DATE NOT NULL,
  year         INT64 NOT NULL,
  quarter      INT64 NOT NULL,
  month        INT64 NOT NULL,
  iso_week     INT64 NOT NULL,
  day_of_week  INT64 NOT NULL,
  is_weekend   BOOL NOT NULL,
  is_holiday   BOOL NOT NULL,
  holiday_name STRING
);

5.6 `analytics_curated.fact_demand_forecast_v1`

CREATE TABLE `proj.analytics_curated.fact_demand_forecast_v1` (
  tenant_id     STRING NOT NULL,
  property_id   STRING NOT NULL,
  room_type_id  STRING NOT NULL,
  business_date DATE NOT NULL,
  horizon_days  INT64 NOT NULL,
  predicted_occupancy_pct NUMERIC,
  predicted_adr  NUMERIC,
  ci80_low_occupancy NUMERIC,
  ci80_high_occupancy NUMERIC,
  model_id       STRING NOT NULL,
  model_version  STRING NOT NULL,
  produced_at    TIMESTAMP NOT NULL,
  _event_id      STRING NOT NULL,
  _ingested_at   TIMESTAMP NOT NULL,
  _schema_version INT64 NOT NULL DEFAULT 1
)
PARTITION BY business_date
CLUSTER BY tenant_id, property_id, room_type_id;

6. Frozen metric SQL (canonical examples)

These are the exact SQL bodies stored in metric_definitions.sql_template. The orchestrator binds @tenant_id, @from, @to, etc.

6.1 `reservation.occupancy_pct` v1

WITH room_nights_available AS (
  SELECT d.date AS business_date, p.rooms_count AS rooms_avail
  FROM `proj.analytics_curated.dim_calendar_v1` d
  CROSS JOIN `proj.analytics_curated.dim_property_v1` p
  WHERE p.tenant_id = @tenant_id
    AND (@property_id IS NULL OR p.property_id = @property_id)
    AND d.date BETWEEN @from AND @to
),
room_nights_sold AS (
  SELECT business_date, SUM(rooms) AS sold
  FROM `proj.analytics_curated.fact_reservation_v1`
  WHERE tenant_id = @tenant_id
    AND status IN ('confirmed','checked_in','checked_out')
    AND (@property_id IS NULL OR property_id = @property_id)
    AND business_date BETWEEN @from AND @to
  GROUP BY business_date
)
SELECT a.business_date AS date,
       SAFE_DIVIDE(SUM(s.sold), SUM(a.rooms_avail)) * 100 AS value
FROM room_nights_available a
LEFT JOIN room_nights_sold s USING (business_date)
GROUP BY date
ORDER BY date;

6.2 `reservation.adr` v1

SELECT business_date AS date,
       SAFE_DIVIDE(SUM(gross_amount - IFNULL(taxes_amount,0)), SUM(rooms)) AS value
FROM `proj.analytics_curated.fact_reservation_v1`
WHERE tenant_id = @tenant_id
  AND status IN ('checked_out','checked_in')
  AND (@property_id IS NULL OR property_id = @property_id)
  AND business_date BETWEEN @from AND @to
GROUP BY date
ORDER BY date;

6.3 `reservation.revpar` v1

WITH adr AS ( /* same as 6.2 */ ),
     occ AS ( /* same as 6.1 */ )
SELECT a.date,
       (adr.value * occ.value / 100) AS value
FROM adr JOIN occ USING (date);

6.4 `funnel.meta_to_book_pct` v1

SELECT business_date AS date,
       SAFE_DIVIDE(
         COUNTIF(channel='meta' AND status IN ('confirmed','checked_in','checked_out')),
         COUNTIF(channel='meta')
       ) * 100 AS value
FROM `proj.analytics_curated.fact_reservation_v1`
WHERE tenant_id = @tenant_id
  AND business_date BETWEEN @from AND @to
GROUP BY date;

7. Authorized views (RLS)

Per SECURITY_MODEL §3, every curated table is exposed to tenant identities only via authorized views in tenant_views.<tenant_id>.<table> (or a single tenant_views.fact_reservation_v1 view that resolves caller identity). Pattern:

CREATE VIEW `proj.tenant_views.fact_reservation_v1` AS
SELECT *
FROM `proj.analytics_curated.fact_reservation_v1`
WHERE tenant_id = SESSION_USER_TENANT_ID();

SESSION_USER_TENANT_ID() is a UDF that resolves the caller's binding from tenant_views.access_bindings (per §8). The Looker Studio Community Connector uses these views; the Query API binds @tenant_id directly and never trusts caller-provided values.

8. Tenant access bindings

CREATE TABLE `proj.tenant_views.access_bindings` (
  google_principal STRING NOT NULL,    -- 'user:foo@hotel.com' or 'serviceAccount:looker-studio-…'
  tenant_id        STRING NOT NULL,
  property_ids     ARRAY<STRING>,
  granted_at       TIMESTAMP NOT NULL,
  granted_by       STRING NOT NULL,
  PRIMARY KEY(google_principal, tenant_id) NOT ENFORCED
);

Binding rows are managed by BootstrapTenantAnalyticsUseCase and RebindAuthorizedViewsUseCase. Removed when a user permission is revoked or a tenant is deleted.

9. Indexes & query patterns

Hot query	Index used
Find due ETL jobs	`ix_etl_jobs_projection` + run scan
Latest run per projection	`ix_etl_runs_status_started`
Dashboard list per user	`ix_dashboards_tenant_owner`
Widget read	PK + dashboard FK
Outbox publish loop	`ix_outbox_unpublished`

10. Backups & retention

Cloud SQL: daily backup, PITR 14 d.
BigQuery: time-travel 7 d (default), table snapshot exports nightly to GCS for catastrophic recovery (last 30 d).
events_raw.* mirrors GCS event archive retention class; deletion respects regulated 10 y object-lock when applicable (we never DELETE; we use partition expiration and tenant-scoped purge).
DQ results retained 2 y in BigQuery.

11. Migrations

Postgres migrations under db/migrations/ (Drizzle), forward-only.
BigQuery DDL applied through Terraform (terraform/bigquery/); curated table changes use the _v<n> coexistence pattern from MIGRATION_PLAN.

Cross-references: SECURITY_MODEL §3, APPLICATION_LOGIC §4, docs/06.

1. Identity strategy​

2. Storage placement​

3. Cloud SQL DDL (metadata)​

3.1 analytics.projections​

3.2 analytics.metric_definitions​

3.3 analytics.dashboards, analytics.widgets, analytics.queries​

3.4 analytics.etl_jobs, analytics.etl_runs​

3.5 analytics.dq_checks, analytics.dq_latest_results​

3.6 analytics.tenant_query_budget​

3.7 analytics.outbox and analytics.inbox_processed​

4. BigQuery datasets​

5. Curated table BigQuery DDL (selected)​

5.1 analytics_curated.fact_reservation_v1​

5.2 analytics_curated.fact_payment_v1​

5.3 analytics_curated.fact_housekeeping_task_v1​

5.4 analytics_curated.fact_lock_action_v1​

5.5 Dimensions​

5.6 analytics_curated.fact_demand_forecast_v1​

6. Frozen metric SQL (canonical examples)​

6.1 reservation.occupancy_pct v1​

6.2 reservation.adr v1​

6.3 reservation.revpar v1​

6.4 funnel.meta_to_book_pct v1​

7. Authorized views (RLS)​

8. Tenant access bindings​

9. Indexes & query patterns​

10. Backups & retention​

11. Migrations​