Document event and Postgres integration contracts

2026-06-10 20:12:27 +00:00
parent 4ac4e401ed
commit abb8f218ec
6 changed files with 658 additions and 407 deletions
--- a/docs/integrations/events.md
+++ b/docs/integrations/events.md
@@ -0,0 +1,238 @@
+# Event Wire Contract
+
+This document is the canonical JSON contract for events emitted by
+`weatherfeeder` JSON sinks, including stdout and NATS. Postgres stores the same
+event envelope fields in parent table columns; see
+[Postgres integration](postgres.md).
+
+Downstream consumers should read the envelope, switch on `schema`, and decode
+`payload` according to that schema.
+
+## Envelope
+
+Every emitted event is a JSON object with these fields:
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `id` | string | yes | Stable event identifier. Treat as opaque. |
+| `kind` | string | yes | Routing kind, such as `observation` or `alert`. |
+| `source` | string | yes | Configured source name. |
+| `emitted_at` | timestamp | yes | When the daemon emitted the event. |
+| `effective_at` | timestamp | no | Timestamp the payload is about, when known. |
+| `schema` | string | no | Schema identifier. Weatherfeeder sources and normalizers set this. |
+| `payload` | object, array, string, or scalar | yes | Schema-specific payload. |
+
+Timestamps are JSON strings using Go `time.Time` JSON encoding, which is
+RFC3339Nano-compatible. Weatherfeeder normalizers use UTC timestamps for
+canonical payloads.
+
+## Kinds And Schemas
+
+Canonical schemas emitted after normalization:
+
+| Kind | Schema | Payload |
+|---|---|---|
+| `observation` | `weather.observation.v1` | `WeatherObservation` |
+| `forecast` | `weather.forecast.v1` | `WeatherForecastRun` |
+| `forecast_discussion` | `weather.forecast_discussion.v1` | `WeatherForecastDiscussion` |
+| `weather_story` | `weather.weather_story.v1` | `WeatherStoryRun` |
+| `alert` | `weather.alert.v1` | `WeatherAlertRun` |
+
+Raw upstream schemas emitted by current sources:
+
+| Kind | Schema | Payload |
+|---|---|---|
+| `observation` | `raw.nws.observation.v1` | NWS observation JSON |
+| `observation` | `raw.openmeteo.current.v1` | Open-Meteo current JSON |
+| `observation` | `raw.openweather.current.v1` | OpenWeather current JSON |
+| `forecast` | `raw.nws.hourly.forecast.v1` | NWS hourly forecast JSON |
+| `forecast` | `raw.nws.narrative.forecast.v1` | NWS narrative forecast JSON |
+| `forecast_discussion` | `raw.nws.forecast_discussion.v1` | NWS forecast discussion HTML string |
+| `weather_story` | `raw.nws.weatherstories.v1` | NWS weather stories JSON |
+| `forecast` | `raw.openmeteo.hourly.forecast.v1` | Open-Meteo hourly forecast JSON |
+| `alert` | `raw.nws.alerts.v1` | NWS alerts JSON |
+
+`standards.SchemaRawOpenWeatherHourlyForecastV1` exists in code, but no current
+registered source emits it.
+
+## Shared Conventions
+
+- Canonical numeric measurements use metric units.
+- Floating-point values in canonical payloads are rounded to 4 digits after the
+  decimal point during normalization.
+- Optional fields use JSON `omitempty`; absent fields should be treated as
+  unknown.
+- `conditionCode` is a WMO weather interpretation code. Unknown observation
+  conditions use `-1`. Forecast period `conditionCode` is optional.
+- Additive fields are compatible within a schema version. Removing, renaming, or
+  changing the meaning of a field requires a new schema identifier.
+
+## `weather.observation.v1`
+
+Payload type: `WeatherObservation`.
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `stationId` | string | no | Provider station/location identifier. |
+| `stationName` | string | no | Human station name. |
+| `timestamp` | timestamp | yes | Observation timestamp. |
+| `conditionCode` | integer | yes | WMO code; `-1` means unknown. |
+| `isDay` | boolean | no | Day/night hint. |
+| `textDescription` | string | no | Short human description. |
+| `temperatureC` | number | no | Celsius. |
+| `dewpointC` | number | no | Celsius. |
+| `windDirectionDegrees` | number | no | Degrees. |
+| `windSpeedKmh` | number | no | Kilometers per hour. |
+| `windGustKmh` | number | no | Kilometers per hour. |
+| `barometricPressurePa` | number | no | Pascals. |
+| `visibilityMeters` | number | no | Meters. |
+| `relativeHumidityPercent` | number | no | Percent from 0 to 100. |
+| `apparentTemperatureC` | number | no | Celsius. |
+| `presentWeather` | array | no | Provider-specific present weather fragments. |
+
+`presentWeather[]` entries contain optional `raw` objects.
+
+## `weather.forecast.v1`
+
+Payload type: `WeatherForecastRun`.
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `locationId` | string | no | Provider location identifier. |
+| `locationName` | string | no | Human location name. |
+| `issuedAt` | timestamp | yes | When the forecast run was generated or issued. |
+| `updatedAt` | timestamp | no | Later provider update time. |
+| `product` | string | yes | `hourly`, `narrative`, or `daily`. |
+| `latitude` | number | no | Degrees. |
+| `longitude` | number | no | Degrees. |
+| `elevationMeters` | number | no | Meters. |
+| `periods` | array | yes | Ordered forecast periods. |
+
+`periods[]` entries:
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `startTime` | timestamp | yes | Period start. |
+| `endTime` | timestamp | yes | Period end. |
+| `name` | string | no | Human label. |
+| `isDay` | boolean | no | Day/night hint. |
+| `conditionCode` | integer | no | WMO code when applicable. |
+| `textDescription` | string | no | Human summary. |
+| `temperatureC` | number | no | Celsius. |
+| `temperatureCMin` | number | no | Celsius. |
+| `temperatureCMax` | number | no | Celsius. |
+| `dewpointC` | number | no | Celsius. |
+| `relativeHumidityPercent` | number | no | Percent from 0 to 100. |
+| `windDirectionDegrees` | number | no | Degrees. |
+| `windSpeedKmh` | number | no | Kilometers per hour. |
+| `windGustKmh` | number | no | Kilometers per hour. |
+| `barometricPressurePa` | number | no | Pascals. |
+| `visibilityMeters` | number | no | Meters. |
+| `apparentTemperatureC` | number | no | Celsius. |
+| `cloudCoverPercent` | number | no | Percent from 0 to 100. |
+| `probabilityOfPrecipitationPercent` | number | no | Percent from 0 to 100. |
+| `precipitationAmountMm` | number | no | Liquid-equivalent millimeters. |
+| `snowfallDepthMm` | number | no | Millimeters. |
+| `uvIndex` | number | no | Unitless index. |
+
+## `weather.forecast_discussion.v1`
+
+Payload type: `WeatherForecastDiscussion`.
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `officeId` | string | no | NWS office identifier. |
+| `officeName` | string | no | Office name. |
+| `product` | string | yes | Current value is `afd`. |
+| `issuedAt` | timestamp | yes | Bulletin issue time. |
+| `updatedAt` | timestamp | no | Later update time. |
+| `keyMessages` | array of strings | no | Extracted key messages. |
+| `shortTerm` | object | no | Short-term section. |
+| `longTerm` | object | no | Long-term section. |
+
+`shortTerm` and `longTerm` sections contain optional `qualifier`, `issuedAt`,
+and `text` fields.
+
+## `weather.weather_story.v1`
+
+Payload type: `WeatherStoryRun`.
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `officeId` | string | no | NWS office identifier. |
+| `asOf` | timestamp | yes | Snapshot time. |
+| `stories` | array | yes | Ordered story cards. |
+
+`stories[]` entries:
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `officeId` | string | no | Office identifier. |
+| `startTime` | timestamp | yes | Story start. |
+| `endTime` | timestamp | yes | Story end. |
+| `updatedAt` | timestamp | yes | Story update time. |
+| `title` | string | no | Story title. |
+| `description` | string | no | Story description. |
+| `altText` | string | no | Image alternate text. |
+| `priority` | boolean | yes | Provider priority flag. |
+| `order` | integer | yes | Provider display order. |
+| `downloadUrl` | string | no | Story image URL. |
+
+## `weather.alert.v1`
+
+Payload type: `WeatherAlertRun`.
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `locationId` | string | no | Provider location identifier. |
+| `locationName` | string | no | Human location name. |
+| `asOf` | timestamp | yes | Snapshot time. |
+| `latitude` | number | no | Degrees. |
+| `longitude` | number | no | Degrees. |
+| `alerts` | array | yes | Active alerts. |
+
+`alerts[]` entries:
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `id` | string | yes | Provider-stable alert identifier. |
+| `event` | string | no | Alert event label. |
+| `headline` | string | no | Alert headline. |
+| `severity` | string | no | Provider severity. |
+| `urgency` | string | no | Provider urgency. |
+| `certainty` | string | no | Provider certainty. |
+| `status` | string | no | Alert status. |
+| `messageType` | string | no | Alert message type. |
+| `category` | string | no | Alert category. |
+| `response` | string | no | Recommended response. |
+| `description` | string | no | Alert description. |
+| `instruction` | string | no | Alert instruction. |
+| `sent` | timestamp | no | Provider sent time. |
+| `effective` | timestamp | no | Effective time. |
+| `onset` | timestamp | no | Onset time. |
+| `expires` | timestamp | no | Expiration time. |
+| `areaDescription` | string | no | Affected area description. |
+| `senderName` | string | no | Provider sender name. |
+| `references` | array | no | Related alerts. |
+
+`references[]` entries contain optional `id`, `identifier`, `sender`, and
+`sent` fields.
+
+## Compact Example
+
+```json
+{
+  "id": "NWSObservationKSTL:2026-06-10T12:00:00Z",
+  "kind": "observation",
+  "source": "NWSObservationKSTL",
+  "emitted_at": "2026-06-10T12:00:05Z",
+  "effective_at": "2026-06-10T12:00:00Z",
+  "schema": "weather.observation.v1",
+  "payload": {
+    "stationId": "KSTL",
+    "timestamp": "2026-06-10T12:00:00Z",
+    "conditionCode": 0,
+    "temperatureC": 22.5
+  }
+}
+```
--- a/docs/integrations/postgres.md
+++ b/docs/integrations/postgres.md
@@ -0,0 +1,403 @@
+# Postgres Integration
+
+This document is the canonical table contract for the optional `postgres` sink.
+It describes the schema created and written by weatherfeeder through feedkit's
+Postgres sink.
+
+Configure the sink as described in [configuration](../config.md#postgres).
+
+## Initialization And Writes
+
+At startup, each configured Postgres sink opens the database and runs
+`CREATE TABLE IF NOT EXISTS` for every weatherfeeder table, followed by
+`CREATE INDEX IF NOT EXISTS` for every configured index.
+
+This initialization creates missing tables and indexes only. It does not alter
+existing tables, migrate column definitions, drop old objects, or backfill data.
+Schema changes require operator-managed database migration.
+
+Events are mapped only for canonical weather schemas:
+
+- `weather.observation.v1`
+- `weather.forecast.v1`
+- `weather.forecast_discussion.v1`
+- `weather.weather_story.v1`
+- `weather.alert.v1`
+
+Unsupported schemas produce no writes for this sink. Mapped events are inserted
+transactionally. Inserts use ordinary `INSERT`; duplicate primary keys fail the
+write.
+
+## Shared Envelope Columns
+
+Parent tables store the feed event envelope:
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT` | no | `event.id` |
+| `event_kind` | `TEXT` | no | `event.kind` |
+| `event_source` | `TEXT` | no | `event.source` |
+| `event_schema` | `TEXT` | no | `event.schema` |
+| `event_emitted_at` | `TIMESTAMPTZ` | no | `event.emitted_at` |
+| `event_effective_at` | `TIMESTAMPTZ` | yes | `event.effective_at` |
+
+## Table Overview
+
+| Table | Primary key | Prune column |
+|---|---|---|
+| `observations` | `event_id` | `observed_at` |
+| `observation_present_weather` | `event_id`, `weather_index` | `observed_at` |
+| `forecasts` | `event_id` | `issued_at` |
+| `forecast_periods` | `run_event_id`, `period_index` | `issued_at` |
+| `forecast_discussions` | `event_id` | `issued_at` |
+| `forecast_discussion_key_messages` | `run_event_id`, `message_index` | `issued_at` |
+| `weather_story_runs` | `event_id` | `as_of` |
+| `weather_stories` | `run_event_id`, `story_index` | `as_of` |
+| `alert_runs` | `event_id` | `as_of` |
+| `alerts` | `run_event_id`, `alert_index` | `as_of` |
+| `alert_references` | `run_event_id`, `alert_index`, `reference_index` | `as_of` |
+
+## Table Contract
+
+### `observations`
+
+Primary key: `event_id`
+
+Prune column: `observed_at`
+
+Indexes:
+
+- `idx_wf_obs_station_observed_at` on `station_id`, `observed_at`
+- `idx_wf_obs_observed_at` on `observed_at`
+- `idx_wf_obs_condition_code` on `condition_code`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT` | no | `event.id` |
+| `event_kind` | `TEXT` | no | `event.kind` |
+| `event_source` | `TEXT` | no | `event.source` |
+| `event_schema` | `TEXT` | no | `event.schema` |
+| `event_emitted_at` | `TIMESTAMPTZ` | no | `event.emitted_at` |
+| `event_effective_at` | `TIMESTAMPTZ` | yes | `event.effective_at` |
+| `station_id` | `TEXT` | yes | `payload.stationId` |
+| `station_name` | `TEXT` | yes | `payload.stationName` |
+| `observed_at` | `TIMESTAMPTZ` | no | `payload.timestamp` |
+| `condition_code` | `INTEGER` | no | `payload.conditionCode` |
+| `is_day` | `BOOLEAN` | yes | `payload.isDay` |
+| `text_description` | `TEXT` | yes | `payload.textDescription` |
+| `temperature_c` | `DOUBLE PRECISION` | yes | `payload.temperatureC` |
+| `dewpoint_c` | `DOUBLE PRECISION` | yes | `payload.dewpointC` |
+| `wind_direction_degrees` | `DOUBLE PRECISION` | yes | `payload.windDirectionDegrees` |
+| `wind_speed_kmh` | `DOUBLE PRECISION` | yes | `payload.windSpeedKmh` |
+| `wind_gust_kmh` | `DOUBLE PRECISION` | yes | `payload.windGustKmh` |
+| `barometric_pressure_pa` | `DOUBLE PRECISION` | yes | `payload.barometricPressurePa` |
+| `visibility_meters` | `DOUBLE PRECISION` | yes | `payload.visibilityMeters` |
+| `relative_humidity_percent` | `DOUBLE PRECISION` | yes | `payload.relativeHumidityPercent` |
+| `apparent_temperature_c` | `DOUBLE PRECISION` | yes | `payload.apparentTemperatureC` |
+
+### `observation_present_weather`
+
+Primary key: `event_id`, `weather_index`
+
+Prune column: `observed_at`
+
+Foreign key: `event_id` references `observations(event_id)` with cascade delete.
+
+Index: `idx_wf_obs_present_observed_at` on `observed_at`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT REFERENCES observations(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `weather_index` | `INTEGER` | no | `payload.presentWeather[]` index. |
+| `observed_at` | `TIMESTAMPTZ` | no | `payload.timestamp` |
+| `raw_text` | `TEXT` | yes | Compact JSON text from `payload.presentWeather[].raw` |
+
+### `forecasts`
+
+Primary key: `event_id`
+
+Prune column: `issued_at`
+
+Indexes:
+
+- `idx_wf_fc_location_product_issued_at` on `location_id`, `product`, `issued_at`
+- `idx_wf_fc_issued_at` on `issued_at`
+- `idx_wf_fc_product_issued_at` on `product`, `issued_at`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT` | no | `event.id` |
+| `event_kind` | `TEXT` | no | `event.kind` |
+| `event_source` | `TEXT` | no | `event.source` |
+| `event_schema` | `TEXT` | no | `event.schema` |
+| `event_emitted_at` | `TIMESTAMPTZ` | no | `event.emitted_at` |
+| `event_effective_at` | `TIMESTAMPTZ` | yes | `event.effective_at` |
+| `location_id` | `TEXT` | yes | `payload.locationId` |
+| `location_name` | `TEXT` | yes | `payload.locationName` |
+| `issued_at` | `TIMESTAMPTZ` | no | `payload.issuedAt` |
+| `updated_at` | `TIMESTAMPTZ` | yes | `payload.updatedAt` |
+| `product` | `TEXT` | no | `payload.product` |
+| `latitude` | `DOUBLE PRECISION` | yes | `payload.latitude` |
+| `longitude` | `DOUBLE PRECISION` | yes | `payload.longitude` |
+| `elevation_meters` | `DOUBLE PRECISION` | yes | `payload.elevationMeters` |
+| `period_count` | `INTEGER` | no | `len(payload.periods)` |
+
+### `forecast_periods`
+
+Primary key: `run_event_id`, `period_index`
+
+Prune column: `issued_at`
+
+Foreign key: `run_event_id` references `forecasts(event_id)` with cascade delete.
+
+Indexes:
+
+- `idx_wf_fc_period_start_time` on `start_time`
+- `idx_wf_fc_period_end_time` on `end_time`
+- `idx_wf_fc_period_run_start` on `run_event_id`, `start_time`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `run_event_id` | `TEXT REFERENCES forecasts(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `period_index` | `INTEGER` | no | `payload.periods[]` index. |
+| `issued_at` | `TIMESTAMPTZ` | no | Parent `payload.issuedAt` |
+| `start_time` | `TIMESTAMPTZ` | no | `payload.periods[].startTime` |
+| `end_time` | `TIMESTAMPTZ` | no | `payload.periods[].endTime` |
+| `name` | `TEXT` | yes | `payload.periods[].name` |
+| `is_day` | `BOOLEAN` | yes | `payload.periods[].isDay` |
+| `condition_code` | `INTEGER` | yes | `payload.periods[].conditionCode` |
+| `text_description` | `TEXT` | yes | `payload.periods[].textDescription` |
+| `temperature_c` | `DOUBLE PRECISION` | yes | `payload.periods[].temperatureC` |
+| `temperature_c_min` | `DOUBLE PRECISION` | yes | `payload.periods[].temperatureCMin` |
+| `temperature_c_max` | `DOUBLE PRECISION` | yes | `payload.periods[].temperatureCMax` |
+| `dewpoint_c` | `DOUBLE PRECISION` | yes | `payload.periods[].dewpointC` |
+| `relative_humidity_percent` | `DOUBLE PRECISION` | yes | `payload.periods[].relativeHumidityPercent` |
+| `wind_direction_degrees` | `DOUBLE PRECISION` | yes | `payload.periods[].windDirectionDegrees` |
+| `wind_speed_kmh` | `DOUBLE PRECISION` | yes | `payload.periods[].windSpeedKmh` |
+| `wind_gust_kmh` | `DOUBLE PRECISION` | yes | `payload.periods[].windGustKmh` |
+| `barometric_pressure_pa` | `DOUBLE PRECISION` | yes | `payload.periods[].barometricPressurePa` |
+| `visibility_meters` | `DOUBLE PRECISION` | yes | `payload.periods[].visibilityMeters` |
+| `apparent_temperature_c` | `DOUBLE PRECISION` | yes | `payload.periods[].apparentTemperatureC` |
+| `cloud_cover_percent` | `DOUBLE PRECISION` | yes | `payload.periods[].cloudCoverPercent` |
+| `probability_of_precipitation_percent` | `DOUBLE PRECISION` | yes | `payload.periods[].probabilityOfPrecipitationPercent` |
+| `precipitation_amount_mm` | `DOUBLE PRECISION` | yes | `payload.periods[].precipitationAmountMm` |
+| `snowfall_depth_mm` | `DOUBLE PRECISION` | yes | `payload.periods[].snowfallDepthMm` |
+| `uv_index` | `DOUBLE PRECISION` | yes | `payload.periods[].uvIndex` |
+
+### `forecast_discussions`
+
+Primary key: `event_id`
+
+Prune column: `issued_at`
+
+Indexes:
+
+- `idx_wf_discussion_office_product_issued_at` on `office_id`, `product`, `issued_at`
+- `idx_wf_discussion_issued_at` on `issued_at`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT` | no | `event.id` |
+| `event_kind` | `TEXT` | no | `event.kind` |
+| `event_source` | `TEXT` | no | `event.source` |
+| `event_schema` | `TEXT` | no | `event.schema` |
+| `event_emitted_at` | `TIMESTAMPTZ` | no | `event.emitted_at` |
+| `event_effective_at` | `TIMESTAMPTZ` | yes | `event.effective_at` |
+| `office_id` | `TEXT` | yes | `payload.officeId` |
+| `office_name` | `TEXT` | yes | `payload.officeName` |
+| `issued_at` | `TIMESTAMPTZ` | no | `payload.issuedAt` |
+| `updated_at` | `TIMESTAMPTZ` | yes | `payload.updatedAt` |
+| `product` | `TEXT` | no | `payload.product` |
+| `short_term_qualifier` | `TEXT` | yes | `payload.shortTerm.qualifier` |
+| `short_term_issued_at` | `TIMESTAMPTZ` | yes | `payload.shortTerm.issuedAt` |
+| `short_term_text` | `TEXT` | yes | `payload.shortTerm.text` |
+| `long_term_qualifier` | `TEXT` | yes | `payload.longTerm.qualifier` |
+| `long_term_issued_at` | `TIMESTAMPTZ` | yes | `payload.longTerm.issuedAt` |
+| `long_term_text` | `TEXT` | yes | `payload.longTerm.text` |
+| `key_message_count` | `INTEGER` | no | `len(payload.keyMessages)` |
+
+### `forecast_discussion_key_messages`
+
+Primary key: `run_event_id`, `message_index`
+
+Prune column: `issued_at`
+
+Foreign key: `run_event_id` references `forecast_discussions(event_id)` with
+cascade delete.
+
+Index: `idx_wf_discussion_message_issued_at` on `issued_at`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `run_event_id` | `TEXT REFERENCES forecast_discussions(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `message_index` | `INTEGER` | no | `payload.keyMessages[]` index. |
+| `issued_at` | `TIMESTAMPTZ` | no | Parent `payload.issuedAt` |
+| `message_text` | `TEXT` | yes | `payload.keyMessages[]` value |
+
+### `weather_story_runs`
+
+Primary key: `event_id`
+
+Prune column: `as_of`
+
+Indexes:
+
+- `idx_wf_story_run_office_as_of` on `office_id`, `as_of`
+- `idx_wf_story_run_as_of` on `as_of`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT` | no | `event.id` |
+| `event_kind` | `TEXT` | no | `event.kind` |
+| `event_source` | `TEXT` | no | `event.source` |
+| `event_schema` | `TEXT` | no | `event.schema` |
+| `event_emitted_at` | `TIMESTAMPTZ` | no | `event.emitted_at` |
+| `event_effective_at` | `TIMESTAMPTZ` | yes | `event.effective_at` |
+| `office_id` | `TEXT` | yes | `payload.officeId` |
+| `as_of` | `TIMESTAMPTZ` | no | `payload.asOf` |
+| `story_count` | `INTEGER` | no | `len(payload.stories)` |
+
+### `weather_stories`
+
+Primary key: `run_event_id`, `story_index`
+
+Prune column: `as_of`
+
+Foreign key: `run_event_id` references `weather_story_runs(event_id)` with
+cascade delete.
+
+Indexes:
+
+- `idx_wf_stories_start_time` on `start_time`
+- `idx_wf_stories_end_time` on `end_time`
+- `idx_wf_stories_updated_at` on `updated_at`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `run_event_id` | `TEXT REFERENCES weather_story_runs(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `story_index` | `INTEGER` | no | `payload.stories[]` index. |
+| `as_of` | `TIMESTAMPTZ` | no | Parent `payload.asOf` |
+| `office_id` | `TEXT` | yes | `payload.stories[].officeId` |
+| `start_time` | `TIMESTAMPTZ` | no | `payload.stories[].startTime` |
+| `end_time` | `TIMESTAMPTZ` | no | `payload.stories[].endTime` |
+| `updated_at` | `TIMESTAMPTZ` | no | `payload.stories[].updatedAt` |
+| `title` | `TEXT` | yes | `payload.stories[].title` |
+| `description` | `TEXT` | yes | `payload.stories[].description` |
+| `alt_text` | `TEXT` | yes | `payload.stories[].altText` |
+| `priority` | `BOOLEAN` | no | `payload.stories[].priority` |
+| `story_order` | `INTEGER` | no | `payload.stories[].order` |
+| `download_url` | `TEXT` | yes | `payload.stories[].downloadUrl` |
+
+### `alert_runs`
+
+Primary key: `event_id`
+
+Prune column: `as_of`
+
+Indexes:
+
+- `idx_wf_alert_run_location_as_of` on `location_id`, `as_of`
+- `idx_wf_alert_run_as_of` on `as_of`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `event_id` | `TEXT` | no | `event.id` |
+| `event_kind` | `TEXT` | no | `event.kind` |
+| `event_source` | `TEXT` | no | `event.source` |
+| `event_schema` | `TEXT` | no | `event.schema` |
+| `event_emitted_at` | `TIMESTAMPTZ` | no | `event.emitted_at` |
+| `event_effective_at` | `TIMESTAMPTZ` | yes | `event.effective_at` |
+| `location_id` | `TEXT` | yes | `payload.locationId` |
+| `location_name` | `TEXT` | yes | `payload.locationName` |
+| `as_of` | `TIMESTAMPTZ` | no | `payload.asOf` |
+| `latitude` | `DOUBLE PRECISION` | yes | `payload.latitude` |
+| `longitude` | `DOUBLE PRECISION` | yes | `payload.longitude` |
+| `alert_count` | `INTEGER` | no | `len(payload.alerts)` |
+
+### `alerts`
+
+Primary key: `run_event_id`, `alert_index`
+
+Prune column: `as_of`
+
+Foreign key: `run_event_id` references `alert_runs(event_id)` with cascade
+delete.
+
+Indexes:
+
+- `idx_wf_alerts_alert_id` on `alert_id`
+- `idx_wf_alerts_severity_expires` on `severity`, `expires`
+- `idx_wf_alerts_as_of` on `as_of`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `run_event_id` | `TEXT REFERENCES alert_runs(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `alert_index` | `INTEGER` | no | `payload.alerts[]` index. |
+| `as_of` | `TIMESTAMPTZ` | no | Parent `payload.asOf` |
+| `alert_id` | `TEXT` | no | `payload.alerts[].id` |
+| `event` | `TEXT` | yes | `payload.alerts[].event` |
+| `headline` | `TEXT` | yes | `payload.alerts[].headline` |
+| `severity` | `TEXT` | yes | `payload.alerts[].severity` |
+| `urgency` | `TEXT` | yes | `payload.alerts[].urgency` |
+| `certainty` | `TEXT` | yes | `payload.alerts[].certainty` |
+| `status` | `TEXT` | yes | `payload.alerts[].status` |
+| `message_type` | `TEXT` | yes | `payload.alerts[].messageType` |
+| `category` | `TEXT` | yes | `payload.alerts[].category` |
+| `response` | `TEXT` | yes | `payload.alerts[].response` |
+| `description` | `TEXT` | yes | `payload.alerts[].description` |
+| `instruction` | `TEXT` | yes | `payload.alerts[].instruction` |
+| `sent` | `TIMESTAMPTZ` | yes | `payload.alerts[].sent` |
+| `effective` | `TIMESTAMPTZ` | yes | `payload.alerts[].effective` |
+| `onset` | `TIMESTAMPTZ` | yes | `payload.alerts[].onset` |
+| `expires` | `TIMESTAMPTZ` | yes | `payload.alerts[].expires` |
+| `area_description` | `TEXT` | yes | `payload.alerts[].areaDescription` |
+| `sender_name` | `TEXT` | yes | `payload.alerts[].senderName` |
+| `reference_count` | `INTEGER` | no | `len(payload.alerts[].references)` |
+
+### `alert_references`
+
+Primary key: `run_event_id`, `alert_index`, `reference_index`
+
+Prune column: `as_of`
+
+Foreign key: `run_event_id` references `alert_runs(event_id)` with cascade
+delete.
+
+Indexes:
+
+- `idx_wf_alert_refs_as_of` on `as_of`
+- `idx_wf_alert_refs_sent` on `sent`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `run_event_id` | `TEXT REFERENCES alert_runs(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `alert_index` | `INTEGER` | no | Parent alert index. |
+| `reference_index` | `INTEGER` | no | `payload.alerts[].references[]` index. |
+| `as_of` | `TIMESTAMPTZ` | no | Parent `payload.asOf` |
+| `id` | `TEXT` | yes | `payload.alerts[].references[].id` |
+| `identifier` | `TEXT` | yes | `payload.alerts[].references[].identifier` |
+| `sender` | `TEXT` | yes | `payload.alerts[].references[].sender` |
+| `sent` | `TIMESTAMPTZ` | yes | `payload.alerts[].references[].sent` |
+
+## Retention
+
+When sink param `prune` is set, every successful write transaction deletes rows
+older than `now - prune` from every table using that table's prune column.
+
+The sink also exposes manual prune helpers in code, but the `weatherfeeder`
+binary does not provide CLI commands for them.
+
+## Reconstructing Canonical Payloads
+
+- `WeatherObservation`: read `observations`, then join
+  `observation_present_weather` by `event_id` ordered by `weather_index`.
+- `WeatherForecastRun`: read `forecasts`, then join `forecast_periods` by
+  `run_event_id` ordered by `period_index`.
+- `WeatherForecastDiscussion`: read `forecast_discussions`, then join
+  `forecast_discussion_key_messages` by `run_event_id` ordered by
+  `message_index`.
+- `WeatherStoryRun`: read `weather_story_runs`, then join `weather_stories` by
+  `run_event_id` ordered by `story_index`.
+- `WeatherAlertRun`: read `alert_runs`, join `alerts` by `run_event_id` ordered
+  by `alert_index`, then join `alert_references` by `run_event_id` and
+  `alert_index` ordered by `reference_index`.