Document current outlook schema behavior

docs/integrations/events.md

@@ -37,7 +37,7 @@ Canonical schemas emitted after normalization:
 | `forecast_discussion` | `weather.forecast_discussion.v1` | `WeatherForecastDiscussion` |
 | `weather_story` | `weather.weather_story.v1` | `WeatherStoryRun` |
 | `alert` | `weather.alert.v1` | `WeatherAlertRun` |
-| `outlook` | `weather.outlook.v1` | `WeatherOutlookRun` |
+| `outlook` | `weather.outlook.v2` | `WeatherOutlookRun` |
 
 Raw upstream schemas emitted by current sources:
 
@@ -220,13 +220,15 @@ Payload type: `WeatherAlertRun`.
 `references[]` entries contain optional `id`, `identifier`, `sender`, and
 `sent` fields.
 
-## `weather.outlook.v1`
+## `weather.outlook.v2`
 
 Payload type: `WeatherOutlookRun`.
 
 The current producer is the SPC convective outlook normalizer. It emits Day 1-3
-convective outlook polygons for categorical, tornado, hail, and wind products.
-All timestamps are UTC.
+convective outlook snapshots for categorical, tornado, hail, and wind products
+that apply to the configured forecast point. Raw SPC bundles remain complete;
+canonical outlook payloads are filtered to local polygons. All timestamps are
+UTC.
 
 | Field | Type | Required | Notes |
 |---|---|:---:|---|
@@ -236,7 +238,8 @@ All timestamps are UTC.
 | `longitude` | number | no | Configured point longitude in decimal degrees. |
 | `asOf` | timestamp | yes | Snapshot time. For SPC, this is the latest outlook issue time when available. |
 | `issuedAt` | timestamp | no | Latest issue time across outlook features when any feature exists. |
-| `outlooks` | array | yes | Ordered outlook polygons. |
+| `outlooks` | array | yes | Ordered outlook polygons containing the configured point. |
+| `discussions` | array | yes | Run-level day discussions for retained outlook days. |
 
 `outlooks[]` entries:
 
@@ -255,19 +258,53 @@ All timestamps are UTC.
 | `issuedAt` | timestamp | yes | Feature issue time. |
 | `expiresAt` | timestamp | yes | Expiration time; currently equal to `validTo`. |
 | `forecaster` | string | no | SPC forecaster text, when present. |
-| `headline` | string | no | Matching Day 1-3 print-page product title. |
-| `summary` | string | no | Text from the print-page `...SUMMARY...` section. |
-| `discussion` | string | no | Cleaned full print-page product text. |
 | `sourceUrl` | string | no | GeoJSON product URL for this outlook feature. |
 | `imageUrl` | string | no | Reserved for provider image URLs; currently empty. |
-| `containsLocation` | boolean | yes | Whether the configured point is inside or on the boundary of the polygon. |
+| `containsLocation` | boolean | yes | Always `true` for emitted v2 outlooks. |
 | `geometry` | object | yes | Compact GeoJSON `Polygon` or `MultiPolygon` geometry. |
 
 `geometry` preserves the SPC feature geometry as compact GeoJSON using
 `[longitude, latitude]` coordinate order. `containsLocation` is computed with
 that geometry and the configured source `latitude`/`longitude`; boundary points
-count as contained. All outlook polygons are emitted, including polygons that do
-not contain the configured point.
+count as contained. Polygons that do not contain the configured point are not
+included in canonical v2 payloads.
+
+`discussions[]` entries:
+
+| Field | Type | Required | Notes |
+|---|---|:---:|---|
+| `day` | integer | yes | SPC outlook day, currently `1`, `2`, or `3`. |
+| `headline` | string | no | Matching Day 1-3 print-page product title. |
+| `summary` | string | no | Text from the print-page `...SUMMARY...` section. |
+| `discussion` | string | no | Cleaned full print-page product text. |
+| `updatedAt` | timestamp | no | Print-page update time, when present. |
+
+When no SPC polygons apply locally, the run is still emitted with `outlooks: []`
+and `discussions: []`. Discussions are included only for days represented by at
+least one retained outlook, and multiple retained outlook types for the same day
+share one discussion entry.
+
+### SPC Outlook Supersession
+
+Consumers should prefer latest-run semantics for current conditions: read the
+latest `WeatherOutlookRun` for the configured location and use its `outlooks`
+and `discussions` arrays together.
+
+Historical SQL consumers that collapse older rows should identify superseded
+outlooks by `provider`, `product`, `outlookType`, `validFrom`, and `validTo`,
+then keep rows with the greatest `issuedAt`. `day` and `label` are not identity
+fields. When multiple retained polygons share that latest `issuedAt`, preserve
+the full group.
+
+## Legacy `weather.outlook.v1`
+
+`weather.outlook.v1` is a historical canonical schema retained as a standards
+constant for older data and consumers. Current SPC normalization emits
+`weather.outlook.v2`.
+
+The v1 payload used `WeatherOutlookRun` and placed `headline`, `summary`, and
+`discussion` on each `outlooks[]` polygon. It also represented the pre-v2 SPC
+canonical behavior, where national polygons were preserved in canonical output.
 
 ## Compact Example
 

docs/integrations/postgres.md

@@ -23,7 +23,7 @@ Events are mapped only for canonical weather schemas:
 - `weather.forecast_discussion.v1`
 - `weather.weather_story.v1`
 - `weather.alert.v1`
-- `weather.outlook.v1`
+- `weather.outlook.v2`
 
 Unsupported schemas produce no writes for this sink. Mapped events are inserted
 transactionally. Inserts use ordinary `INSERT`; duplicate primary keys fail the
@@ -59,6 +59,7 @@ Parent tables store the feed event envelope:
 | `alert_references` | `run_event_id`, `alert_index`, `reference_index` | `as_of` |
 | `outlook_runs` | `event_id` | `as_of` |
 | `outlooks` | `run_event_id`, `outlook_index` | `as_of` |
+| `outlook_discussions` | `run_event_id`, `discussion_index` | `as_of` |
 
 ## Table Contract
 
@@ -408,6 +409,7 @@ Indexes:
 | `as_of` | `TIMESTAMPTZ` | no | `payload.asOf` |
 | `issued_at` | `TIMESTAMPTZ` | yes | `payload.issuedAt` |
 | `outlook_count` | `INTEGER` | no | `len(payload.outlooks)` |
+| `discussion_count` | `INTEGER` | no | `len(payload.discussions)` |
 
 ### `outlooks`
 
@@ -442,14 +444,36 @@ Indexes:
 | `issued_at` | `TIMESTAMPTZ` | no | `payload.outlooks[].issuedAt` |
 | `expires_at` | `TIMESTAMPTZ` | no | `payload.outlooks[].expiresAt` |
 | `forecaster` | `TEXT` | yes | `payload.outlooks[].forecaster` |
-| `headline` | `TEXT` | yes | `payload.outlooks[].headline` |
-| `summary` | `TEXT` | yes | `payload.outlooks[].summary` |
-| `discussion` | `TEXT` | yes | `payload.outlooks[].discussion` |
 | `source_url` | `TEXT` | yes | `payload.outlooks[].sourceUrl` |
 | `image_url` | `TEXT` | yes | `payload.outlooks[].imageUrl` |
 | `contains_location` | `BOOLEAN` | no | `payload.outlooks[].containsLocation` |
 | `geometry_json` | `TEXT` | no | Compact JSON from `payload.outlooks[].geometry` |
 
+### `outlook_discussions`
+
+Primary key: `run_event_id`, `discussion_index`
+
+Prune column: `as_of`
+
+Foreign key: `run_event_id` references `outlook_runs(event_id)` with cascade
+delete.
+
+Indexes:
+
+- `idx_wf_outlook_discussions_day_as_of` on `day`, `as_of`
+- unique `idx_wf_outlook_discussions_run_day` on `run_event_id`, `day`
+
+| Column | Type | Null | Source |
+|---|---|:---:|---|
+| `run_event_id` | `TEXT REFERENCES outlook_runs(event_id) ON DELETE CASCADE` | no | Parent event ID. |
+| `discussion_index` | `INTEGER` | no | `payload.discussions[]` index. |
+| `as_of` | `TIMESTAMPTZ` | no | Parent `payload.asOf` |
+| `day` | `INTEGER` | no | `payload.discussions[].day` |
+| `headline` | `TEXT` | yes | `payload.discussions[].headline` |
+| `summary` | `TEXT` | yes | `payload.discussions[].summary` |
+| `discussion` | `TEXT` | yes | `payload.discussions[].discussion` |
+| `updated_at` | `TIMESTAMPTZ` | yes | `payload.discussions[].updatedAt` |
+
 ## Retention
 
 When sink param `prune` is set, every successful write transaction deletes rows
@@ -472,5 +496,6 @@ binary does not provide CLI commands for them.
 - `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`.
-- `WeatherOutlookRun`: read `outlook_runs`, then join `outlooks` by
-  `run_event_id` ordered by `outlook_index`.
+- `WeatherOutlookRun`: read `outlook_runs`, join `outlooks` by `run_event_id`
+  ordered by `outlook_index`, then join `outlook_discussions` by
+  `run_event_id` ordered by `discussion_index`.

docs/integrations/spc.md

@@ -13,7 +13,7 @@ events are documented in [event wire contract](events.md).
 
 | Driver | Kind | Raw schema | Canonical schema |
 | --- | --- | --- | --- |
-| `spc_convective_outlook` | `outlook` | `raw.spc.convective_outlook.v1` | `weather.outlook.v1` |
+| `spc_convective_outlook` | `outlook` | `raw.spc.convective_outlook.v1` | `weather.outlook.v2` |
 
 ## Config Requirements
 
@@ -52,8 +52,8 @@ It also fetches three required print pages:
 - Day 3 convective outlook print page
 
 GeoJSON products are authoritative for outlook polygons, valid windows, issue
-times, labels, and severity rank. Print pages are authoritative for headline,
-summary, and discussion text.
+times, labels, and severity rank. Print pages are authoritative for run-level
+day discussion headline, summary, and discussion text.
 
 ## Accept Headers
 
@@ -99,23 +99,33 @@ Raw source `effective_at` prefers:
 4. fetch time.
 
 The normalizer sets canonical `asOf` and normalized event `effective_at` from
-the latest valid outlook feature `issuedAt`, with fallback to print-page update
-time and then the incoming event envelope.
+the latest valid GeoJSON issue time across the complete raw bundle, including
+empty no-risk placeholders, with fallback to print-page update time and then the
+incoming event envelope.
 
 ## Mapping Notes
 
-Each GeoJSON feature becomes one canonical outlook. Products are ordered by day,
-then categorical, tornado, hail, and wind. Feature order is preserved within
-each product.
+The raw source fetches and envelopes the complete SPC bundle. The normalizer
+decodes every configured GeoJSON product, skips empty no-risk
+`GeometryCollection` placeholders, and emits canonical outlooks only when the
+configured point is inside or on the boundary of a real feature. Products are
+ordered by day, then categorical, tornado, hail, and wind. Retained feature order
+is preserved within each product.
 
 The normalizer computes `containsLocation` with the configured latitude and
 longitude against compact GeoJSON `Polygon` or `MultiPolygon` geometry.
 Coordinates use GeoJSON order, `[longitude, latitude]`, and boundary points
 count as contained.
 
-All outlook polygons are preserved, including polygons that do not contain the
-configured point. Matching print-page headline, summary, and discussion text is
-attached to every outlook for the same day.
+Canonical outlook runs are emitted even when no polygons apply locally. In that
+case the payload contains empty `outlooks` and `discussions` arrays.
+
+Print-page prose is represented as run-level day discussions. Discussions are
+included only for days represented by at least one retained outlook. Multiple
+retained outlook types for the same day share one discussion entry.
+
+For downstream current-state and historical supersession guidance, see the
+[event wire contract](events.md#spc-outlook-supersession).
 
 ## Failure Behavior