fix(overview): 成本分析“提供商用量”展示提供商名称而非 ID

成本分析页「提供商用量」表格的“提供商”列渲染接口返回的 label，而后端在
UsageAnalyticsView::Breakdown + group_by=provider 时把 label 直接写成了
分组键 provider_id，页面因此显示 provider-1 这类内部编号。

- label 改为按「providers.name → 使用记录里的 provider_name 快照 →
  provider_id 兜底」解析；分组键 id 仍是 provider_id，“查看”链接、按
  提供商匹配的已登记支出列、时间序列图表均不受影响。
- 空白与 unknown/unknow/pending 不当作名称；provider_id 为空时保持空标签，
  前端沿用“未归属提供商”文案。
- 内存仓储（无数据库部署与测试）同步同一口径，避免两种实现表现不一致。
- 新增用例：内存侧 overview_provider_breakdown_labels_rows_with_provider_name，
  SQL 侧 provider_breakdown_labels_resolve_catalog_name_before_recorded_name_and_id。

crates/aether-data/adapters/postgres/src/usage/analytics.rs

@@ -40,6 +40,20 @@ count(*) FILTER (WHERE actor_user_id IS NOT NULL)::bigint AS trusted_attribution
 count(*) FILTER (WHERE status = 'failed' AND failure_origin IS NOT NULL AND failure_origin <> 'unknown')::bigint AS classified_failure_count
 "#;
 
+// 提供商分组的展示标签：分组键仍然是 provider_id，但页面上要展示“提供商名称”。
+// 解析顺序与用量审计聚合保持一致：提供商目录中的当前名称 → 使用记录里的名称快照 → 原始 provider_id。
+// 'unknown' / 'unknow' / 'pending' 是历史占位值，不能当成名称展示；
+// provider_id 为空说明这条记录本身无法归属，保持空标签让前端显示“未归属提供商”。
+pub(super) const ANALYTICS_PROVIDER_LABEL_SQL: &str = r#"COALESCE(
+        NULLIF(BTRIM(provider_catalog.name), ''),
+        CASE
+          WHEN page.group_id IS NULL THEN NULL
+          WHEN lower(BTRIM(COALESCE(page.provider_name, ''))) IN ('', 'unknown', 'unknow', 'pending') THEN NULL
+          ELSE BTRIM(page.provider_name)
+        END,
+        page.group_id::text
+      )"#;
+
 pub(super) fn dashboard_total_metrics_sql() -> &'static str {
     r#"count(*)::bigint AS request_count,
 COALESCE(sum(total_tokens),0)::bigint AS total_tokens,
@@ -359,6 +373,10 @@ impl SqlxUsageReadRepository {
                 | UsageAnalyticsView::DashboardCharts
                 | UsageAnalyticsView::Breakdown => {
                     let timeseries = query.view != UsageAnalyticsView::Breakdown;
+                    // 提供商分组的行需要额外带出名称快照，并在最外层关联提供商目录解析展示名。
+                    // 只有明细分组（Breakdown）才需要，时间序列仍然直接用 provider_id 作为标签。
+                    let provider_labels =
+                        !timeseries && query.group_by == UsageAnalyticsGroupBy::Provider;
                     let group = if timeseries {
                         let granularity = match query.granularity {
                             UsageAnalyticsGranularity::Hour => "hour",
@@ -393,6 +411,11 @@ impl SqlxUsageReadRepository {
                         .push(", grouped AS (SELECT ")
                         .push(group)
                         .push(" AS group_id, ")
+                        .push(if provider_labels {
+                            "max(provider_name) AS provider_name, "
+                        } else {
+                            ""
+                        })
                         .push(&metrics_sql)
                         .push(if timeseries {
                             " FROM dated GROUP BY "
@@ -414,9 +437,22 @@ impl SqlxUsageReadRepository {
                             .push(", group_id ASC NULLS LAST");
                     }
                     builder.push(" LIMIT ").push_bind(if timeseries { 10_001 } else { i64::from(query.limit) }).push(" OFFSET ").push_bind(if timeseries { 0 } else { query.offset as i64 })
-                        .push(") SELECT (SELECT count(*) FROM grouped) AS total, COALESCE(jsonb_agg(jsonb_build_object('id', group_id::text, 'label', group_id::text, 'bucket_start', ")
+                        .push(") SELECT (SELECT count(*) FROM grouped) AS total, COALESCE(jsonb_agg(jsonb_build_object('id', page.group_id::text, 'label', ")
+                        .push(if provider_labels { ANALYTICS_PROVIDER_LABEL_SQL } else { "page.group_id::text" })
+                        .push(", 'bucket_start', ")
                         .push(if timeseries { "to_char(group_id AT TIME ZONE 'UTC', 'YYYY-MM-DD\"T\"HH24:MI:SS\"Z\"')" } else { "NULL" })
-                        .push(", 'metrics', to_jsonb(page) - 'group_id')), '[]'::jsonb) AS items FROM page");
+                        .push(", 'metrics', to_jsonb(page) - 'group_id'")
+                        .push(if provider_labels {
+                            " - 'provider_name'"
+                        } else {
+                            ""
+                        })
+                        .push(")), '[]'::jsonb) AS items FROM page")
+                        .push(if provider_labels {
+                            " LEFT JOIN public.providers AS provider_catalog ON provider_catalog.id = page.group_id"
+                        } else {
+                            ""
+                        });
                     let row = builder
                         .build()
                         .fetch_one(&mut *tx)

crates/aether-data/adapters/postgres/src/usage/tests.rs

@@ -5414,3 +5414,21 @@ fn attach_usage_settlement_pricing_snapshot_metadata_adds_missing_values_without
         })
     );
 }
+
+#[test]
+fn provider_breakdown_labels_resolve_catalog_name_before_recorded_name_and_id() {
+    let sql = super::analytics::ANALYTICS_PROVIDER_LABEL_SQL;
+    let catalog = sql
+        .find("provider_catalog.name")
+        .expect("provider catalog name");
+    let recorded = sql
+        .find("page.provider_name")
+        .expect("recorded provider name snapshot");
+    let raw_id = sql
+        .find("page.group_id::text")
+        .expect("raw provider id fallback");
+    // 成本分析“提供商”列的解析顺序：目录名称 → 使用记录里的名称快照 → 原始 provider_id。
+    assert!(catalog < recorded && recorded < raw_id);
+    // 历史占位值不能被当成提供商名称展示。
+    assert!(sql.contains("'unknown', 'unknow', 'pending'"));
+}

crates/aether-data/runtime/src/repository/usage/memory/analytics.rs

@@ -59,6 +59,27 @@ fn available(row: &StoredRequestUsageAudit, key: &str) -> bool {
         .and_then(serde_json::Value::as_bool)
         != Some(false)
 }
+/// 提供商明细分组的展示名：分组键依旧是 provider_id（保证与 PostgreSQL 实现一致），
+/// 只是把展示标签换成使用记录里的提供商名称快照，名称缺失或为历史占位值时回退到 provider_id。
+/// provider_id 为空说明无法归属，保持 None 让前端显示“未归属提供商”。
+fn provider_display_label(
+    rows: &[&StoredRequestUsageAudit],
+    group_id: Option<&str>,
+) -> Option<String> {
+    let id = group_id?;
+    rows.iter()
+        .map(|row| row.provider_name.trim())
+        .filter(|name| {
+            !name.is_empty()
+                && !matches!(
+                    name.to_ascii_lowercase().as_str(),
+                    "unknown" | "unknow" | "pending"
+                )
+        })
+        .max()
+        .map(str::to_owned)
+        .or_else(|| Some(id.to_owned()))
+}
 // The legacy audit contract stores epoch seconds despite its historical field name.
 fn usage_started_ms(row: &StoredRequestUsageAudit) -> u64 {
     row.created_at_unix_ms.saturating_mul(1000)
@@ -656,10 +677,17 @@ impl InMemoryUsageReadRepository {
                     };
                     groups.entry(group).or_default().push(row);
                 }
+                // 提供商明细分组需要单独解析展示名，其余分组仍然用分组键本身作为标签。
+                let provider_breakdown = query.view == UsageAnalyticsView::Breakdown
+                    && query.group_by == UsageAnalyticsGroupBy::Provider;
                 let mut grouped = groups
                     .into_iter()
                     .map(|(id, rows)| UsageAnalyticsRow {
-                        label: id.clone(),
+                        label: if provider_breakdown {
+                            provider_display_label(&rows, id.as_deref())
+                        } else {
+                            id.clone()
+                        },
                         bucket_start: (query.view != UsageAnalyticsView::Breakdown)
                             .then(|| id.clone())
                             .flatten(),

crates/aether-data/runtime/src/repository/usage/memory/tests.rs

@@ -160,6 +160,52 @@ async fn overview_model_performance_merges_provider_samples_without_pagination()
     assert_eq!(filtered.model_rows, vec![result.model_rows[0].clone()]);
 }
 
+#[tokio::test]
+async fn overview_provider_breakdown_labels_rows_with_provider_name() {
+    use aether_data_contracts::repository::usage::*;
+    let at = chrono::DateTime::parse_from_rfc3339("2026-09-12T10:05:00Z").unwrap();
+    // 同一 provider_id 的两条记录，展示标签应解析成提供商名称而不是 provider_id。
+    let mut first = sample_usage("provider-label-1", at.timestamp());
+    first.provider_id = Some("provider-1".into());
+    first.provider_name = "Provider One".into();
+    let mut second = sample_usage("provider-label-2", at.timestamp());
+    second.provider_id = Some("provider-1".into());
+    second.provider_name = "Provider One".into();
+    // 名称为历史占位值时回退到 provider_id。
+    let mut unnamed = sample_usage("provider-label-3", at.timestamp());
+    unnamed.provider_id = Some("provider-2".into());
+    unnamed.provider_name = "unknown".into();
+    // provider_id 为空说明无法归属，标签保持为空，由前端显示“未归属提供商”。
+    let mut unattributed = sample_usage("provider-label-4", at.timestamp());
+    unattributed.provider_id = None;
+    unattributed.provider_name = "legacy".into();
+    let repo = InMemoryUsageReadRepository::seed([first, second, unnamed, unattributed]);
+    let result = repo
+        .query_usage_analytics(&UsageAnalyticsQuery {
+            from_unix_ms: (at - chrono::Duration::minutes(5)).timestamp_millis() as u64,
+            to_unix_ms: (at + chrono::Duration::minutes(55)).timestamp_millis() as u64,
+            timezone: "UTC".into(),
+            view: UsageAnalyticsView::Breakdown,
+            group_by: UsageAnalyticsGroupBy::Provider,
+            limit: 25,
+            descending: true,
+            ..Default::default()
+        })
+        .await
+        .unwrap();
+    assert_eq!(result.total, 3);
+    let rows = result
+        .rows
+        .iter()
+        .map(|row| (row.id.as_deref(), row.label.as_deref()))
+        .collect::<Vec<_>>();
+    assert!(rows.contains(&(Some("provider-1"), Some("Provider One"))));
+    assert!(rows.contains(&(Some("provider-2"), Some("provider-2"))));
+    assert!(rows.contains(&(None, None)));
+    // 明细分组不填充 bucket_start。
+    assert!(result.rows.iter().all(|row| row.bucket_start.is_none()));
+}
+
 #[tokio::test]
 async fn overview_memory_chart_hour_buckets_are_utc_in_half_hour_zones() {
     use aether_data_contracts::repository::usage::*;