agent-usage-all-in-one 0.9.0 → 0.10.0

package/README.md

@@ -17,8 +17,6 @@ It offers advice but never switches agents automatically.
 
 ## Compared
 
-![Category comparison: provider UIs, token exporters, FinOps dashboards, and Agent Usage](docs/design/compare-sheet.png)
-
 Honest category axes and when to prefer Agent Usage vs provider UIs, exporters, or FinOps dashboards: [docs/comparison.md](docs/comparison.md).
 
 Sourced named competitors (ccusage, CodeBurn, Token Monitor, TokenTracker): [docs/named-competitors.md](docs/named-competitors.md).
@@ -35,7 +33,8 @@ The dashboard has two primary tabs:
   selected window, metric, and currency sit in the same sticky header as the
   primary tabs. The summary board leads with the headline amount beside a
   GitHub-style last-year usage wall of headline-included recorded Tokens, then
-  charts Provider share and the interactive daily trend, and shows model
+  charts each known model separately by agent and billing domain on the share
+  pie and the interactive daily trend, and shows model
   rankings with visual share bars and the public API retail equivalent of
   eligible token usage. The year wall is independent of the selected 24-hour,
   7-day, or 30-day window.
@@ -51,8 +50,6 @@ unclassified or unpriced instead of being guessed or displayed as zero.
 Grok Build/SuperGrok and xAI API are independent billing domains. Their quotas,
 tokens, and costs are never added together.
 
-![Cost honesty: actual charges, estimates, subscriptions, and API retail equivalent stay separate](docs/design/cost-semantics-4.png)
-
 ## Fast, progressive startup
 
 The loopback web service starts before connector discovery or data processing.
@@ -60,11 +57,13 @@ Cached results are available immediately while discovery, provider usage, model
 pricing, and retention run as independent background modules. Each dashboard tab
 shows only its own update indicator and completed sections remain usable.
 
-Transcript scans use a persistent, path-redacted file index. Historical retail
-pricing is recalculated only when the pricing catalog version changes. SQLite
-time/provider/model indexes and retention compaction run in a worker after
-provider collection, while price backfill processes bounded pages. Settings
-includes an explicitly confirmed
+Transcript scans use a persistent, path-redacted file index. Refresh derives API
+retail equivalent only for new or changed Token observations. Historical retail
+pricing backfills remaining unpriced observations when the pricing catalog
+version changes. SQLite time/provider/model indexes and retention compaction run
+in a worker after provider collection only when those jobs have work. Price
+backfill processes bounded pages. Token and cost refresh bars follow price
+derivation, not connector collection or retention. Settings includes an explicitly confirmed
 **Hard rebuild all data** action for troubleshooting. It ignores these caches, can
 use substantial resources, and may take a long time without blocking the web UI.
 
@@ -89,8 +88,6 @@ oversight — see [ADR 017](docs/adr/017-macos-node24-npm-runtime.md) and the
 [platform roadmap](docs/platform-roadmap.md). Linux and Homebrew / DMG are not
 shipped.
 
-![Quick start: install and open the local dashboard in three steps](docs/design/install-3-steps.png)
-
 ```bash
 npx agent-usage-all-in-one
 ```
@@ -148,9 +145,7 @@ evidence remain visibly distinct.
 
 ## First launch / empty state
 
-Until connectors find local client data, the dashboard shows an honest empty state — connect or use an installed agent first:
-
-![Empty state when no agent data is connected yet](docs/design/empty-state-unconnected.png)
+Until connectors find local client data, the dashboard shows an honest empty state — connect or use an installed agent first.
 
 ## Credentials and privacy
 
@@ -188,8 +183,6 @@ Linux, Homebrew, and DMG stay off the product until the conditions in
 
 > **VISION only — not shipped.** Menu bar / always-available glance is tracked in [issue #19](https://github.com/Blackman99/agent-usage-all-in-one/issues/19). Do not treat the mock as a current feature.
 
-![VISION (not shipped): menu bar glance concept — see issue #19](docs/design/menubar-vision.png)
-
 ## License and community
 
 MIT — see [LICENSE](LICENSE). Also see [CONTRIBUTING.md](CONTRIBUTING.md),

package/README.zh-CN.md

@@ -16,8 +16,6 @@ API 对等费用、历史与诊断。它只提供建议，不会自动切换 Age
 
 ## 对比说明
 
-![类别对比：各平台官网、Token 导出工具、FinOps 看板与 Agent Usage](docs/design/compare-sheet.png)
-
 按类别对比何时更适合用 Agent Usage，而非各平台官网、导出工具或通用 FinOps 看板：[docs/comparison.md](docs/comparison.md)。
 
 具名竞品对照（ccusage、CodeBurn、Token Monitor、TokenTracker）：[docs/named-competitors.md](docs/named-competitors.md)。
@@ -29,7 +27,7 @@ API 对等费用、历史与诊断。它只提供建议，不会自动切换 Age
 页面只有两个主要标签页：
 
 - **Agent 用量**：保留各平台原生的 5 小时、周、月、All models 和 Fable only 等额度标签与刷新时间。
-- **Token 与模型费用**：支持 24 小时、7 天和 30 天范围。所选窗口、指标与币种和主标签页共用同一条吸顶导航。汇总卡片左侧是金额或 Token 头条，右侧是过去一年的 GitHub 风格用量墙（只统计计入摘要的已记录 Token），随后用图表展示 Provider 占比和可交互的每日趋势，再以占比条呈现模型排行及符合条件的 Token 按公开 API 价格计算出的对等费用。用量墙与所选的 24 小时 / 7 天 / 30 天窗口相互独立。
+- **Token 与模型费用**：支持 24 小时、7 天和 30 天范围。所选窗口、指标与币种和主标签页共用同一条吸顶导航。汇总卡片左侧是金额或 Token 头条，右侧是过去一年的 GitHub 风格用量墙（只统计计入摘要的已记录 Token），随后用图表按 agent、计费域和模型分开展示占比与可交互的每日趋势，再以占比条呈现模型排行及符合条件的 Token 按公开 API 价格计算出的对等费用。用量墙与所选的 24 小时 / 7 天 / 30 天窗口相互独立。
 - **设置面板**：采用居中双栏结构，将连接管理、自定义模型费率、运行监控、系统诊断与数据隐私划分为 5 个独立分类，方便高效维护各项配置。
 
 实际账单、平台报告估算、固定订阅费和 API 对等零售价是四类互不混合的证据。
@@ -43,9 +41,9 @@ Grok Build/SuperGrok 与 xAI API 是两个独立计费域，其额度、Token
 本地 Web 服务会先启动，再在后台运行连接发现、平台用量、模型定价和保留整理四个独立模块。
 已有缓存会立即展示；每个标签页只在自己的数据区域显示更新状态，已经完成的内容始终可用。
 
-本地会话记录扫描使用跨进程持久化且不保存原始个人路径的文件索引。历史费用只在价格目录版本变化时
-重新计算，并按有限批次处理；时间、平台、模型和计费域索引及保留期整理会在平台采集完成后交给后台
-工作线程执行。设置里提供需要明确确认的
+本地会话记录扫描使用跨进程持久化且不保存原始个人路径的文件索引。刷新只为新增或定价输入已变化的
+Token 观测计算 API 对等零售价；历史费用只在价格目录版本变化时补算仍未定价的观测，并按有限批次处理。
+时间、平台、模型和计费域索引及保留期整理会在平台采集完成后交给后台工作线程，且仅在确有工作（索引尚未建立、或存在超过 90 天的原始观测）时执行。Token 与费用区域的刷新条只在价格推导实际运行时显示，不等待连接器采集或保留整理。设置里提供需要明确确认的
 **硬重算全部数据**：它会忽略缓存、消耗较多资源，并且可能等待很久，但不会阻塞页面。
 
 ## 开发调试

package/dist/cli.js

@@ -789,17 +789,44 @@ function mergeCatalogWithCustomRates(catalog, customRates) {
     entries: [...customEntries, ...catalog.entries]
   };
 }
+function catalogModelKey(providerId, model) {
+  return `${providerId}\0${model.trim().toLowerCase()}`;
+}
+function indexRetailPriceCatalog(catalog) {
+  const cached = catalogIndexes.get(catalog);
+  if (cached) return cached;
+  const byProviderAndModel = /* @__PURE__ */ new Map();
+  for (const entry of catalog.entries) {
+    const models = /* @__PURE__ */ new Set([entry.canonicalModel, ...entry.aliases]);
+    for (const model of models) {
+      const key = catalogModelKey(entry.providerId, model);
+      const list = byProviderAndModel.get(key);
+      if (list) list.push(entry);
+      else byProviderAndModel.set(key, [entry]);
+    }
+  }
+  const index = { byProviderAndModel };
+  catalogIndexes.set(catalog, index);
+  return index;
+}
+function matchingCatalogEntries(index, providerId, observation, normalizedModel) {
+  const entries = index.byProviderAndModel.get(catalogModelKey(providerId, normalizedModel)) ?? [];
+  return entries.filter(
+    (entry) => entry.billingDomainId === observation.billingDomainId || entry.billingDomainId === "*"
+  );
+}
 function deriveRetailEquivalentCosts(snapshot2, catalog = ANTHROPIC_PRICING_CATALOG, calculatedAt = snapshot2.observedAt) {
   const costs = [];
   const decisions = [];
+  const index = indexRetailPriceCatalog(catalog);
   for (const observation of snapshot2.usage) {
-    const result = priceObservation(snapshot2.provider.id, observation, catalog, calculatedAt);
+    const result = priceObservation(snapshot2.provider.id, observation, index, calculatedAt);
     decisions.push(result.decision);
     if (result.cost) costs.push(result.cost);
   }
   return { costs, decisions };
 }
-function priceObservation(providerId, observation, catalog, calculatedAt) {
+function priceObservation(providerId, observation, index, calculatedAt) {
   const unavailable = (reason) => ({
     cost: null,
     decision: {
@@ -814,10 +841,11 @@ function priceObservation(providerId, observation, catalog, calculatedAt) {
     return unavailable("model-unclassified");
   }
   const normalizedModel = normalized.model.trim().toLowerCase();
-  const matchingCandidates = catalog.entries.filter(
-    (entry2) => entry2.providerId === providerId && (entry2.billingDomainId === observation.billingDomainId || entry2.billingDomainId === "*") && [entry2.canonicalModel, ...entry2.aliases].some(
-      (model) => model.trim().toLowerCase() === normalizedModel
-    )
+  const matchingCandidates = matchingCatalogEntries(
+    index,
+    providerId,
+    observation,
+    normalizedModel
   );
   let selectedEntries;
   const customExact = matchingCandidates.filter(
@@ -977,7 +1005,7 @@ function matchContext(rule, observation) {
 function preciseMoney(value) {
   return Number(value.toFixed(12));
 }
-var ANTHROPIC_PRICING_CATALOG, OPENCODE_GO_SOURCE, XAI_PRICING_SOURCE, OPENAI_GPT_56_SOURCE, OPENAI_GPT_6_ASTRA_SOURCE, OPEN_CODE_GO_SOURCE_PATH, openCodeHistory, OPEN_CODE_BASE_HISTORY, OPEN_CODE_PRICE_HISTORY, OPEN_CODE_FLAT_MODELS, OPEN_CODE_CONTEXT_MODELS, DEEPSEEK_GO_MODELS, WEEKDAY_PEAK_RANGES, DEEPSEEK_OFFICIAL_SOURCE, DEEPSEEK_OFFICIAL_MODELS, DEEPSEEK_OFFICIAL_EFFECTIVE_FROM, GOOGLE_ANTIGRAVITY_SOURCE, GOOGLE_ANTIGRAVITY_EFFECTIVE_FROM, DIRECT_OFFICIAL_PRICING_CATALOG, OFFICIAL_PRICING_CATALOG;
+var ANTHROPIC_PRICING_CATALOG, OPENCODE_GO_SOURCE, XAI_PRICING_SOURCE, OPENAI_GPT_56_SOURCE, OPENAI_GPT_6_ASTRA_SOURCE, OPEN_CODE_GO_SOURCE_PATH, openCodeHistory, OPEN_CODE_BASE_HISTORY, OPEN_CODE_PRICE_HISTORY, OPEN_CODE_FLAT_MODELS, OPEN_CODE_CONTEXT_MODELS, DEEPSEEK_GO_MODELS, WEEKDAY_PEAK_RANGES, DEEPSEEK_OFFICIAL_SOURCE, DEEPSEEK_OFFICIAL_MODELS, DEEPSEEK_OFFICIAL_EFFECTIVE_FROM, GOOGLE_ANTIGRAVITY_SOURCE, GOOGLE_ANTIGRAVITY_EFFECTIVE_FROM, DIRECT_OFFICIAL_PRICING_CATALOG, OFFICIAL_PRICING_CATALOG, catalogIndexes;
 var init_retail_pricing = __esm({
   "src/core/retail-pricing.ts"() {
     "use strict";
@@ -1447,6 +1475,7 @@ var init_retail_pricing = __esm({
         ...openCodeLocalHistoryEntries(DIRECT_OFFICIAL_PRICING_CATALOG.entries)
       ]
     };
+    catalogIndexes = /* @__PURE__ */ new WeakMap();
   }
 });
 
@@ -1492,7 +1521,7 @@ function yieldToEventLoop() {
 }
 function yieldForPendingHttp() {
   return new Promise((resolve5) => {
-    setTimeout(resolve5, PENDING_HTTP_YIELD_MS);
+    setTimeout(() => setImmediate(resolve5), PENDING_HTTP_YIELD_MS);
   });
 }
 function providerForConnector(id, definitions) {
@@ -1639,23 +1668,16 @@ var init_usage_application = __esm({
           let cursor = null;
           do {
             const page = this.#repository.getRetailPricingBackfillPage(cursor, 250);
-            for (const snapshot2 of page.snapshots) {
-              const costs = deriveRetailEquivalentCosts(
-                snapshot2,
-                this.#priceCatalog,
-                calculatedAt
-              ).costs;
-              if (costs.length > 0) this.#repository.saveDerivedCosts(snapshot2.provider.id, costs);
-            }
+            this.#saveDerivedRetailCosts(page.snapshots, calculatedAt);
             cursor = page.nextCursor;
             if (cursor) await yieldToEventLoop();
           } while (cursor);
         } else {
-          for (const snapshot2 of this.#repository.getRetailPricingBackfillSnapshots?.() ?? []) {
-            const costs = deriveRetailEquivalentCosts(snapshot2, this.#priceCatalog, calculatedAt).costs;
-            if (costs.length > 0) this.#repository.saveDerivedCosts(snapshot2.provider.id, costs);
-            await yieldToEventLoop();
-          }
+          this.#saveDerivedRetailCosts(
+            this.#repository.getRetailPricingBackfillSnapshots?.() ?? [],
+            calculatedAt
+          );
+          await yieldToEventLoop();
         }
         this.#repository.saveApplicationState?.(catalogVersionKey, this.#priceCatalog.version);
       }
@@ -2195,9 +2217,23 @@ var init_usage_application = __esm({
         );
         return result;
       }
+      #saveDerivedRetailCosts(snapshots, calculatedAt) {
+        if (!this.#priceCatalog) return;
+        for (const snapshot2 of snapshots) {
+          const costs = deriveRetailEquivalentCosts(snapshot2, this.#priceCatalog, calculatedAt).costs;
+          if (costs.length > 0) this.#repository.saveDerivedCosts?.(snapshot2.provider.id, costs);
+        }
+      }
       #withRetailCosts(snapshot2) {
-        const retailCosts = this.#priceCatalog ? deriveRetailEquivalentCosts(snapshot2, this.#priceCatalog, this.#clock().toISOString()).costs : [];
-        return { ...snapshot2, costs: [...snapshot2.costs, ...retailCosts] };
+        if (!this.#priceCatalog) return snapshot2;
+        const usage = this.#repository.observationsNeedingRetailDerivation ? this.#repository.observationsNeedingRetailDerivation(snapshot2.provider.id, snapshot2.usage) : snapshot2.usage;
+        if (usage.length === 0) return snapshot2;
+        const retailCosts = deriveRetailEquivalentCosts(
+          { ...snapshot2, usage },
+          this.#priceCatalog,
+          this.#clock().toISOString()
+        ).costs;
+        return retailCosts.length === 0 ? snapshot2 : { ...snapshot2, costs: [...snapshot2.costs, ...retailCosts] };
       }
       async discoverConnectors() {
         if (!this.#discoveryProbe) return this.getConnectorStatuses();
@@ -2300,7 +2336,7 @@ var init_usage_application = __esm({
         return status;
       }
     };
-    PENDING_HTTP_YIELD_MS = 20;
+    PENDING_HTTP_YIELD_MS = 75;
   }
 });
 
@@ -6006,7 +6042,6 @@ var init_antigravity_sqlite_usage_client = __esm({
   "src/server/antigravity-sqlite-usage-client.ts"() {
     "use strict";
     init_token_normalization();
-    init_retail_pricing();
     ({ DatabaseSync: DatabaseSync2 } = createRequire2(import.meta.url)(
       "node:sqlite"
     ));
@@ -6015,7 +6050,6 @@ var init_antigravity_sqlite_usage_client = __esm({
       #clock;
       #lookbackDays;
       #cachePath;
-      #priceCatalog;
       #fileCache = /* @__PURE__ */ new Map();
       #cacheLoaded = false;
       constructor(options) {
@@ -6023,7 +6057,6 @@ var init_antigravity_sqlite_usage_client = __esm({
         this.#clock = options.clock ?? (() => /* @__PURE__ */ new Date());
         this.#lookbackDays = options.lookbackDays ?? 90;
         this.#cachePath = options.cachePath;
-        this.#priceCatalog = options.priceCatalog ?? OFFICIAL_PRICING_CATALOG;
       }
       async readUsage(options = { mode: "incremental" }) {
         await this.#loadCache();
@@ -6048,21 +6081,10 @@ var init_antigravity_sqlite_usage_client = __esm({
           }
         }
         usage.sort((a, b) => a.observedAt.localeCompare(b.observedAt) || a.id.localeCompare(b.id));
-        const { costs } = deriveRetailEquivalentCosts(
-          {
-            provider: { id: "antigravity", displayName: "Antigravity" },
-            billingDomains: [{ id: "code-assist-subscription", displayName: "Gemini Code Assist" }],
-            quotaBuckets: [],
-            usage,
-            costs: [],
-            observedAt: this.#clock().toISOString()
-          },
-          this.#priceCatalog
-        );
         await this.#persistCache();
         return {
           usage,
-          costs,
+          costs: [],
           complete: allComplete
         };
       }
@@ -8120,6 +8142,23 @@ function pricingBackfillSnapshots(rows) {
   }
   return [...snapshots.values()];
 }
+function unpricedRetailPredicate(alias) {
+  return `NOT EXISTS (
+    SELECT 1 FROM cost_records priced
+    WHERE priced.provider_id = ${alias}.provider_id
+      AND priced.usage_observation_id = ${alias}.id
+      AND priced.kind = 'retail-equivalent'
+  )`;
+}
+function chunked(items, size) {
+  if (items.length === 0) return [];
+  if (items.length <= size) return [items];
+  const chunks = [];
+  for (let index = 0; index < items.length; index += size) {
+    chunks.push(items.slice(index, index + size));
+  }
+  return chunks;
+}
 function normalizeCostForPersistence(cost) {
   if (cost.kind !== "subscription") return cost;
   return {
@@ -8473,50 +8512,17 @@ function buildTokenMoneyWorkbench(providers, now, query) {
     };
   });
   const buckets = emptyIntervals.map((emptyInterval, index) => {
-    const segments = allHistories.flatMap(({ provider, domain, history, includedInHeadline }) => {
-      const interval = history.intervals[index];
-      if (!interval) return [];
-      const retailEquivalent2 = buildWorkbenchMetric(
-        interval.costs,
-        "retail-equivalent",
-        comparisonCurrency,
-        interval.tokenEvidence.recordedTokens,
-        history.exchangeRates
-      );
-      const reportedEstimate2 = buildWorkbenchMetric(
-        interval.costs,
-        "reported-estimate",
-        comparisonCurrency,
-        interval.tokenEvidence.recordedTokens,
-        history.exchangeRates
-      );
-      if (interval.tokenEvidence.observationCount === 0 && retailEquivalent2.records === 0 && reportedEstimate2.records === 0)
-        return [];
-      return [
-        {
-          providerId: provider.id,
-          providerDisplayName: provider.displayName,
-          billingDomainId: domain.id,
-          billingDomainDisplayName: domain.displayName,
-          includedInHeadline,
-          recordedTokens: interval.tokenEvidence.recordedTokens,
-          observationCount: interval.tokenEvidence.observationCount,
-          timePrecisions: interval.tokenEvidence.timePrecisions,
-          authorities: interval.authorities ?? [],
-          lastObservedAt: interval.lastObservedAt ?? null,
-          retailEquivalent: {
-            status: retailEquivalent2.status,
-            amount: retailEquivalent2.amount,
-            currency: comparisonCurrency
-          },
-          reportedEstimate: {
-            status: reportedEstimate2.status,
-            amount: reportedEstimate2.amount,
-            currency: comparisonCurrency
-          }
-        }
-      ];
-    });
+    const segments = allHistories.flatMap(
+      ({ provider, domain, history, includedInHeadline }) => workbenchTrendSegmentsForInterval(
+        provider,
+        domain,
+        history,
+        includedInHeadline,
+        emptyInterval,
+        index,
+        comparisonCurrency
+      )
+    );
     return {
       start: emptyInterval.start.toISOString(),
       end: emptyInterval.end.toISOString(),
@@ -8554,6 +8560,133 @@ function buildTokenMoneyWorkbench(providers, now, query) {
     )
   };
 }
+function workbenchTrendSegmentsForInterval(provider, domain, history, includedInHeadline, emptyInterval, index, comparisonCurrency) {
+  const interval = history.intervals[index];
+  if (!interval) return [];
+  const start = emptyInterval.start.toISOString();
+  const end = emptyInterval.end.toISOString();
+  const models = history.models.flatMap((model) => {
+    const observations = (model.observations ?? []).filter(
+      (observation) => observation.observedAt >= start && observation.observedAt < end
+    );
+    const priceEvidence = (model.priceEvidence ?? []).filter(
+      (cost) => (cost.observedAt ?? "") >= start && (cost.observedAt ?? "") < end
+    );
+    const tokenEvidence = emptyTokenEvidence();
+    const authorities = /* @__PURE__ */ new Set();
+    let lastObservedAt = null;
+    for (const observation of observations) {
+      tokenEvidence.recordedTokens += observation.recordedTokens;
+      tokenEvidence.observationCount += 1;
+      tokenEvidence.timePrecisions.add(observation.timePrecision);
+      authorities.add(observation.authority);
+      if (!lastObservedAt || observation.observedAt > lastObservedAt) {
+        lastObservedAt = observation.observedAt;
+      }
+    }
+    const retailEquivalent = buildWorkbenchMetric(
+      priceEvidence,
+      "retail-equivalent",
+      comparisonCurrency,
+      tokenEvidence.recordedTokens,
+      history.exchangeRates
+    );
+    const reportedEstimate = buildWorkbenchMetric(
+      priceEvidence,
+      "reported-estimate",
+      comparisonCurrency,
+      tokenEvidence.recordedTokens,
+      history.exchangeRates
+    );
+    if (tokenEvidence.observationCount === 0 && retailEquivalent.records === 0 && reportedEstimate.records === 0) {
+      return [];
+    }
+    return [
+      {
+        providerId: provider.id,
+        providerDisplayName: provider.displayName,
+        billingDomainId: domain.id,
+        billingDomainDisplayName: domain.displayName,
+        model: model.model,
+        includedInHeadline,
+        recordedTokens: tokenEvidence.recordedTokens,
+        observationCount: tokenEvidence.observationCount,
+        timePrecisions: [...tokenEvidence.timePrecisions].sort(),
+        authorities: [...authorities].sort(),
+        lastObservedAt,
+        retailEquivalent: {
+          status: retailEquivalent.status,
+          amount: retailEquivalent.amount,
+          currency: comparisonCurrency
+        },
+        reportedEstimate: {
+          status: reportedEstimate.status,
+          amount: reportedEstimate.amount,
+          currency: comparisonCurrency
+        }
+      }
+    ];
+  });
+  const unclassifiedObservations = (history.unclassified.observations ?? []).filter(
+    (observation) => observation.observedAt >= start && observation.observedAt < end
+  );
+  const unclassifiedTokenEvidence = emptyTokenEvidence();
+  const unclassifiedAuthorities = /* @__PURE__ */ new Set();
+  let unclassifiedLastObservedAt = null;
+  for (const observation of unclassifiedObservations) {
+    unclassifiedTokenEvidence.recordedTokens += observation.recordedTokens;
+    unclassifiedTokenEvidence.observationCount += 1;
+    unclassifiedTokenEvidence.timePrecisions.add(observation.timePrecision);
+    unclassifiedAuthorities.add(observation.authority);
+    if (!unclassifiedLastObservedAt || observation.observedAt > unclassifiedLastObservedAt) {
+      unclassifiedLastObservedAt = observation.observedAt;
+    }
+  }
+  const leftoverCosts = models.length === 0 ? interval.costs : [];
+  const leftoverRetailEquivalent = buildWorkbenchMetric(
+    leftoverCosts,
+    "retail-equivalent",
+    comparisonCurrency,
+    unclassifiedTokenEvidence.recordedTokens,
+    history.exchangeRates
+  );
+  const leftoverReportedEstimate = buildWorkbenchMetric(
+    leftoverCosts,
+    "reported-estimate",
+    comparisonCurrency,
+    unclassifiedTokenEvidence.recordedTokens,
+    history.exchangeRates
+  );
+  if (unclassifiedTokenEvidence.observationCount === 0 && leftoverRetailEquivalent.records === 0 && leftoverReportedEstimate.records === 0) {
+    return models;
+  }
+  return [
+    ...models,
+    {
+      providerId: provider.id,
+      providerDisplayName: provider.displayName,
+      billingDomainId: domain.id,
+      billingDomainDisplayName: domain.displayName,
+      model: null,
+      includedInHeadline,
+      recordedTokens: unclassifiedTokenEvidence.recordedTokens,
+      observationCount: unclassifiedTokenEvidence.observationCount,
+      timePrecisions: [...unclassifiedTokenEvidence.timePrecisions].sort(),
+      authorities: [...unclassifiedAuthorities].sort(),
+      lastObservedAt: unclassifiedLastObservedAt,
+      retailEquivalent: {
+        status: leftoverRetailEquivalent.status,
+        amount: leftoverRetailEquivalent.amount,
+        currency: comparisonCurrency
+      },
+      reportedEstimate: {
+        status: leftoverReportedEstimate.status,
+        amount: leftoverReportedEstimate.amount,
+        currency: comparisonCurrency
+      }
+    }
+  ];
+}
 function allDomainHistories(providers) {
   return providers.flatMap(
     (provider) => provider.billingDomains.map((domain) => ({
@@ -9342,13 +9475,18 @@ function freshnessStatus(lastSuccessAt, now) {
   if (!lastSuccessAt) return "unavailable";
   return now.getTime() - new Date(lastSuccessAt).getTime() <= FRESHNESS_WINDOW_MS ? "fresh" : "stale";
 }
-var FRESHNESS_WINDOW_MS, QUERY_INDEXES_SQL, DELETE_DERIVED_RETAIL_COSTS_SQL, DELETE_DEMO_PROVIDER_DATA_SQL, DatabaseSync3, RETENTION_AGGREGATE_SQL, SqliteUsageRepository;
+var FRESHNESS_WINDOW_MS, SQLITE_VARIABLE_CHUNK, USAGE_PRICING_INPUT_COLUMNS, QUERY_INDEXES_SQL, DELETE_DERIVED_RETAIL_COSTS_SQL, DELETE_DEMO_PROVIDER_DATA_SQL, DatabaseSync3, RETENTION_AGGREGATE_SQL, SqliteUsageRepository;
 var init_sqlite_usage_repository = __esm({
   "src/server/sqlite-usage-repository.ts"() {
     "use strict";
     init_token_normalization();
     init_quota_normalization();
     FRESHNESS_WINDOW_MS = 15 * 60 * 1e3;
+    SQLITE_VARIABLE_CHUNK = 400;
+    USAGE_PRICING_INPUT_COLUMNS = `id, billing_domain_id, model, observed_at, total_tokens, input_tokens,
+  output_tokens, reasoning_tokens, cache_read_tokens, cache_write_tokens, cache_write_5m_tokens,
+  cache_write_1h_tokens, unclassified_tokens, reasoning_semantics, cache_read_semantics,
+  cache_write_semantics, model_attribution, time_precision, aggregation_temporality`;
     QUERY_INDEXES_SQL = `
   CREATE INDEX IF NOT EXISTS usage_observed_at_idx
     ON usage_observations(observed_at);
@@ -9627,14 +9765,9 @@ var init_sqlite_usage_repository = __esm({
            usage_scope = excluded.usage_scope,
            aggregation_temporality = excluded.aggregation_temporality`
           );
-          const existingPricedUsageStatement = this.#database.prepare(
-            `SELECT billing_domain_id, model, observed_at, total_tokens, input_tokens, output_tokens,
-                reasoning_tokens, cache_read_tokens, cache_write_tokens, cache_write_5m_tokens,
-                cache_write_1h_tokens, unclassified_tokens,
-                reasoning_semantics, cache_read_semantics, cache_write_semantics, model_attribution,
-                time_precision, aggregation_temporality
-         FROM usage_observations
-         WHERE provider_id = ? AND id = ?`
+          const existingById = this.#usagePricingInputs(
+            snapshot2.provider.id,
+            snapshot2.usage.map((observation) => observation.id)
           );
           const deleteDerivedCostsStatement = this.#database.prepare(
             `DELETE FROM cost_records
@@ -9642,7 +9775,7 @@ var init_sqlite_usage_repository = __esm({
           );
           for (const observation of snapshot2.usage) {
             const normalized = normalizeTokenObservation(observation);
-            const existing = existingPricedUsageStatement.get(snapshot2.provider.id, normalized.id);
+            const existing = existingById.get(normalized.id);
             if (existing && pricingInputsChanged(existing, normalized)) {
               deleteDerivedCostsStatement.run(snapshot2.provider.id, normalized.id);
             }
@@ -9812,6 +9945,7 @@ var init_sqlite_usage_repository = __esm({
          JOIN billing_domains b
            ON b.provider_id = u.provider_id AND b.id = u.billing_domain_id
          WHERE ${additiveUsagePredicate("u")}
+           AND ${unpricedRetailPredicate("u")}
          ORDER BY u.provider_id, u.observed_at, u.id`
         ).all();
         return pricingBackfillSnapshots(rows);
@@ -9834,6 +9968,7 @@ var init_sqlite_usage_repository = __esm({
          JOIN billing_domains b
            ON b.provider_id = u.provider_id AND b.id = u.billing_domain_id
          WHERE ${additiveUsagePredicate("u")}
+           AND ${unpricedRetailPredicate("u")}
            ${cursorPredicate}
          ORDER BY u.provider_id, u.observed_at, u.id
          LIMIT ?`
@@ -9851,6 +9986,30 @@ var init_sqlite_usage_repository = __esm({
           } : null
         };
       }
+      observationsNeedingRetailDerivation(providerId, observations) {
+        if (observations.length === 0) return [];
+        const existing = this.#usagePricingInputs(
+          providerId,
+          observations.map((observation) => observation.id)
+        );
+        return observations.filter((observation) => {
+          const normalized = normalizeTokenObservation(observation);
+          const prior = existing.get(normalized.id);
+          return !prior || pricingInputsChanged(prior, normalized);
+        });
+      }
+      #usagePricingInputs(providerId, observationIds) {
+        const byId = /* @__PURE__ */ new Map();
+        for (const ids of chunked(observationIds, SQLITE_VARIABLE_CHUNK)) {
+          const rows = this.#database.prepare(
+            `SELECT ${USAGE_PRICING_INPUT_COLUMNS}
+           FROM usage_observations
+           WHERE provider_id = ? AND id IN (${ids.map(() => "?").join(", ")})`
+          ).all(providerId, ...ids);
+          for (const row of rows) byId.set(String(row.id), row);
+        }
+        return byId;
+      }
       saveDerivedCosts(providerId, costs) {
         if (costs.length === 0) return;
         if (costs.some((cost) => cost.kind !== "retail-equivalent")) {
@@ -10333,6 +10492,7 @@ var init_sqlite_usage_repository = __esm({
       }
       compactUsageHistory(now) {
         const cutoff = new Date(now.getTime() - 90 * 24 * 60 * 60 * 1e3).toISOString();
+        if (!this.#hasObservationsOlderThan(cutoff)) return this.getRetentionStatus();
         this.#database.exec("BEGIN IMMEDIATE");
         try {
           this.#database.prepare(RETENTION_AGGREGATE_SQL).run(cutoff);
@@ -10405,6 +10565,7 @@ var init_sqlite_usage_repository = __esm({
         ).run(key, value);
       }
       async ensureQueryIndexes() {
+        if (this.#queryIndexesReady()) return;
         await runDatabaseWorker({
           databasePath: this.#databasePath,
           operation: "indexes",
@@ -10413,6 +10574,7 @@ var init_sqlite_usage_repository = __esm({
       }
       async maintainUsageHistory(now) {
         const cutoff = new Date(now.getTime() - 90 * 24 * 60 * 60 * 1e3).toISOString();
+        if (!this.#hasObservationsOlderThan(cutoff)) return this.getRetentionStatus();
         return runDatabaseWorker({
           databasePath: this.#databasePath,
           operation: "retention",
@@ -10422,6 +10584,14 @@ var init_sqlite_usage_repository = __esm({
           now: now.toISOString()
         });
       }
+      #queryIndexesReady() {
+        const indexes = this.#database.prepare("PRAGMA index_list('usage_observations')").all();
+        return indexes.some((index) => index.name === "usage_provider_model_time_idx");
+      }
+      #hasObservationsOlderThan(cutoff) {
+        const row = this.#database.prepare("SELECT 1 AS present FROM usage_observations WHERE observed_at < ? LIMIT 1").get(cutoff);
+        return row !== void 0;
+      }
       close() {
         this.#database.close();
       }
@@ -10852,7 +11022,7 @@ var init_sqlite_usage_repository = __esm({
           models: [...byModel.entries()].map(([model, value]) => {
             const observationIds = new Set(value.observations.map((observation) => observation.id));
             const priceEvidence = costs.filter(
-              (cost) => (cost.kind === "retail-equivalent" || cost.kind === "reported-estimate") && cost.usage_observation_id !== null && observationIds.has(cost.usage_observation_id)
+              (cost) => (cost.kind === "retail-equivalent" || cost.kind === "reported-estimate") && (cost.usage_observation_id !== null && observationIds.has(cost.usage_observation_id) || cost.usage_observation_id === null && cost.model === model)
             ).map((cost) => {
               const summarized = summarizeCosts([cost], Number(cost.priced_tokens ?? 0))[0];
               return {