fix(usage): count cached tokens in every total — Codebase/Today/Yesterday were ~99% short

CHANGELOG.md

@@ -20,6 +20,7 @@ All notable changes to the **OpenCode Go BYOK Provider** extension are documente
 
 ### Fixed
 
+- **`[Usage]` Token counting now includes cached tokens — Codebase/Today/Yesterday totals were massively undercounted.** The CLI's per-message `tokens.input` EXCLUDES cache reads, and the sum used for the charts and rows (`input + output + reasoning`) dropped `cache.read` entirely. DeepSeek V4 sessions routinely carry ~700K cached prompt tokens per message, so the displayed totals were ~99% short. Every counting site (`sumDailyUsage`, `buildUsageSeries`, `codebaseUsage`) now uses the full total `input + output + reasoning + cache.read`, verified to match the DB's authoritative `tokens.total`. Requests and costs were already correct (1 per assistant message; cost from the CLI's own per-message field, extension-side billing via billable = prompt − cached).
 - **`[Usage]` DeepSeek V4 reasoning no longer leaks into visible text.** With thinking "off" the DeepSeek V4 family still produces genuine chain-of-thought (only the effort parameter is omitted), so the Go-gateway "thinking-off" workaround (#37635) misclassified it as content and printed it as visible text. `reasoning_content` from reasoning-first models now always lands in the thinking block; the workaround stays scoped to models that genuinely stop reasoning when no effort is sent (verified case: MiMo). Dead branches removed around the same logic (redundant `isGoGateway` wrapper and the implied-empty `!visible` check).
 
 - **`[Usage]` SQLite reads no longer depend on the `sqlite3` binary.** The zero-usage mystery was the Android SDK's `sqlite3` (`~/Android/Sdk/platform-tools/sqlite3`) being on the PATH only when VS Code launches from a shell that exports it — desktop-launched windows silently lost all CLI history (Today/Yesterday/Codebase = 0 while the fetched quota kept working). The CLI history is now read through Node's built-in `node:sqlite` first (zero external dependencies, retried twice on busy WAL states), falling back to the `sqlite3` binary resolved from PATH **plus** known locations (system, Homebrew, Android SDK). Failures are logged with the exact error to the "OpenCode Go Usage" output channel.

src/goUsageTracker.ts

@@ -330,6 +330,14 @@ export interface HistoryRow {
   cwd?: string;
   /** Model that produced the message (OpenCode CLI data). */
   modelId?: string;
+  /**
+   * Total tokens for the message: input + output + reasoning + cache.read.
+   * The CLI's `tokens.input` EXCLUDES cached tokens — the authoritative
+   * `tokens.total` matches input + output + reasoning + cache.read — so this
+   * sum is what any "tokens used" display must count (parity with the
+   * extension's own promptTokens, which include cached tokens).
+   */
+  tokensTotal: number;
 }
 
 /** Non-negative finite integer (tokens can legitimately be 0). */
@@ -358,7 +366,7 @@ export function sumDailyUsage(
       if (row.createdMs < dayStartMs) continue;
       cost += row.cost;
       requests += 1;
-      tokens += row.tokensInput + row.tokensOutput + row.tokensReasoning;
+      tokens += row.tokensTotal;
     }
   }
 
@@ -468,7 +476,7 @@ export function buildUsageSeries(
 
   if (source !== "extension") {
     for (const row of rows) {
-      add(row.modelId, row.createdMs, row.cost, row.tokensInput + row.tokensOutput + row.tokensReasoning);
+      add(row.modelId, row.createdMs, row.cost, row.tokensTotal);
     }
   }
   if (source !== "cli") {
@@ -539,16 +547,23 @@ function normalizeHistoryRows(rows: unknown): HistoryRow[] {
         typeof candidate.createdMs === "number" && candidate.createdMs > 0 && typeof candidate.cost === "number" && candidate.cost >= 0
       );
     })
-    .map((row) => ({
-      createdMs: row.createdMs,
-      cost: row.cost,
-      tokensInput: positiveNumberish(row.tokensInput),
-      tokensOutput: positiveNumberish(row.tokensOutput),
-      tokensReasoning: positiveNumberish(row.tokensReasoning),
-      tokensCacheRead: positiveNumberish(row.tokensCacheRead),
-      cwd: typeof row.cwd === "string" && row.cwd.trim() ? row.cwd : undefined,
-      modelId: typeof row.modelId === "string" && row.modelId.trim() ? row.modelId : undefined,
-    }));
+    .map((row) => {
+      const tokensInput = positiveNumberish(row.tokensInput);
+      const tokensOutput = positiveNumberish(row.tokensOutput);
+      const tokensReasoning = positiveNumberish(row.tokensReasoning);
+      const tokensCacheRead = positiveNumberish(row.tokensCacheRead);
+      return {
+        createdMs: row.createdMs,
+        cost: row.cost,
+        tokensInput,
+        tokensOutput,
+        tokensReasoning,
+        tokensCacheRead,
+        tokensTotal: tokensInput + tokensOutput + tokensReasoning + tokensCacheRead,
+        cwd: typeof row.cwd === "string" && row.cwd.trim() ? row.cwd : undefined,
+        modelId: typeof row.modelId === "string" && row.modelId.trim() ? row.modelId : undefined,
+      };
+    });
 }
 
 /**
@@ -892,7 +907,7 @@ export class GoUsageTracker {
       if (!isCwdInWorkspace(row.cwd, folders)) continue;
       cost += row.cost;
       requests += 1;
-      tokens += row.tokensInput + row.tokensOutput + row.tokensReasoning;
+      tokens += row.tokensTotal;
     }
     return { cost, requests, tokens };
   }

src/test/goUsageTracker.test.ts

@@ -636,8 +636,26 @@ describe("sumDailyUsage", () => {
   const now = new Date();
   const dayMs = Date.UTC(now.getUTCFullYear(), now.getUTCMonth(), now.getUTCDate());
   const rows: HistoryRow[] = [
-    { createdMs: dayMs + 1000, cost: 0.1, tokensInput: 100, tokensOutput: 50, tokensReasoning: 20, tokensCacheRead: 10, cwd: "/repo" },
-    { createdMs: dayMs - 60_000, cost: 0.2, tokensInput: 200, tokensOutput: 100, tokensReasoning: 0, tokensCacheRead: 0, cwd: "/repo" },
+    {
+      createdMs: dayMs + 1000,
+      cost: 0.1,
+      tokensInput: 100,
+      tokensOutput: 50,
+      tokensReasoning: 20,
+      tokensCacheRead: 10,
+      cwd: "/repo",
+      tokensTotal: 180,
+    },
+    {
+      createdMs: dayMs - 60_000,
+      cost: 0.2,
+      tokensInput: 200,
+      tokensOutput: 100,
+      tokensReasoning: 0,
+      tokensCacheRead: 0,
+      cwd: "/repo",
+      tokensTotal: 300,
+    },
   ];
   const entries: UsageLogEntry[] = [
     {
@@ -654,14 +672,14 @@ describe("sumDailyUsage", () => {
   it("merges CLI rows and extension entries in auto mode", () => {
     const total = sumDailyUsage(rows, entries, dayMs, "auto");
     assert.equal(total.requests, 2);
-    assert.equal(total.tokens, 210);
+    assert.equal(total.tokens, 220, "row total includes cache.read (180) plus the entry (40)");
     assert.ok(Math.abs(total.cost - 0.15) < 1e-9, `expected ~0.15, got ${String(total.cost)}`);
   });
 
   it("excludes rows before the day window", () => {
     const total = sumDailyUsage(rows, [], dayMs, "cli");
     assert.equal(total.requests, 1, "only the row inside the window counts");
-    assert.equal(total.tokens, 170, "input + output + reasoning");
+    assert.equal(total.tokens, 180, "input + output + reasoning + cache.read");
   });
 
   it("cli source ignores extension entries", () => {
@@ -747,6 +765,7 @@ describe("buildUsageSeries", () => {
       tokensOutput: 50,
       tokensReasoning: 0,
       tokensCacheRead: 0,
+      tokensTotal: 150,
       cwd: "/repo",
       modelId: "qwen3.6-plus",
     },
@@ -757,6 +776,7 @@ describe("buildUsageSeries", () => {
       tokensOutput: 100,
       tokensReasoning: 0,
       tokensCacheRead: 0,
+      tokensTotal: 300,
       cwd: "/repo",
       modelId: "deepseek-v4-flash",
     },
@@ -767,6 +787,7 @@ describe("buildUsageSeries", () => {
       tokensOutput: 150,
       tokensReasoning: 0,
       tokensCacheRead: 0,
+      tokensTotal: 450,
       cwd: "/repo",
       modelId: "qwen3.6-plus",
     },
@@ -777,6 +798,7 @@ describe("buildUsageSeries", () => {
       tokensOutput: 200,
       tokensReasoning: 0,
       tokensCacheRead: 0,
+      tokensTotal: 600,
       cwd: "/repo",
       modelId: "qwen3.6-plus",
     },
@@ -825,6 +847,24 @@ describe("buildUsageSeries", () => {
     assert.ok(!series.byModel.some((p) => p.model === "glm-5"));
   });
 
+  it("counts cached tokens in daily totals (tokens.input excludes cache)", () => {
+    const cached: HistoryRow[] = [
+      {
+        createdMs: dayMs,
+        cost: 0.1,
+        tokensInput: 152,
+        tokensOutput: 209,
+        tokensReasoning: 0,
+        tokensCacheRead: 699_392,
+        tokensTotal: 699_753,
+        cwd: "/repo",
+        modelId: "deepseek-v4-flash",
+      },
+    ];
+    const series = buildUsageSeries(cached, [], 1, dayMs, "cli");
+    assert.equal(series.days[0].tokens, 699_753, "cache.read must be part of the token total");
+  });
+
   it("lifetime windows (days=0) span from the earliest usage day", () => {
     const series = buildUsageSeries(rows, entries, 0, dayMs, "auto");
     // earliest row = dayMs - DAY → 2 buckets: yesterday + today