vscode-extension: v0.5.1 — fix negative savings on dashboard

Mirror of upstream PR #36 (fix/savings-includes-media-cost). The
"Saved vs Opus" panel hero would show negative dollar amounts as
soon as a user spent meaningfully on ImageGen / VideoGen, e.g.

    $-8.79
    You spent $20.4896 instead of $11.70

Root cause: getStatsSummary() compared an Opus-token baseline (chat
only — image/video log inputTokens=0/outputTokens=0) against
totalCostUsd (chat + media combined), so once media spend exceeded
the chat-vs-Opus delta the difference flipped negative.

Fix: split byModel into chatOnlyCost (rows with tokens) and mediaCost
(rows without). opusCost on the display side now equals
opusChatCost + mediaCost so "you spent X instead of Y" stays
apples-to-apples; saved = max(0, opusChatCost - chatOnlyCost) is
the chat-side delta only and is clamped non-negative.

Bumps vscode-extension to 0.5.1; updates README changelog.

Author: KillerQueen-Z

src/panel/html.ts

@@ -684,8 +684,13 @@ async function loadOverview() {
     document.getElementById('period-info').textContent = stats.period || '';
 
     if (stats.opusCost > 0) {
-      const saved = stats.saved || (stats.opusCost - stats.totalCostUsd);
-      const pct = stats.savedPct || ((1 - stats.totalCostUsd / stats.opusCost) * 100);
+      // tracker.ts now returns saved already clamped to >= 0 and opusCost
+      // already inclusive of media (so comparing to totalCostUsd is
+      // apples-to-apples). Older summaries fall back to clamped recompute.
+      const saved = stats.saved != null
+        ? Math.max(0, stats.saved)
+        : Math.max(0, stats.opusCost - stats.totalCostUsd);
+      const pct = stats.opusCost > 0 ? (saved / stats.opusCost) * 100 : 0;
       document.getElementById('savings-hero').style.display = 'flex';
       document.getElementById('savings-amount').textContent = usdBig(saved);
       document.getElementById('savings-pct').textContent = pct.toFixed(0) + '%';

src/stats/tracker.ts

@@ -257,19 +257,45 @@ export function recordUsage(
 export function getStatsSummary(): {
   stats: Stats;
   opusCost: number;
+  /** All chat / token-billed model spend (excludes image / video / music). */
+  chatOnlyCost: number;
+  /** Per-image / per-second / per-track media generation spend. */
+  mediaCost: number;
   saved: number;
   savedPct: number;
   avgCostPerRequest: number;
   period: string;
 } {
   const stats = loadStats();
 
-  // Calculate what it would cost with the Opus-tier baseline
-  const opusCost =
+  // Hypothetical "if you'd used Opus for everything" baseline. Opus is a
+  // chat model — it can't replace ImageGen / VideoGen / Music (per_image,
+  // per_second, per_track billing), so for those rows the Opus-equivalent
+  // cost IS just the actual cost (no alternative). For chat rows, the
+  // baseline is the same tokens repriced at Opus rates.
+  //
+  // Walk byModel: rows with zero tokens are media (recordUsage stores
+  // image/video calls with inputTokens=0 outputTokens=0). Those count
+  // towards both sides equally; chat rows count at actual price on the
+  // "actual" side and at Opus rates on the "baseline" side.
+  let chatOnlyCost = 0;
+  let mediaCost = 0;
+  for (const m of Object.values(stats.byModel)) {
+    if ((m.inputTokens + m.outputTokens) > 0) chatOnlyCost += m.costUsd;
+    else mediaCost += m.costUsd;
+  }
+  const opusChatCost =
     (stats.totalInputTokens / 1_000_000) * OPUS_PRICING.input +
     (stats.totalOutputTokens / 1_000_000) * OPUS_PRICING.output;
-
-  const saved = opusCost - stats.totalCostUsd;
+  // What the bill would have been if every chat request had hit Opus
+  // pricing — media stays the same on both sides because there's no
+  // Opus alternative for image/video gen.
+  const opusCost = opusChatCost + mediaCost;
+
+  // Saved is the chat-side delta only (media nets to zero in the diff).
+  // Clamp to 0 so a session where the user paid more than Opus would have
+  // (e.g. Sonnet 4.6 with thinking enabled) doesn't show negative savings.
+  const saved = Math.max(0, opusChatCost - chatOnlyCost);
   const savedPct = opusCost > 0 ? (saved / opusCost) * 100 : 0;
   const avgCostPerRequest =
     stats.totalRequests > 0 ? stats.totalCostUsd / stats.totalRequests : 0;
@@ -285,5 +311,5 @@ export function getStatsSummary(): {
     else period = `${days} days`;
   }
 
-  return { stats, opusCost, saved, savedPct, avgCostPerRequest, period };
+  return { stats, opusCost, chatOnlyCost, mediaCost, saved, savedPct, avgCostPerRequest, period };
 }

vscode-extension/README.md

@@ -41,6 +41,9 @@ Franklin is an autonomous AI agent that runs directly in VS Code. It doesn't jus
 
 ## Changelog
 
+### 0.5.1
+- **Fix: \"Saved vs Opus\" no longer goes negative** — when meaningful spend hit ImageGen / VideoGen, the savings widget showed numbers like `$-8.79` because chat-only Opus baseline was being compared against total spend (chat + media). Now compared apples-to-apples (media counts on both sides; saved is the chat-side delta, clamped to >= 0). Mirror of upstream PR #36.
+
 ### 0.5.0
 - **Major core sync** — extension now rides on Franklin core v3.10.0, picking up two months of upstream work since 0.4.5 (no more cherry-pick lag):
   - **Detached background tasks** (v3.10.0) — `Detach` capability spawns long-running work as a separate process; `franklin task list / tail / wait / cancel` to drive them

vscode-extension/out/extension.cjs

@@ -147707,8 +147707,17 @@ function recordUsage(model, inputTokens, outputTokens, costUsd, latencyMs, fallb
 }
 function getStatsSummary() {
   const stats = loadStats();
-  const opusCost = stats.totalInputTokens / 1e6 * OPUS_PRICING.input + stats.totalOutputTokens / 1e6 * OPUS_PRICING.output;
-  const saved = opusCost - stats.totalCostUsd;
+  let chatOnlyCost = 0;
+  let mediaCost = 0;
+  for (const m2 of Object.values(stats.byModel)) {
+    if (m2.inputTokens + m2.outputTokens > 0)
+      chatOnlyCost += m2.costUsd;
+    else
+      mediaCost += m2.costUsd;
+  }
+  const opusChatCost = stats.totalInputTokens / 1e6 * OPUS_PRICING.input + stats.totalOutputTokens / 1e6 * OPUS_PRICING.output;
+  const opusCost = opusChatCost + mediaCost;
+  const saved = Math.max(0, opusChatCost - chatOnlyCost);
   const savedPct = opusCost > 0 ? saved / opusCost * 100 : 0;
   const avgCostPerRequest = stats.totalRequests > 0 ? stats.totalCostUsd / stats.totalRequests : 0;
   let period = "No data";
@@ -147721,7 +147730,7 @@ function getStatsSummary() {
     else
       period = `${days} days`;
   }
-  return { stats, opusCost, saved, savedPct, avgCostPerRequest, period };
+  return { stats, opusCost, chatOnlyCost, mediaCost, saved, savedPct, avgCostPerRequest, period };
 }
 var import_node_fs3, import_node_path2, import_node_os2, resolvedStatsFile, EMPTY_STATS, cachedStats, flushTimer, FLUSH_DELAY_MS;
 var init_tracker = __esm({

vscode-extension/out/extension.cjs.map

@@ -2,7 +2,7 @@
   "name": "franklin-vscode",
   "displayName": "Franklin",
   "description": "The AI agent with a wallet. Autonomous marketing & trading agent powered by x402 micropayments.",
-  "version": "0.5.0",
+  "version": "0.5.1",
   "publisher": "blockrun",
   "icon": "images/icon.png",
   "repository": {