diff --git a/CHANGELOG.md b/CHANGELOG.md
index 06ae374..e4170e1 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,16 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Added
+
+- **Token breakdown: decompose the "in" headline into fresh vs cached.** The headline
+  "NN in" is dominated by cheap `cache_read` tokens; a new `Input  X new · Y cached`
+  line separates freshly-billed input (`input + cache_creation`) from cache reads, so the
+  cost line is interpretable (a big "in" at 97% cache is mostly $1.50/M reads, not $15/M
+  fresh input). Applied to the single-session and live views.
+
 ## [1.0.0] - 2026-02-18
 
 ### Added
diff --git a/package-lock.json b/package-lock.json
index 8b66636..b5aa8a4 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -186,7 +186,6 @@
       "integrity": "sha512-PI+bOkutJnO7hsPZALHKWFj83K/45p4ABODUf0kKIoJS8tXJgDDhjYJGg6Eh/x177OokLfBPJUV9z3OspGt99g==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@algolia/client-common": "5.48.2",
         "@algolia/requester-browser-xhr": "5.48.2",
@@ -1371,7 +1370,6 @@
       "integrity": "sha512-BH7YwL6rA93ReqeQS1c4bsPpcfOmJasG+Fkr6Y59q83f9M1WcBRHR2vM+P9eOisYRcN3ujQoiZY8uk5W+1WL8w==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "undici-types": "~6.21.0"
       }
@@ -1779,7 +1777,6 @@
       "integrity": "sha512-U5U2dCy+ei/elHp7jMdYOPWBO0sjiUEyIW0PEblm4cL8GvDNiyiOlhZmKgwWbyPPj0w0s6HXeMZBAofTdVWgIQ==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@algolia/abtesting": "1.14.2",
         "@algolia/client-abtesting": "5.48.2",
@@ -2055,7 +2052,6 @@
       "integrity": "sha512-/yNdlIkpWbM0ptxno3ONTuf+2g318kh2ez3KSeZN5dZ8YC6AAmgeWz+GasYYiBJPFaYcSAPeu4GfhUaChzIJXA==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "tabbable": "^6.4.0"
       }
@@ -2360,7 +2356,6 @@
       "integrity": "sha512-5gTmgEY/sqK6gFXLIsQNH19lWb4ebPDLA4SdLP7dsWkIXHWlG66oPuVvXSGFPppYZz8ZDZq0dYYrbHfBCVUb1Q==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "engines": {
         "node": ">=12"
       },
@@ -2671,7 +2666,6 @@
       "integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
       "dev": true,
       "license": "Apache-2.0",
-      "peer": true,
       "bin": {
         "tsc": "bin/tsc",
         "tsserver": "bin/tsserver"
@@ -2796,7 +2790,6 @@
       "integrity": "sha512-w+N7Hifpc3gRjZ63vYBXA56dvvRlNWRczTdmCBBa+CotUzAPf5b7YMdMR/8CQoeYE5LX3W4wj6RYTgonm1b9DA==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "esbuild": "^0.27.0",
         "fdir": "^6.5.0",
@@ -3358,7 +3351,6 @@
       "integrity": "sha512-o5a9xKjbtuhY6Bi5S3+HvbRERmouabWbyUcpXXUA1u+GNUKoROi9byOJ8M0nHbHYHkYICiMlqxkg1KkYmm25Sw==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "esbuild": "^0.21.3",
         "postcss": "^8.4.43",
@@ -3497,7 +3489,6 @@
       "integrity": "sha512-BRdrNfeoccSoIZeIhyPBfvWSLFP4q8J3u8Ju8Ug5vu3LdD+yTM13Sg4sKtljxozbnuMu1NB1X5HBHRYUzFocKg==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@vue/compiler-dom": "3.5.28",
         "@vue/compiler-sfc": "3.5.28",
diff --git a/src/formatter.test.ts b/src/formatter.test.ts
index e5e3453..0c029b3 100644
--- a/src/formatter.test.ts
+++ b/src/formatter.test.ts
@@ -413,3 +413,19 @@ describe('formatter: insights', () => {
     expect(output).toContain('Warmup overhead');
   });
 });
+
+describe('formatSession: input decomposed into fresh vs cached', () => {
+  it('shows freshly-billed input (input + cache writes) and cache reads separately', () => {
+    // input 5K + cacheCreation 10K = 15K fresh; cacheRead 50K cached.
+    const output = strip(formatSession(makeAnalysis()));
+    expect(output).toContain('Input');
+    expect(output).toContain('15.0K new');
+    expect(output).toContain('50.0K cached');
+  });
+
+  it('the headline "in" remains the full context total (fresh + cached)', () => {
+    // input+cacheRead+cacheCreation = 65K — the line a user sees as scale.
+    const output = strip(formatSession(makeAnalysis()));
+    expect(output).toContain('65.0K in');
+  });
+});
diff --git a/src/formatter.ts b/src/formatter.ts
index b226a59..43cefdf 100644
--- a/src/formatter.ts
+++ b/src/formatter.ts
@@ -162,6 +162,12 @@ export function formatSession(analysis: SessionAnalysis): string {
     const inBar = Math.round(inFrac * BAR_WIDTH);
     const outBar = BAR_WIDTH - inBar;
     lines.push(` Tokens  ${chalk.cyan('\u2588'.repeat(inBar))}${chalk.green('\u2588'.repeat(outBar))}  ${chalk.cyan(formatTokens(tokens.input + tokens.cacheRead + tokens.cacheCreation) + ' in')} ${chalk.green(formatTokens(tokens.output) + ' out')}`);
+    // Decompose "in": the headline is dominated by cheap cache reads. Show the
+    // freshly-billed input (input + cache writes) separately so the cost line
+    // is interpretable (a huge "in" at 97% cache is mostly $1.50/M reads, not
+    // $15/M fresh input).
+    const freshIn = tokens.input + tokens.cacheCreation;
+    lines.push(` Input   ${chalk.gray(formatTokens(freshIn) + ' new')} ${chalk.gray('\u00b7')} ${chalk.gray(formatTokens(tokens.cacheRead) + ' cached')}`);
   }
   const cachePct = Math.round(analysis.cacheHitRate * 100);
   lines.push(` Cache   ${renderCacheBar(analysis.cacheHitRate)}  ${cachePct}% hit`);
@@ -263,6 +269,8 @@ export function formatSessionLive(analysis: SessionAnalysis): string {
     const inBar = Math.round(inFrac * BAR_WIDTH);
     const outBar = BAR_WIDTH - inBar;
     lines.push(` Tokens  ${chalk.cyan('\u2588'.repeat(inBar))}${chalk.green('\u2588'.repeat(outBar))}  ${chalk.cyan(formatTokens(liveTok.input + liveTok.cacheRead + liveTok.cacheCreation) + ' in')} ${chalk.green(formatTokens(liveTok.output) + ' out')}`);
+    const liveFreshIn = liveTok.input + liveTok.cacheCreation;
+    lines.push(` Input   ${chalk.gray(formatTokens(liveFreshIn) + ' new')} ${chalk.gray('\u00b7')} ${chalk.gray(formatTokens(liveTok.cacheRead) + ' cached')}`);
   }
   const liveCachePct = Math.round(analysis.cacheHitRate * 100);
   lines.push(` Cache   ${renderCacheBar(analysis.cacheHitRate)}  ${liveCachePct}% hit`);