@@ -128,6 +128,32 @@ export function checkConclusion(outcome) {
 	}
 }
 
+/**
+ * One mark per outcome, so the heading reads at a glance.
+ *
+ * "## Test Companion — Run broke" and "## Test Companion — Succeeded" look the
+ * same in a crowded pull request thread, and the one that needs attention is
+ * the one that blends in.
+ *
+ * Deliberately the four GitHub itself uses for the same ideas — a tick for
+ * done, a cross for a failing test, a warning for "somebody has to look" —
+ * rather than anything decorative. An outcome this Action has never heard of,
+ * which a newer CLI could report, gets no mark rather than a crash: losing the
+ * mark is cosmetic, losing the report is not.
+ */
+const OUTCOME_MARKS = Object.freeze({
+	[OUTCOMES.SUCCEEDED]: "✅",
+	[OUTCOMES.TESTS_FAILED]: "❌",
+	[OUTCOMES.RUN_BROKE]: "💥",
+	[OUTCOMES.STOPPED_AT_LIMIT]: "🛑",
+	[OUTCOMES.COULD_NOT_RUN]: "⚠️",
+	[OUTCOMES.SKIPPED]: "⏭️",
+})
+
+export function emojiFor(outcome) {
+	return OUTCOME_MARKS[outcome] ?? ""
+}
+
 /** What a repository can ask a bad outcome to do to its build — FR-9. */
 export const BUILD_EFFECTS = Object.freeze({
 	/** Never red. The check still reports truthfully. */
 
@@ -8,6 +8,17 @@
  * preinstalled on every runner and already how this product talks to GitHub.
  */
 import { CAPABILITIES } from "./intent.mjs"
+import { emojiFor } from "./outcome.mjs"
+
+/**
+ * The outcome with its mark in front — never instead of the words, so a reader
+ * with images off, or a screen reader, still gets the whole outcome. An
+ * outcome with no mark reads exactly as it did before.
+ */
+const withMark = (outcome) => {
+	const mark = emojiFor(outcome)
+	return mark ? `${mark} ${outcome}` : String(outcome ?? "")
+}
 
 /**
  * How a later run finds the comment it wrote earlier — FR-29 wants ONE comment,
@@ -140,7 +151,9 @@ export function renderJobSummary({
 	warnings = [],
 } = {}) {
 	const lines = []
-	lines.push(`## Test Companion — ${outcome}`)
+	// The mark goes in front of the words, never instead of them: a reader with
+	// images off, or a screen reader, still gets the whole outcome.
+	lines.push(`## Test Companion — ${withMark(outcome)}`)
 	lines.push("")
 
 	// FR-43: both asking and approving need write access, so one person doing
@@ -176,7 +189,7 @@ export function renderJobSummary({
 	const rows = [
 		["Job", workflow ?? summary.workflow],
 		["Trigger", summary.event],
-		["Outcome", outcome],
+		["Outcome", withMark(outcome)],
 		["Took", summary.duration],
 		// Credits, not dollars: it is the unit the bill and the pipeline file's
 		// own ceiling are written in, so a figure in dollars would not line up
 
@@ -0,0 +1,48 @@
+import { test, describe } from "node:test"
+import assert from "node:assert/strict"
+import { OUTCOMES, emojiFor } from "../src/outcome.mjs"
+import { renderJobSummary } from "../src/report.mjs"
+
+/**
+ * The outcome should be readable at a glance.
+ *
+ * "## Test Companion — Run broke" and "## Test Companion — Succeeded" look the
+ * same in a crowded pull request thread, and the one that needs attention is
+ * the one that blends in.
+ */
+
+describe("every outcome has a mark of its own", () => {
+	test("no two outcomes share one", () => {
+		const marks = Object.values(OUTCOMES).map(emojiFor)
+		assert.equal(new Set(marks).size, marks.length, `marks collide: ${marks.join(" ")}`)
+	})
+
+	test("every outcome has one", () => {
+		for (const o of Object.values(OUTCOMES)) {
+			assert.ok(emojiFor(o), `${o} has no mark`)
+		}
+	})
+
+	test("an outcome we do not know does not break the heading", () => {
+		// A newer CLI could report something this Action has never heard of.
+		// A missing mark is a cosmetic loss; a crash is a lost report.
+		assert.equal(emojiFor("Something new"), "")
+		assert.equal(emojiFor(undefined), "")
+	})
+
+	test("the heading carries it", () => {
+		const body = renderJobSummary({ workflow: "fix", outcome: OUTCOMES.RUN_BROKE, summary: {}, runUrl: "https://run" })
+		assert.match(body.split("\n")[0], /^## Test Companion — .+ Run broke$/u)
+		assert.ok(body.split("\n")[0].includes(emojiFor(OUTCOMES.RUN_BROKE)))
+	})
+
+	test("and so does the row, so the table reads on its own", () => {
+		const body = renderJobSummary({ workflow: "fix", outcome: OUTCOMES.TESTS_FAILED, summary: {}, runUrl: "https://run" })
+		assert.match(body, new RegExp(`\\| Outcome \\| ${emojiFor(OUTCOMES.TESTS_FAILED)} Tests failed \\|`, "u"))
+	})
+
+	test("a success still reads as a success", () => {
+		const body = renderJobSummary({ workflow: "generate", outcome: OUTCOMES.SUCCEEDED, summary: {}, runUrl: "https://run" })
+		assert.match(body.split("\n")[0], /Succeeded$/)
+	})
+})
@@ -1191,7 +1191,9 @@ describe("renderComment keeps the body postable", () => {
 	test("the table still renders above the cut answer", () => {
 		const body = renderComment(view("word ".repeat(40_000)))
 		assert.ok(body.includes("| Job | fix |"))
-		assert.ok(body.includes("Test Companion — Succeeded"))
+		// The heading carries a mark in front of the words now, so match the
+		// shape rather than the exact string — see outcome-emoji.test.mjs.
+		assert.match(body, /## Test Companion — .*Succeeded/u)
 	})
 
 	// Not the answer's fault. The listed failures are capped, but a test NAME