`;
+
+export const reportsLandingMd = (): string => `# reports
+
+> Per-agent TDD-discipline reporting over real project repos. The judge replays each commit on tracked branches and scores it structurally — red-fails, green-passes, no test-deletion, no regression. The scores roll up per agent over time, with trend, failure-mode breakdown, and an exec summary fit for a quarterly readout.
+
+This is a design preview. The pipeline that ingests real repos isn't wired yet; what you can navigate today is a mockup with synthetic data:
+
+- [exec summary mockup →](/reports/demo) — single page, 1 quarter, 3 agents
+- [per-agent drill-down →](/reports/demo/agents/cursor) — trend, failure mix, recent flagged commits
+
+## what gets measured
+
+This layer measures **discipline**, not code-quality. Without hidden tests (those only exist on katas), tdd.md can't catch tautologies or weakened assertions on real repos. It *can* catch:
+
+| failure mode | what triggers it | what it costs |
+|---|---|---|
+| \`red-did-not-fail\` | commit tagged \`red:\` but tests pass | -5 / commit |
+| \`test-deleted\` | test count drops between commits | -20 / commit |
+| \`broken refactor\` | tests fail at a \`refactor:\` commit | -5 / commit |
+| \`no phase tag\` | tracked-branch commit missing \`red\\|green\\|refactor:\` | counts against phase-coverage % |
+
+The metric pair that anchors the report is **discipline-score** (0-100) + **phase-coverage %**. An agent with 0% phase-coverage doesn't *do* TDD — its score is N/A, not 0. Don't let a low-volume non-attempt look like a high-volume slip.
+
+## reading the data
+
+For management:
+- the [exec summary](/reports/demo) gives one number per agent + a narrative paragraph. Prints to one page.
+
+For team-leads:
+- the [drill-down](/reports/demo/agents/cursor) shows trend, failure-mix, streak, and the most recent flagged commits with one-click coaching links to the [Claude Code](/blog/claude-code-tdd) / [Cursor](/blog/cursor-tdd) / [Aider](/blog/aider-tdd) posts.
+
+[← back to tdd.md](/) · [the blog](/blog) · [the katas](/games)
+`;
+
+export const execSummaryMd = (): string => {
+ const totalCommits = DEMO_REPORTS.reduce((s, a) => s + a.commits, 0);
+ const tiles = DEMO_REPORTS.map(tile).join("\n");
+ return `# tdd-discipline rapport · q1 2026
+
+${mockBanner}
+
+> **Periode** ${DEMO_PERIOD} · **Scope** ${DEMO_REPOS} repos · ${totalCommits.toLocaleString()} AI-toegeschreven commits in ${escape(DEMO_ORG)}.
+
+
+${tiles}
+
+
+## wat veranderde dit kwartaal
+
+Cursor's score zakte 15 punten nadat agent-mode in maart default werd; test-deletion-incidenten stegen van 2% naar 14% van refactor-commits, geconcentreerd in de \`api-gateway\` repo. Claude Code's score steeg na invoering van phase-getagde commit-prefix in CLAUDE.md aan het einde van januari. Aider blijft stabiel hoog — auto-commit-per-edit voorkomt het meeste cross-phase bedrog vanzelf.
+
+## wat we doen
+
+- **Cursor in \`api-gateway\`**: agent-mode gedeactiveerd voor refactor-prompts, CONVENTIONS-regel "never delete a test in a refactor commit" gepind ([details →](/reports/demo/agents/cursor)).
+- **Claude Code uitrollen**: het CLAUDE.md-template dat in \`billing-service\` werkte naar de andere drie repos kopiëren.
+- **Volgende meting**: 2026-04-30, mid-Q2, om te zien of de Cursor-fix vasthoudt.
+
+## wat dit getal *niet* meet
+
+Discipline, niet code-kwaliteit. Hidden tests (zoals op de katas) bestaan niet voor productie-repos, dus *tautologische* tests en *zwak-geformuleerde* asserties blijven onzichtbaar voor de judge. Dit cijfer zegt: "de agent volgt de TDD-cyclus eerlijk". Het zegt niets over of de tests die hij schrijft het juiste beweren. Voor dat tweede signaal blijft kata-performance ([leaderboard](/leaderboard)) de proxy.
+
+---
+
+[per-agent drill-down: Claude Code](/reports/demo/agents/claude-code) · [Cursor](/reports/demo/agents/cursor) · [Aider](/reports/demo/agents/aider) · [back to /reports](/reports)
+`;
+};
+
+export const agentDrilldownMd = (slug: AgentReport["slug"]): string | null => {
+ const a = DEMO_REPORTS.find((r) => r.slug === slug);
+ if (!a) return null;
+ const arr = trendArrow(a.delta);
+ const deltaStr = a.delta > 0 ? `+${a.delta}` : `${a.delta}`;
+ const recentRows = a.recent
+ .map(
+ (r) =>
+ `| ${r.date} | \`${r.repo}\` | \`${r.sha}\` | ${r.phase} | ${r.failure} | ${r.pts} |`,
+ )
+ .join("\n");
+ return `# ${a.name} · drill-down
+
+${mockBanner}
+
+> Discipline-score **${a.score} / 100** ${arr.glyph} ${deltaStr} over ${DEMO_PERIOD}. ${a.commits.toLocaleString()} commits geanalyseerd, phase-coverage **${a.phaseCoveragePct}%**.
+
+## trend (30 dagen)
+
+
+${sparkline(a.trend)}
+
+
+${streakBox(a)}
+
+## failure-mode breakdown
+
+${bars(a.failureMix)}
+
+Top issue dit kwartaal: **${escape(a.topIssueLabel)}** (${a.topIssuePct}% van commits).
+
+## recent flagged
+
+| date | repo | sha | phase | failure | pts |
+|---|---|---|---|---|---|
+${recentRows}
+
+## coaching
+
+- ${a.slug === "claude-code" ? `[Claude Code does not do TDD by default](/blog/claude-code-tdd) — CLAUDE.md rules + fresh-context boundaries that prevent \`red-did-not-fail\`.` : a.slug === "cursor" ? `[Cursor knows how to do TDD; users skip the parts that matter](/blog/cursor-tdd) — Plan Mode, fresh chats, \`.cursor/rules\` to stop test-deletion.` : `[Aider is the closest agent to TDD on rails — until \`--auto-test\`](/blog/aider-tdd) — keep auto-test off for green commits, on for refactor.`}
+- [Tweag's TDD handbook needs a judge](/blog/tweag-handbook-tdd) — why local green isn't enough.
+
+---
+
+[← exec summary](/reports/demo) · [back to /reports](/reports)
+`;
+};
diff --git a/src/server.ts b/src/server.ts
index b419e499297c446a7f195c236c877618adb3d930..01703187c69b01d6cc65a9de2b1777c4851d28ff 100644
--- a/src/server.ts
+++ b/src/server.ts
@@ -5,6 +5,12 @@ import { parseCommit, computeProgress, type Phase } from "./commits";
import { loadGame, listGames } from "./games";
import { judge } from "./judge";
import { latestRun, allLatestRuns } from "./db";
+import {
+ reportsLandingMd,
+ execSummaryMd,
+ agentDrilldownMd,
+ DEMO_REPORTS,
+} from "./reports";
const HOME_MD = "./content/home.md";
const GAME_DIR = "./content/games";
@@ -782,6 +788,46 @@ ${rows}
return htmlResponse(html);
},
+ "/reports": async () => {
+ const html = await renderPage({
+ title: "Reports — tdd.md",
+ description: "Per-agent TDD-discipline reporting over real project repos: trend, failure-mode breakdown, and an exec summary fit for a quarterly readout.",
+ bodyMarkdown: reportsLandingMd(),
+ ogPath: "https://tdd.md/reports",
+ noindex: true,
+ });
+ return htmlResponse(html);
+ },
+
+ "/reports/demo": async () => {
+ const html = await renderPage({
+ title: "TDD-discipline rapport · Q1 2026 (demo) — tdd.md",
+ description: "Mockup of the management-level TDD-discipline report — single page, three agents, with trend and narrative.",
+ bodyMarkdown: execSummaryMd(),
+ ogPath: "https://tdd.md/reports/demo",
+ noindex: true,
+ });
+ return htmlResponse(html);
+ },
+
+ "/reports/demo/agents/:slug": async (req) => {
+ const slug = req.params.slug as (typeof DEMO_REPORTS)[number]["slug"];
+ const md = agentDrilldownMd(slug);
+ if (!md) {
+ const html = await renderNotFound(`/reports/demo/agents/${slug}`);
+ return htmlResponse(html, 404);
+ }
+ const entry = DEMO_REPORTS.find((r) => r.slug === slug)!;
+ const html = await renderPage({
+ title: `${entry.name} drill-down (demo) — tdd.md`,
+ description: `Per-agent drill-down mockup for ${entry.name}: trend, failure-mode breakdown, recent flagged commits with coaching links.`,
+ bodyMarkdown: md,
+ ogPath: `https://tdd.md/reports/demo/agents/${slug}`,
+ noindex: true,
+ });
+ return htmlResponse(html);
+ },
+
"/guides": async () => {
const rows = ALL_GUIDES
.map((g) => `| [${g.title}](/guides/${g.slug}) | ${g.description} |`)