cert-coach-web 0.1.0 → 0.1.1

package/AGENTS.md

@@ -54,7 +54,9 @@ Load the extension: `pi --extension ./.pi/extensions/cert-coach.ts`.
 
 ## Current certs
 
-- `DP600/` — Microsoft Fabric Analytics Engineer (DP-600). Blueprint verified 2026-10-21 (skills measured 2026-10-19). Bank: 117 items, 74 gradeable, 110 verified (107 confirmed, 3 corrected), 7 unverifiable, 0 pending; 12 authored gap questions are medium confidence and need a targeted docs pass. Free half: 21/43 pages. `DP600/theory.md` is a manual `DEEP_PASS`; `build_theory` refuses to overwrite it. Direct Lake on OneLake has **no** DirectQuery fallback; only the SQL analytics endpoint variant does.
-- `PL300/` — Microsoft Power BI Data Analyst (PL-300). Blueprint rebuilt with the current **four** domains (the old three-domain/45–50% visualize split was obsolete). Bank: 239 items, 159 gradeable, 236 verified (235 confirmed, 1 corrected), 0 pending, 3 unverifiable; 4 authored gap questions are medium confidence; free half 47/95 pages.
+- `DP600/` — Microsoft Fabric Analytics Engineer (DP-600). Blueprint verified 2026-10-21 (skills measured 2026-10-19). Bank: 117 items, 74 gradeable, 110 verified (105 confirmed, 5 corrected), 7 unverifiable, 0 pending; all 16 authored gap questions human-reviewed. Free half: 21/43 pages. `DP600/theory.md` is a manual `DEEP_PASS`. Direct Lake on OneLake has **no** DirectQuery fallback; only the SQL analytics endpoint variant does.
+- `PL300/` — Microsoft Power BI Data Analyst (PL-300). Current **four** domains (25–30/25–30/25–30/15–20). Bank: 234 items, 155 gradeable, 231 verified (230 confirmed, 1 corrected), 3 unverifiable, 0 pending. Free half 47/95 pages.
+- `DATABRICKSDA/` — Databricks Certified Data Analyst Associate (DA-ASSOC, 45 MCQ / 90 min). Blueprint verified 2026-10-26: nine domains 11/8/5/20/15/16/12/5/8. **Dump ingested, verified, and gap-filled 2026-10-26**: bank **106 items, 91 gradeable, all answer-verified** (88 confirmed · 3 corrected · 15 unverifiable), **0 pending**, **0 unmapped**; **36-node DAG, all 36 used**, a hands-on `lab` per node, thin-bank gate `minCoverage: 0.5`. Every domain is now within ~±5 pts of blueprint share. 10 items are medium confidence — a targeted docs pass would firm them up. Note: docs have rebranded Delta Sharing → **OpenSharing** and Genie → a family (Genie One / Genie Agents / Genie Code); stems use exam-era wording, keywords carry both.
+- `SAPC02/` — AWS Certified Solutions Architect – Professional (SAP-C02: 75 questions delivered = 65 scored + 10 unscored, 180 min, pass 750/1000, $300, 3-yr). Blueprint **fetched and verified 2026-10-26**: four domains 26/29/25/20. Bootstrap starter bank: 20 gradeable scenario items (all doc-verified), 31-node DAG with a hands-on AWS `lab` per node, thin-bank gate `minCoverage: 0.5`. Scenario/trade-off exam; Domain 2 is heaviest (genAI/Bedrock, event-driven, Control Tower), and scored content is 65 — not 75. ⚠️ **SAP-C02 last test day 2026-11-16; SAP-C03 registration opens 2026-10-27** (`cert_meta.lifecycle`) — confirm the booked version. Dump still to ingest.
 
 Do not trust counts written in prose; `cert_status` and `validate.mjs` are the source of truth.

package/README.md

@@ -23,9 +23,10 @@ Counts are live in `progress.json` / `/api/status`; this is a snapshot.
 |---|---:|---:|---|---:|
 | **DP-600** | 117 | 74 | 105 confirmed · 5 corrected · 7 unverifiable · 0 pending | 22% |
 | **PL-300** | 234 | 155 | 230 confirmed · 1 corrected · 3 unverifiable · 0 pending | 23% |
+| **Databricks DA** | 24 | 24 | 24 confirmed · 0 corrected · 0 unverifiable · 0 pending | bootstrap |
 
 - **Works today:** TUI (`cert_*` tools) + web console (Study, Path, Dashboard, Theory, Logs, teaching chat), Chat v2 (read + research + verify), full mock, live Mermaid, browser cert scaffold, `pi` auto-install, OpenRouter free-model config.
-- **Green:** 44 tests, both certs `validate` PASS, `npm pack` 61 files / no banks.
+- **Green:** 67 tests, both certs `validate` PASS, `npm pack` 64 files / no banks.
 - **Next:** npm publish; generated-question review; optional extension slimming.
 
 ## Quickstart
@@ -189,7 +190,7 @@ server/                    web console server (server-owned StudyLoop, SSE, auth
 public/                    study SPA (Study, Path, Dashboard, Theory, Logs, chat)
 bin/                       web console entry point
 tests/                     deterministic core/server/web/chat/mock tests
-DP600/  PL300/             internal data (shareable as-is; progress.json is owner-only)
+DP600/  PL300/  DATABRICKSDA/  SAPC02/   internal data (shareable as-is; progress.json is owner-only)
 ```
 
 ## Move your own workspace to another PC
@@ -200,7 +201,8 @@ For a private move (your own second machine), use the archive script:
 powershell -ExecutionPolicy Bypass -File engine/scripts/export_private.ps1
 ```
 
-It writes a zip to your Desktop containing the engine **and** `DP600/`, `PL300/`
+It writes a zip to your Desktop containing the engine **and** `DP600/`, `PL300/`,
+`DATABRICKSDA/`, `SAPC02/`
 including `raw/`, verified `bank.json`, and `vision_assets/`. Progress is reset
 in the archive and study logs are removed by default. Page state (`.cert-coach/`)
 and `node_modules/` are excluded. Use `-KeepProgress` to include attempts/history.
@@ -260,6 +262,8 @@ npm run test:e2e:real
 
 The Chrome test checks question render, no key before submit, feedback, Next, Path, and chat/tool cards. Playwright uses the installed Chrome channel. The header **+ New** button scaffolds a workspace from the browser; research and ingest still run through the agent (see `BROWSER_CERT_CREATION.md`).
 
+Long chat sessions are compacted automatically so a small-context model never dies mid-study. After **12 learner turns** the server starts a fresh pi session and carries a recap of the earlier questions forward; if the model reports a context-limit error anyway, the same compaction happens on the spot and the chat keeps working. Tune the threshold with `CERT_COACH_CHAT_MAX_TURNS` (or `maxTurns` on `ChatBridge`). The chat log shows a dashed *session compacted* line when it happens.
+
 Mock: choose **10 / 20 / 65 questions** and **untimed / 90 / 130 minutes** before starting. The summary is isolated from `progress.json`.
 
 Mermaid fences in agent chat/theory now render as live diagrams via a local `public/vendor/mermaid.min.js`.
@@ -317,5 +321,5 @@ Pi: `cert_setup`, `cert_status`, `cert_ask`, `cert_next`, `cert_answer`, `cert_v
 - **DP-600:** 74/74 gradeable items verified; ledger 117 items (105 confirmed, 5 corrected, 7 unverifiable, 0 pending). Free half only (21/43 pages) — full coverage needs a paid account and `MAX_PAGES`.
 - **PL-300:** 155/155 gradeable items verified; ledger 234 items (230 confirmed, 1 corrected, 3 unverifiable, 0 pending). Free half (47/95 pages). `manage_secure` is ~14.5% of the bank vs the 15–20% blueprint.
 - **Image items are not lost.** 42 DP-600 / 79 PL-300 items depend on exhibits or code images. They are marked `needs_vision` with asset URLs; a vision pass can recover them.
-- **Data is internal-shareable, as-is.** `DP600/` and `PL300/` (banks, `raw/`, `vision_assets/`) ship in the *internal* package only, never the public npm tarball. Scraping is still subject to the source site's terms, so keep the internal export inside your org. `progress.json` is owner-only and reset on export.
+- **Data is internal-shareable, as-is.** `DP600/`, `PL300/`, `DATABRICKSDA/`, and `SAPC02/` (banks, `raw/`, `vision_assets/`) ship in the *internal* package only, never the public npm tarball. Scraping is still subject to the source site's terms, so keep the internal export inside your org. `progress.json` is owner-only and reset on export.
 - **The extension is pi-only.** The portable protocol is the certification-coach skill; the deterministic state layer requires pi.

package/engine/core/cert-core.mjs

@@ -143,12 +143,17 @@ export function computeReadiness(c) {
   const gradeable = c.bank.filter((q) => q.gradeable).length || 1;
   const seen = Object.values(by).reduce((a, b) => a + b.seen, 0);
   const coverage = seen / gradeable;
-  const label = value >= 0.85 && coverage >= 0.8 ? "ready" : value >= 0.65 ? "borderline" : "not_ready";
+  const gate = c.meta.readiness || {};
+  const minAccuracy = Number.isFinite(gate.minAccuracy) ? gate.minAccuracy : 0.85;
+  const minCoverage = Number.isFinite(gate.minCoverage) ? gate.minCoverage : 0.8;
+  const label =
+    value >= minAccuracy && coverage >= minCoverage ? "ready" : value >= 0.65 ? "borderline" : "not_ready";
   c.progress.readiness = {
     method: "blueprint_weighted_last_answer_accuracy",
     value: Number(value.toFixed(4)),
     label,
     coverage: Number(coverage.toFixed(4)),
+    gate: { minAccuracy, minCoverage },
     by_domain: Object.fromEntries(
       Object.entries(by).map(([k, v]) => [k, { ...v, accuracy: v.seen ? Number((v.lastCorrect / v.seen).toFixed(3)) : null }])
     ),

package/engine/extractor.js

@@ -29,7 +29,7 @@
   const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
 
   function detectCert() {
-    if (INJECTED_CERT && INJECTED_CERT !== "{{CERT}}") return INJECTED_CERT;
+    if (INJECTED_CERT && !INJECTED_CERT.includes("{")) return INJECTED_CERT;
     if (window.CERT_CODE) return window.CERT_CODE;
     const text = `${location.href} ${document.title} ${document.body?.innerText?.slice(0, 4000) || ""}`;
     const m = text.match(/\b([A-Z]{2,4}-\d{3})\b/);

package/engine/scripts/export_private.ps1

@@ -17,10 +17,9 @@ if (Test-Path $stage) { Remove-Item $stage -Recurse -Force }
 New-Item -ItemType Directory -Force $stage | Out-Null
 
 $items = @(
-  "AGENTS.md", "README.md", ".pi/skills/", ".pi/prompts/", "EERO_PORT.md",
-  "cert-coach.ts", "extractor.js", "package.json",
-  "core", "server", "public", "prompts", "bin", "scripts", ".agents",
-  "DP600", "PL300"
+  "AGENTS.md", "README.md", "EERO_PORT.md", "BROWSER_CERT_CREATION.md", "LICENSE", "package.json",
+  ".pi", "engine", "web", "tests",
+  "DP600", "PL300", "DATABRICKSDA"
 )
 foreach ($item in $items) {
   $src = Join-Path $Repo $item
@@ -28,10 +27,13 @@ foreach ($item in $items) {
     Copy-Item $src (Join-Path $stage $item) -Recurse -Force
   }
 }
+# Never ship dependencies; they are reinstalled on the other PC.
+Get-ChildItem $stage -Recurse -Directory -Filter "node_modules" -ErrorAction SilentlyContinue |
+  Remove-Item -Recurse -Force -ErrorAction SilentlyContinue
 
 # Reset learner state in the archive (not in the real workspace) unless asked.
 if (-not $KeepProgress) {
-  foreach ($cert in @("DP600", "PL300")) {
+  foreach ($cert in @("DP600", "PL300", "DATABRICKSDA")) {
     $certDir = Join-Path $stage $cert
     if (-not (Test-Path $certDir)) { continue }
     $metaPath = Join-Path $certDir "cert_meta.json"
@@ -80,7 +82,7 @@ if (-not $KeepProgress) {
 @"
 PRIVATE CERT COACH WORKSPACE — DO NOT REDISTRIBUTE
 
-This archive includes the engine plus DP600/PL300 cert data (raw dumps,
+This archive includes the engine plus DP600, PL300, and DATABRICKSDA cert data (raw dumps,
 verified banks, vision assets). Progress is reset; study logs are removed.
 
 On the other PC:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cert-coach-web",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "Local web study console for a pi certification coach: server-owned quiz loop, images, and teaching chat.",
   "type": "module",
   "private": false,
@@ -12,7 +12,7 @@
   },
   "scripts": {
     "start": "node web/server/main.mjs",
-    "test": "node --test tests/core.test.mjs tests/server.test.mjs tests/web.test.mjs tests/chat.test.mjs tests/mock.test.mjs",
+    "test": "node --test tests/core.test.mjs tests/server.test.mjs tests/web.test.mjs tests/chat.test.mjs tests/mock.test.mjs tests/spa.test.mjs tests/spa-render.test.mjs tests/ingest.test.mjs tests/validate.test.mjs tests/extractor.test.mjs",
     "pack:check": "npm pack --dry-run",
     "test:e2e:chrome": "node tests/e2e-chrome.mjs",
     "test:package": "node tests/package-smoke.mjs",
@@ -64,6 +64,7 @@
     "access": "public"
   },
   "devDependencies": {
-    "@playwright/test": "^1.64.1"
+    "@playwright/test": "^1.64.1",
+    "jsdom": "^30.1.3"
   }
 }

package/web/public/app.js

@@ -1,3 +1,6 @@
+import { safeHttpUrl, imageSrc } from "./spa-logic.js";
+import { markdownFragment } from "./spa-render.js";
+
 const $ = (id) => document.getElementById(id);
 
 const el = {
@@ -165,22 +168,6 @@ function setPhase(phase) {
   el.phasePill.textContent = phase || "idle";
 }
 
-function safeHttpUrl(value) {
-  try {
-    const u = new URL(value);
-    return u.protocol === "http:" || u.protocol === "https:" ? u.href : "";
-  } catch {
-    return "";
-  }
-}
-
-function imageSrc(im) {
-  const src = String(im?.src || "");
-  if (/^https?:\/\//i.test(src)) return "/api/asset?src=" + encodeURIComponent(src);
-  if (src.startsWith("/assets/")) return src;
-  return "";
-}
-
 function renderImages(images) {
   el.images.replaceChildren();
   for (const im of images || []) {
@@ -456,108 +443,8 @@ async function startPractice(mode = "practice") {
   }
 }
 
-function inlineNodes(text) {
-  const frag = document.createDocumentFragment();
-  const re = /(`[^`]+`|\*\*[^*]+\*\*|\[[^\]]+\]\(https?:\/\/[^)\s]+\))/g;
-  let last = 0;
-  for (const m of text.matchAll(re)) {
-    if (m.index > last) frag.append(document.createTextNode(text.slice(last, m.index)));
-    const tok = m[0];
-    if (tok.startsWith("`")) {
-      const code = document.createElement("code");
-      code.textContent = tok.slice(1, -1);
-      frag.append(code);
-    } else if (tok.startsWith("**")) {
-      const b = document.createElement("strong");
-      b.textContent = tok.slice(2, -2);
-      frag.append(b);
-    } else {
-      const mm = tok.match(/^\[([^\]]+)\]\((https?:\/\/[^)\s]+)\)$/);
-      const a = document.createElement("a");
-      a.textContent = mm?.[1] || tok;
-      a.href = safeHttpUrl(mm?.[2] || "");
-      a.target = "_blank";
-      a.rel = "noreferrer";
-      frag.append(a);
-    }
-    last = m.index + tok.length;
-  }
-  if (last < text.length) frag.append(document.createTextNode(text.slice(last)));
-  return frag;
-}
-
-function markdownFragment(text) {
-  const frag = document.createDocumentFragment();
-  const lines = String(text || "").split(/\r?\n/);
-  let i = 0;
-  let para = [];
-  const flush = () => {
-    if (!para.length) return;
-    const p = document.createElement("p");
-    p.append(inlineNodes(para.join(" ")));
-    frag.append(p);
-    para = [];
-  };
-  while (i < lines.length) {
-    const line = lines[i];
-    if (/^```/.test(line)) {
-      flush();
-      const lang = line.slice(3).trim();
-      i++;
-      const code = [];
-      while (i < lines.length && !/^```/.test(lines[i])) code.push(lines[i++]);
-      if (i < lines.length) i++;
-      const pre = document.createElement("pre");
-      if (lang === "mermaid") pre.className = "diagram";
-      const c = document.createElement("code");
-      c.textContent = code.join("\n");
-      pre.append(c);
-      frag.append(pre);
-      continue;
-    }
-    const h = line.match(/^(#{1,3})\s+(.*)$/);
-    if (h) {
-      flush();
-      const elh = document.createElement("h" + h[1].length);
-      elh.append(inlineNodes(h[2]));
-      frag.append(elh);
-      i++;
-      continue;
-    }
-    if (/^\s*[-*]\s+/.test(line)) {
-      flush();
-      const ul = document.createElement("ul");
-      while (i < lines.length && /^\s*[-*]\s+/.test(lines[i])) {
-        const li = document.createElement("li");
-        li.append(inlineNodes(lines[i].replace(/^\s*[-*]\s+/, "")));
-        ul.append(li);
-        i++;
-      }
-      frag.append(ul);
-      continue;
-    }
-    if (/^>\s?/.test(line)) {
-      flush();
-      const quote = document.createElement("blockquote");
-      quote.append(inlineNodes(line.replace(/^>\s?/, "")));
-      frag.append(quote);
-      i++;
-      continue;
-    }
-    if (!line.trim()) {
-      flush();
-      i++;
-      continue;
-    }
-    para.push(line);
-    i++;
-  }
-  flush();
-  return frag;
-}
-
 function renderMarkdownInto(node, text) {
-  node.replaceChildren(markdownFragment(text));
+  node.replaceChildren(markdownFragment(document, text));
   void renderMermaidInto(node);
 }
 
@@ -569,7 +456,7 @@ function chatScroll(force = false) {
 
 function appendChatMessage(role, text) {
   const div = document.createElement("div");
-  div.className = "chat-msg " + (role === "user" ? "user" : "assistant");
+  div.className = "chat-msg " + (role === "user" ? "user" : role === "system" ? "system" : "assistant");
   if (role === "assistant") renderMarkdownInto(div, text);
   else div.textContent = text;
   el.chatLog.append(div);
@@ -680,6 +567,16 @@ function handleChatEvent(event) {
       showNotice(event.error || "Chat failed.");
       break;
     }
+    case "chat.rotated": {
+      // The server compacted the session before it ran out of context.
+      appendChatMessage(
+        "system",
+        event.reason === "context-overflow"
+          ? "Context limit reached - earlier turns were summarised. Keep going."
+          : "Long session - starting a fresh context with a recap of earlier turns."
+      );
+      break;
+    }
     default:
       break;
   }

package/web/public/index.html

@@ -52,7 +52,7 @@
           <h1>Ready when you are</h1>
           <p id="idle-note" class="muted"></p>
           <div class="mock-controls">
-            <label>Mock <select id="mock-size"><option value="10">10 questions</option><option value="20">20 questions</option><option value="65" selected>65 questions</option></select></label>
+            <label>Mock <select id="mock-size"><option value="10">10 questions</option><option value="20">20 questions</option><option value="45">45 questions</option><option value="65" selected>65 questions</option></select></label>
             <label>Time <select id="mock-minutes"><option value="0">untimed</option><option value="90">90 min</option><option value="130" selected>130 min</option></select></label>
           </div>
           <div class="actions">

package/web/public/spa-logic.js

@@ -0,0 +1,20 @@
+// Pure, DOM-free SPA helpers. Kept in a module so they are unit-testable in
+// node (app.js itself touches `document` at load, so it cannot be imported).
+
+/** Allow only http(s) URLs; anything else (javascript:, data:, relative) -> "". */
+export function safeHttpUrl(value) {
+  try {
+    const u = new URL(value);
+    return u.protocol === "http:" || u.protocol === "https:" ? u.href : "";
+  } catch {
+    return "";
+  }
+}
+
+/** Remote exhibits go through the guarded proxy; local /assets/ pass through. */
+export function imageSrc(im) {
+  const src = String(im?.src || "");
+  if (/^https?:\/\//i.test(src)) return "/api/asset?src=" + encodeURIComponent(src);
+  if (src.startsWith("/assets/")) return src;
+  return "";
+}

package/web/public/spa-render.js

@@ -0,0 +1,107 @@
+// DOM markdown renderer, extracted from app.js so it can be unit-tested in
+// jsdom. `doc` is injected (rather than using the global) so the same code runs
+// in the browser (documents) and in tests (jsdom window.document).
+import { safeHttpUrl } from "./spa-logic.js";
+
+const INLINE = /(`[^`]+`|\*\*[^*]+\*\*|\[[^\]]+\]\(https?:\/\/[^)\s]+\))/g;
+
+/** Inline markdown (code, bold, http links) -> DocumentFragment. */
+export function inlineNodes(doc, text) {
+  const frag = doc.createDocumentFragment();
+  let last = 0;
+  for (const m of text.matchAll(INLINE)) {
+    if (m.index > last) frag.append(doc.createTextNode(text.slice(last, m.index)));
+    const tok = m[0];
+    if (tok.startsWith("`")) {
+      const code = doc.createElement("code");
+      code.textContent = tok.slice(1, -1);
+      frag.append(code);
+    } else if (tok.startsWith("**")) {
+      const b = doc.createElement("strong");
+      b.textContent = tok.slice(2, -2);
+      frag.append(b);
+    } else {
+      const mm = tok.match(/^\[([^\]]+)\]\((https?:\/\/[^)\s]+)\)$/);
+      const a = doc.createElement("a");
+      a.textContent = mm?.[1] || tok;
+      a.href = safeHttpUrl(mm?.[2] || "");
+      a.target = "_blank";
+      a.rel = "noreferrer";
+      frag.append(a);
+    }
+    last = m.index + tok.length;
+  }
+  if (last < text.length) frag.append(doc.createTextNode(text.slice(last)));
+  return frag;
+}
+
+/** Block markdown (fenced code, headings, lists, quotes, paragraphs) -> DocumentFragment. */
+export function markdownFragment(doc, text) {
+  const frag = doc.createDocumentFragment();
+  const lines = String(text || "").split(/\r?\n/);
+  let i = 0;
+  let para = [];
+  const flush = () => {
+    if (!para.length) return;
+    const p = doc.createElement("p");
+    p.append(inlineNodes(doc, para.join(" ")));
+    frag.append(p);
+    para = [];
+  };
+  while (i < lines.length) {
+    const line = lines[i];
+    if (/^```/.test(line)) {
+      flush();
+      const lang = line.slice(3).trim();
+      i++;
+      const code = [];
+      while (i < lines.length && !/^```/.test(lines[i])) code.push(lines[i++]);
+      if (i < lines.length) i++;
+      const pre = doc.createElement("pre");
+      if (lang === "mermaid") pre.className = "diagram";
+      const c = doc.createElement("code");
+      c.textContent = code.join("\n");
+      pre.append(c);
+      frag.append(pre);
+      continue;
+    }
+    const h = line.match(/^(#{1,3})\s+(.*)$/);
+    if (h) {
+      flush();
+      const elh = doc.createElement("h" + h[1].length);
+      elh.append(inlineNodes(doc, h[2]));
+      frag.append(elh);
+      i++;
+      continue;
+    }
+    if (/^\s*[-*]\s+/.test(line)) {
+      flush();
+      const ul = doc.createElement("ul");
+      while (i < lines.length && /^\s*[-*]\s+/.test(lines[i])) {
+        const li = doc.createElement("li");
+        li.append(inlineNodes(doc, lines[i].replace(/^\s*[-*]\s+/, "")));
+        ul.append(li);
+        i++;
+      }
+      frag.append(ul);
+      continue;
+    }
+    if (/^>\s?/.test(line)) {
+      flush();
+      const quote = doc.createElement("blockquote");
+      quote.append(inlineNodes(doc, line.replace(/^>\s?/, "")));
+      frag.append(quote);
+      i++;
+      continue;
+    }
+    if (!line.trim()) {
+      flush();
+      i++;
+      continue;
+    }
+    para.push(line);
+    i++;
+  }
+  flush();
+  return frag;
+}

package/web/public/styles.css

@@ -324,6 +324,7 @@ a:hover { color: #183677; }
 .chat-msg { max-width: 92%; padding: 9px 11px; border-radius: 11px; font-size: 12px; line-height: 1.5; overflow-wrap: anywhere; }
 .chat-msg.user { align-self: flex-end; background: var(--cobalt-soft); border: 1px solid #cbd7f7; color: #183677; }
 .chat-msg.assistant { align-self: flex-start; background: var(--surface-2); border: 1px solid var(--rule); }
+.chat-msg.system { align-self: center; max-width: 100%; background: transparent; border: 1px dashed var(--rule); color: var(--ink-3, #6b6257); font-size: 11px; font-style: italic; padding: 4px 10px; }
 .chat-msg.assistant p { margin: 0 0 7px; }
 .chat-msg.assistant p:last-child { margin-bottom: 0; }
 .chat-msg.assistant h1, .chat-msg.assistant h2, .chat-msg.assistant h3 { font-family: var(--font-display); font-size: 16px; margin: 10px 0 5px; }

package/web/server/chat.mjs

@@ -21,6 +21,15 @@ function textOf(message) {
     .join("");
 }
 
+// A long study session eventually overflows a small-context model. When the child
+// dies for that reason we compact instead of surfacing a dead chat.
+const CONTEXT_OVERFLOW = /context (?:window|length|size)|context (?:is )?too (?:big|large|long)|maximum context|token limit|too many tokens|prompt is too long/i;
+const DEFAULT_MAX_TURNS = 12;
+
+function isContextOverflow(message) {
+  return CONTEXT_OVERFLOW.test(String(message || ""));
+}
+
 function contextBlock(state) {
   const lines = [];
   lines.push(`[cert] ${state?.cert || "?"} · ${state?.mode || "practice"} · phase=${state?.phase || "idle"}`);
@@ -46,7 +55,7 @@ function contextBlock(state) {
 }
 
 export class ChatBridge extends EventEmitter {
-  constructor({ root, cert, cliPath, sessionDir = "", session = false }) {
+  constructor({ root, cert, cliPath, sessionDir = "", session = false, maxTurns }) {
     super();
     this.root = root;
     this.cert = cert;
@@ -56,6 +65,14 @@ export class ChatBridge extends EventEmitter {
     this.rpc = null;
     this.busy = false;
     this.history = [];
+    // Context hygiene: rotate the pi child before it outgrows a small-context model,
+    // carrying a short recap of earlier turns into the fresh session.
+    const configured = Number(maxTurns ?? process.env.CERT_COACH_CHAT_MAX_TURNS ?? DEFAULT_MAX_TURNS);
+    this.maxTurns = Number.isFinite(configured) && configured > 0 ? Math.floor(configured) : DEFAULT_MAX_TURNS;
+    this.turns = 0;
+    this.rotations = 0;
+    this.recap = "";
+    this.lastRotateReason = "";
     this.error = cliPath ? null : "Teaching chat is unavailable: no pi CLI found. Pass --pi <pi|path> or set PI_BIN.";
     this._assistant = "";
     this._tools = new Map();
@@ -67,6 +84,10 @@ export class ChatBridge extends EventEmitter {
       running: Boolean(this.rpc?.running),
       busy: this.busy,
       error: this.error,
+      turns: this.turns,
+      maxTurns: this.maxTurns,
+      rotations: this.rotations,
+      lastRotateReason: this.lastRotateReason,
       stderr: (this.rpc?.stderr || "").slice(-1200),
     };
   }
@@ -95,8 +116,19 @@ export class ChatBridge extends EventEmitter {
     rpc.on("event", (ev) => this._onRpc(ev));
     rpc.on("exit", (e) => {
       this.busy = false;
-      this.error = e.message;
       this.rpc = null;
+      if (isContextOverflow(e.message)) {
+        // Compact rather than dead-end: keep the learner's thread alive.
+        this.recap = this._buildRecap();
+        this.turns = 0;
+        this.rotations += 1;
+        this.lastRotateReason = "context-overflow";
+        this.error = null;
+        this._emitEvent({ type: "chat.rotated", reason: "context-overflow", rotations: this.rotations, hasRecap: Boolean(this.recap) });
+        this._emitEvent({ type: "chat.error", error: "Context limit reached. Earlier turns were summarised - your next message continues the conversation.", recoverable: true });
+        return;
+      }
+      this.error = e.message;
       this._emitEvent({ type: "chat.error", error: e.message });
     });
     await rpc.start();
@@ -161,17 +193,55 @@ export class ChatBridge extends EventEmitter {
     }
   }
 
+  /** Short recap of earlier turns, injected once into a fresh session. */
+  _buildRecap(limit = 6) {
+    const users = this.history.filter((m) => m.role === "user").slice(-limit);
+    const lastAssistant = [...this.history].reverse().find((m) => m.role === "assistant");
+    const lines = [];
+    if (users.length) {
+      lines.push("Earlier in this session the learner asked:");
+      for (const m of users) lines.push(`- ${String(m.text).replace(/\s+/g, " ").slice(0, 160)}`);
+    }
+    if (lastAssistant?.text) {
+      lines.push(`Where we left off: ${String(lastAssistant.text).replace(/\s+/g, " ").slice(0, 400)}`);
+    }
+    return lines.join("\n");
+  }
+
+  /** Stop the child so the next prompt starts a fresh pi session. */
+  async rotate(reason = "turn-limit") {
+    if (this.rpc) {
+      try { await this.rpc.stop(); } catch { /* ignore */ }
+      this.rpc = null;
+    }
+    this.recap = this._buildRecap();
+    this.turns = 0;
+    this.rotations += 1;
+    this.lastRotateReason = reason;
+    this.busy = false;
+    this._emitEvent({ type: "chat.rotated", reason, rotations: this.rotations, hasRecap: Boolean(this.recap) });
+    return this.recap;
+  }
+
   async prompt(text, state) {
     const clean = String(text || "").trim();
     if (!clean) throw new CoreError("EMPTY_PROMPT", "Message is empty.");
+    if (this.turns >= this.maxTurns) await this.rotate("turn-limit");
     await this.ensure();
     this.history.push({ role: "user", text: clean, at: new Date().toISOString() });
     if (this.history.length > 200) this.history = this.history.slice(-200);
     this._emitEvent({ type: "chat.user", text: clean });
     this.busy = true;
     this.error = null;
-    const message = `${contextBlock(state)}\n\n[learner]\n${clean}`;
-    const disposition = await this.rpc.prompt(message);
+    const parts = [];
+    if (this.recap) {
+      parts.push(`[session recap - earlier turns were summarised to keep this conversation responsive]\n${this.recap}`);
+      this.recap = "";
+    }
+    parts.push(contextBlock(state));
+    parts.push(`[learner]\n${clean}`);
+    this.turns += 1;
+    const disposition = await this.rpc.prompt(parts.join("\n\n"));
     return { disposition };
   }
 

package/web/server/http.mjs

@@ -386,6 +386,12 @@ export function createStudyServer({ root, loop, token = "", chat = null, runtime
       if (method === "GET" && url.pathname === "/app.js") {
         return sendFile(res, path.join(PUBLIC, "app.js"), "text/javascript; charset=utf-8");
       }
+      if (method === "GET" && url.pathname === "/spa-logic.js") {
+        return sendFile(res, path.join(PUBLIC, "spa-logic.js"), "text/javascript; charset=utf-8");
+      }
+      if (method === "GET" && url.pathname === "/spa-render.js") {
+        return sendFile(res, path.join(PUBLIC, "spa-render.js"), "text/javascript; charset=utf-8");
+      }
       if (method === "GET" && url.pathname === "/styles.css") {
         return sendFile(res, path.join(PUBLIC, "styles.css"), "text/css; charset=utf-8");
       }