#!/usr/bin/env python3 """ nav-tree spike validation — 4-scenario 2-test - vanilla vs CDP - real-user simulation (5 pages, with mouse) vs direct-link bot (1 page) """ import sys, subprocess, time, os, json from playwright.sync_api import sync_playwright BASE = "http://127.0.0.1:3457/spikes/nav-tree/site" DEMO = "http://127.0.0.1:3457/spikes/nav-tree/demo.html" PORT = 9360 CHROME = "/usr/bin/google-chrome" def run(condition, scenario, use_cdp=False): """Run one scenario. condition='vanilla'|'cdp', scenario='realuser'|'directbot'""" print(f"\n--- {condition.upper()} / {scenario.upper()} ---") if use_cdp: udir = f"/tmp/cdp_navtree_{condition}_{scenario}" proc = subprocess.Popen([ CHROME, "--headless=new", "--no-sandbox", "--disable-gpu", f"--remote-debugging-port={PORT}", f"--user-data-dir={udir}", "about:blank" ], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) time.sleep(3) try: with sync_playwright() as p: browser = p.chromium.connect_over_cdp(f"http://localhost:{PORT}") ctx = browser.contexts[0] if browser.contexts else browser.new_context() page = ctx.pages[0] if ctx.pages else ctx.new_page() result = _run_scenario(page, scenario) browser.close() finally: proc.terminate() time.sleep(0.5) if proc.poll() is None: proc.kill() else: with sync_playwright() as p: b = p.chromium.launch(headless=True, args=[ "--no-sandbox", "--disable-setuid-sandbox", "--disable-dev-shm-usage" ]) ctx = b.new_context() page = ctx.new_page() result = _run_scenario(page, scenario) b.close() print(f" pageCount: {result['pageCount']}") print(f" childCount: {result['childCount']}") print(f" rootPath: {result['rootPath']}") print(f" childPaths: {result['childPaths']}") print(f" interactionCount: {result['interactionCount']}") print(f" verdict: {result['verdict']}") return result def _run_scenario(page, scenario): if scenario == "realuser": # Simulate real user: visit pages in order, with brief dwell time page.goto(f"{BASE}/home.html", timeout=15000) page.wait_for_timeout(800) # Click some links (with mouse movement) page.mouse.move(400, 300) page.wait_for_timeout(200) try: page.click("#link-products", timeout=3000) except: page.goto(f"{BASE}/products.html", timeout=15000) page.wait_for_timeout(1000) page.mouse.move(500, 400) page.wait_for_timeout(200) try: page.click("#link-product-42", timeout=3000) except: page.goto(f"{BASE}/product.html?id=42", timeout=15000) page.wait_for_timeout(800) page.mouse.move(600, 500) page.wait_for_timeout(200) try: page.click("#btn-add-cart", timeout=3000) except: pass page.wait_for_timeout(500) try: page.click("#link-cart", timeout=3000) except: page.goto(f"{BASE}/cart.html", timeout=15000) page.wait_for_timeout(800) page.mouse.move(700, 350) page.wait_for_timeout(200) try: page.click("#link-checkout", timeout=3000) except: page.goto(f"{BASE}/checkout.html", timeout=15000) page.wait_for_timeout(800) elif scenario == "directbot": # Bot: directly load checkout, no prior pages page.goto(f"{BASE}/checkout.html", timeout=15000) page.wait_for_timeout(500) # Now load the demo page to read the tree state page.goto(DEMO, timeout=15000) page.wait_for_function("window.__dbReady === true", timeout=10000) page.wait_for_timeout(500) # let the setInterval refresh result = page.evaluate("window.__dbResult") return result if __name__ == "__main__": results = {} for condition in ["vanilla", "cdp"]: for scenario in ["directbot", "realuser"]: key = f"{condition}_{scenario}" results[key] = run(condition, scenario, use_cdp=(condition == "cdp")) # Compare print("\n" + "=" * 60) print("COMPARISON MATRIX") print("=" * 60) print(f"{'Condition':<25} {'Pages':<8} {'Children':<10} {'Root':<15} {'Verdict'}") print("-" * 60) for key, r in results.items(): print(f"{key:<25} {r['pageCount']:<8} {r['childCount']:<10} {r['rootPath'] or '(none)':<15} {r['verdict']}") # Verdict logic print() print("VERIFICATION:") # Both real-user scenarios should show pageCount >= 4 ru_v = results["vanilla_realuser"] ru_c = results["cdp_realuser"] db_v = results["vanilla_directbot"] db_c = results["cdp_directbot"] print(f" Real-user simulation:") print(f" vanilla: {ru_v['pageCount']} pages, {ru_v['childCount']} children, verdict={ru_v['verdict']}") print(f" cdp: {ru_c['pageCount']} pages, {ru_c['childCount']} children, verdict={ru_c['verdict']}") print(f" Direct-link bot:") print(f" vanilla: {db_v['pageCount']} pages, {db_v['childCount']} children, verdict={db_v['verdict']}") print(f" cdp: {db_c['pageCount']} pages, {db_c['childCount']} children, verdict={db_c['verdict']}") # The signal works if: # 1. Real-user scenarios show multi-page navigation (>=3 pages) # 2. Direct-link bot scenarios show only 1 page (only checkout) # 3. Both vanilla AND CDP produce consistent results (not CDP-divergent) if ru_v["pageCount"] >= 4 and ru_c["pageCount"] >= 4: print(f"\n ✅ Real-user simulation correctly shows multi-page journey") else: print(f"\n ❌ Real-user simulation FAILED — expected 4+ pages, got vanilla={ru_v['pageCount']}, cdp={ru_c['pageCount']}") if db_v["pageCount"] == 1 and db_c["pageCount"] == 1: print(f" ✅ Direct-link bot correctly shows single page") else: print(f" ❌ Direct-link bot FAILED — expected 1 page, got vanilla={db_v['pageCount']}, cdp={db_c['pageCount']}") if ru_v["verdict"] == ru_c["verdict"] and db_v["verdict"] == db_c["verdict"]: print(f" ✅ Verdicts identical between vanilla and CDP (not browser-mode-specific)") else: print(f" ❌ Verdicts DIFFER between vanilla and CDP") if (ru_v["pageCount"] >= 4 and ru_c["pageCount"] >= 4 and db_v["pageCount"] == 1 and db_c["pageCount"] == 1): print(f"\n >>> SIGNAL VALIDATED <<<") sys.exit(0) else: print(f"\n >>> SIGNAL FAILED VALIDATION <<<") sys.exit(1)