1. Article paragraphs not in server HTML — Critical
What: The server sends every plain-text paragraph as an empty container; the text exists only inside the JSON data <script>. Paragraphs appear only after the browser runs JavaScript.
<div id="text-element-with-ad"><div></div></div>
Evidence
- PV: 62/62 live article pages with text blocks empty in server HTML (text, syndicated-print, syndicated, pv-exclusive); 42/42 in an earlier 64-URL sample. Share of body text in server HTML on 3 measured articles: 0%, 25% (big-fact blocks only), 0%.
- DH: 31/31 sampled articles empty; same markup.
- The Quint and Swarajya (also Quintype) serve paragraphs in server HTML → the problem is the custom TPML component, not the platform.
- Files:
evidence/empty_body_urls.csv,docs/QA_test_plan_article_body_and_CLS.md§1b.
Cause: story-templates/story-elements/index.js → CustomStoryElement. To insert ads between paragraphs, the text was split with the browser-only DOMParser and <p> tags counted inside a useEffect (never runs on the server).
Since when (git, PV tpml upstream/master; checked 16 Nov 2026): not since launch. At launch (d0db20fd, 28 Feb 2023) text elements went through Quintype's StoryElement, which renders paragraphs on the server. The client-side split arrived on 8 Oct 2024 (f9d0a0e3, "add new custom story element component to render ad after 4th para": getParagraphsArr regex + useEffect); d45f010f (15 Oct 2024) swapped the regex for DOMParser. So every PV story crawled since ~Oct 2024 is affected, including Jan 2026 (before print-to-web).
Traffic around the change (GA4, Google organic sessions incl. Discover; checks/traffic_funnel/ga4_monthly_organic_pv_2024-2026.json): article-landing sessions Aug 2024 4.38M → Nov 2024 2.79M (−36%) → Jan 2025 3.76M → Jul 2025 4.85M. The fall began in Sep 2024 (before the change) and Nov 2024 had a Google core update, so it can't be attributed to #1 alone; there is no lasting step down in totals because Discover (headline and image) dominates. The cost of #1 is mainly long-tail search that PV articles cannot rank for. Varta Setu, the automatic print-to-web system (syndicated-print template, from late Mar 2026), nearly doubled output (11–13K → ~20K stories/month) with no rise in web clicks to articles (Jan–Mar 546–619K/month vs Apr–Oct 474–586K; checks/traffic_funnel/monthly_articles_web_pv_2026.json).
Fix (built, PV): branch fix/ssr-story-paragraphs, commit bcd5d909 (local, not pushed). Server-safe paragraph splitter + per-story ad plan computed during render. Verified: 143 stories / 633 text blocks identical split and ad positions; local server HTML has 0 empty containers vs 1–8 per page in production; same ad/Taboola positions after JavaScript; hydration keeps all paragraph nodes.
DH: confirmed same code in quintype/tpml-dh main (0bc245a1, 23 Oct 2026): story-elements/index.js uses DOMParser in getParagraphsArr (line 134) and the module-level countPara / hasRenderedMidAd1/2 (lines 236–238). The PV fix can be ported.
Status (19 Nov 2026): ✅ DH fix live in production (per user); PV still to ship.
DH production verified (19 Nov 2026; checks/ssr_compare/checklist_www.deccanherald.com_2026-11-19_pre-deploy.md → checklist_www.deccanherald.com_2026-11-19.md, dh_live_sample_2026-11-19.json):
- Code:
quintype/tpml-dhorigin/main250ef126("ssr fix (#2110)", 18 Nov 2026), a squash of branchssr-fix(b7f3215b,2fc0e447). Paragraph splitting moved to newstory-elements/html-utils.js(string scanner, noDOMParser). No browser-only API is left in the render path;document/windowonly insideuseEffectortypeof windowguards. Body rendered during render, not inuseEffect. The module-level ad counters are gone (per-elementstartParaCountfrom auseMemo). The premiumSpinnerLoadergate is removed: server now renders the 115-word preview (#9). The commit does not touch "Also read", the heroWithLazyor related links. - Same 3 stories, Googlebot UA, before JavaScript: this morning (pre-deploy) 1/13, 1/15, 0/10 body paragraphs; after deploy 1/13, 15/15, 10/10. Story 1 was a Cloudflare cache HIT (
age383 s); the same URL with a cache-busting query had 0 of 13 containers empty. The 3 training reference stories (…4222260, …4222274) are 0 empty on live. - 12 recent stories (text, syndicated, premium, login): every MISS/EXPIRED response and most HITs have 0 empty containers; 3 of 12 (all Cloudflare HIT) still carried old HTML (13/13, 11/11, 25/25 empty). Cache-control is
s-maxage=900, stale-while-revalidate=1000, stale-if-error=14400, so they should age out; re-check on 20 Nov and ask for a purge if not. - Fresh HTML of 3 stories: one
FirstAdContainerand oneSecondAdContainereach, no duplicate GPT slot IDs, no/https:links. Premium story …4223169: 2 containers, 0 empty. - Not fixed by this release: the DH "Also read" card is still not in server HTML (0 cards on …4223587, whose linked story is not linked anywhere in the HTML);
also-read/index.js:50still builds/${linkedStorySlug}(#41, P1 item 3). Hero image still insideWithLazy(hero-image/index.js:72-74). - Every template in the latest 600 DH stories, 2 stories each, fresh fetch (
checks/ssr_compare/dh_templates_2026-11-19.json). Templates: text 268, syndicated 272, dh-exclusive 27, brandspot-client-story 13, listicle 9, photo 3, horoscope 3, news-briefs 2, video 2, visual-story 1. Result: 0 empty containers on all of them, and every free story had all its paragraphs in the HTML except the "Also read" text. Premium (subscription) stories have only their preview (5/13, 3/29), by design. Photo stories have no text elements. Horoscope keeps sign text instory.metadata, not in text elements. It was 0 of 12 signs on 12 Nov (#27) and is 12 of 12 now, because the commit imports the template statically instead of throughloadable(pages/story/index.js). Not tested: live-blog, podcast, cartoon, panchanga, embed, immersive (none in the latest 600 stories). - Covers old stories too (the page is rendered on each request): 3 stories from the 15 Mar 2024 daily sitemap (…2939006, …2939069, …2939001) had 0 of 4, 0 of 15 and 0 of 10 containers empty, both plain and cache-busted (Cloudflare MISS). Google's stored copy changes only when it recrawls each URL.
- Side effect: the e-paper promo after paragraph 1 is now in server HTML, and its
<h2>ANYTIME, ANYWHERE.</h2>is the first<h2>of every article (…-4223637,evidence/dh_ssr_2026-11-19/4223637_fresh.html). Fix:atoms/new-body-banner/index.js:44, render the heading as a<div>/<p>with the same class. - Code risks to pass to tech (not seen live): (1) the mid-ad checks
currentPara === 1(index.js:147) andcurrentPara === 4(:182) have no "already rendered" guard, so a non-<p>chunk (heading, list, blockquote) right after paragraph 1 or 4 would repeat the ad block with the same slot ID. (2)isWebView(:131) is false on the server and true on the client for?app=true, so the app webview gets a hydration mismatch around the second mid ad. (3)html-utils.jsmis-splits edge cases: an HTML comment or stray closing tag becomes visible text (!-- x --,/div>); a blockquote with several<p>counts as one paragraph. (4)console.log("[Taboola Debug] …")left in (:250,:268).
Status (16 Nov 2026, per user): 🚧 tech team is testing the fix on DH first; PV to follow shortly after.
Status (11 Nov 2026): 🚧 in progress with tech. Not yet released; re-test with the steps below once it ships.
Verify after release: view-source / JavaScript-disabled test (QA plan T1); Search Console "View crawled page → HTML".
DH fix verified on training (18 Nov 2026; checks/ssr_compare/checklist_deccanherald-training-web.quintype.io_2026-11-18.md vs checklist_www.deccanherald.com_2026-11-18.md): the same 3 stories fetched as Googlebot, before JavaScript. Training: 29 of 29 body paragraphs in the server HTML (16/16, 8/8, 5/5); live DH: 2 of 29. Nothing else regressed on training: H1, title, canonical, hero image, byline, NewsArticle JSON-LD and AMP link unchanged; HTML 1.30 MB (live 1.29 MB). Training is blocked in robots.txt (Disallow: /) and canonicals point to www, so it can't leak into Google. Also seen on training: og:type is now article (live: website, #34). After release: re-run checklist.py --host www.deccanherald.com, then URL Inspection "View crawled page" on 3 stories, then the body-indexing test (#2) after ~4 weeks.