Repository navigation
site: home page redesign (one CTA, use-case blocks, aligned layout) - #3034
Merged
Merged
Conversation
Why: the owner reviewed xerj.org as a first-time investor and the home page failed it. Measured on main 34a2f24 at 1440 px: four different left edges (x = 64, 280, 360, 544), a centred hero over left-aligned sections, 3,166 words over 12 screens, three copy buttons, four use cases squeezed into columns with blog thumbnails, a "Looking for a CEO" ad, and a footer whose bottom bar repeated five of its own links next to twelve legacy pages (pricing with passed SOC 2 dates, industries, a mock dashboard). What changed: Home page (landing/index.html + new landing/home.css) - One container (1200 px of content) and one 12-column grid. Every heading, paragraph and figure in <main> starts at x = 120 at 1440 and x = 16 at 390; nothing is centred (headless probe, 61 of 62 text elements on the edge, the 62nd is the link beside the Email button). - One copy button on the whole page: the hero prompt, which covers install with the model, index, search, ask XERJ Query-1 and reference coding, with a verify line and a no-shell line (llms.txt study rules: no obligation words, no PR ask, no shell assumed). The install line is plain text. - Why now + system diagram, one proof table (the six corrected comparisons from /blog/xerj-query-ai-model, each with how it was obtained), seven full-width use-case blocks (text left, real capture right), what is open, dated GitHub counts, one ask. 1,393 words in <main> (was 3,166). - Removed: centred hero and its caps eyebrow, the install COPY button, the six metric cards, the format-4 caveat paragraph, the 4-column use-case grid and its second prompt, the reference-coding and autoindex long forms (July capture, 94% aside, 80/81 tiles), the HTTP API walkthrough and recipe cards, the feature table, the blog thumbnail strip, the GET XERJ band, the CEO ad, the hero-video script, charts.js and chart-data.js. Use-case images (landing/assets/home/, 32 webp, wide/narrow x day/night) - Rendered from real runs of the rc.101 release (sha256 checked) with the model bundle on a private port: facts ask/UNKNOWN/CONFLICT/correction (I1), xerj def over sled (I2), xerj autoindex over 13 public NIST/CMS PDFs, 22.9 MB, wall 4.6 s, then 0.94 ms p50 over 200 distinct queries (I3), memory + graph ego + the 24-tool MCP list (I5), the official Python elasticsearch 8.19.3 client (I7). I4 (xerj connect) and the 787 MB figure come from the 2026-11-05 dev-build run and are tagged "Ships in rc.102". - Found while capturing I7: elasticsearch-py's helpers.bulk sends PUT /_bulk and rc.101 (and main) answers 405; the router registers POST only (engine/crates/xerj-api/src/router.rs:311, :419). Block 07 says so. Nav and footer, site-wide (scripts/seo/set_nav.py, fix_links.py) - Nav: Query-1, Use cases, Benchmarks, Docs, Blog, Hub + the GitHub badge, one row; at 720 px and below a <details> Menu with the same links, no JS. The badge count is re-read from the API (18,495, dated in its title). - Footer: generated brand line, 4 groups / 27 links, each once, then one bar with the copyright and the DAY/NIGHT toggle. The hand-written <footer class="footer"> bar is removed from every page. /answers/ and /compare/ stay because build_articles.py requires every article hub in the footer; /product, /use-cases and /for-agents stay because seo_lint rule 54 needs the home page to link them. - Pages that left nav and footer: pricing, security, solutions, resources, industries (5 pages), demo, aise-demo are 301'd (_redirects) and noindex; playground gate, architecture, agent-search and brand are noindex only. docs/index.html now links /docs/agents/ (it was only reachable by footer). New generator: scripts/seo/home_counts.py writes the dated contributor, release and download counts (no stars) and the three newest posts. Checks: build_articles/fix_heads/fix_links/set_nav/stars_badge/home_counts --check ok; seo_lint 0 gate violations; landing-constants-guard passed; test_verify_agent_discovery, test_llms_field_report (updated for the new single prompt) and the other seo tests OK; node landing tests 28/28.
This was referenced Nov 7, 2026
…hero proof strip) Why: the critique of the 8ef5418 preview (landing-redesign-2026-11-07 CRITIQUE.md) found one layout bug, two proof rows that broke the honest-claims rules, two claims missing their fair context, and an investor read that saw only developer documentation above the fold. Must-fix - M1 #open grid: the heading spanned 8 of 12 columns, so "The engine" card landed beside it at x=936 (1440) and the next two wrapped with an empty slot. The heading is now full width; the three cards share one row. - M2 proof table: dropped "17 min vs 62 to 97 h" (measured vs estimated, and a keyword index vs embeddings) and "2 ms vs 30 ms" (measured vs arithmetic). The blog keeps both with their caveats. - M3 wrong answers: the usual-way cell now says RAG searched one merged store, and that Opus 5.5 given the same records got 5 wrong, which our sample cannot tell apart from 1 (CORRECTIONS.md item 2). - M4 cost: under the $42 the page now says writing the facts is not in the number and that today you or your agent write them; the table cell says "not counting writing the facts". - M6 block 07 links the new issue #3036 (PUT /_bulk 405); the fix is PR #3037. The sentence stays until a release carries it. - M5 (llms.txt auto-send + PR-first field report) and M7 (rc.102 timing) are owner decisions and are not in this commit. M5 is pinned by test_llms_field_report and the installers, so it is a policy change, not a copy edit. Should-fix / nice-to-have - S8a hero: the empty right third at 1280+ now holds a three-number proof strip (220 ms vs 1.5 to 2.3 s API time; $42 vs $13,400+; 1 wrong and 35 declined vs 9 to 15, with the 5-wrong same-records line), each scoped to the same 160 questions. The proof section's duplicate big number is gone. Below 960 px the strip sits under the prompt. - S8b each use-case heading is now the outcome ("Your coding agent stops guessing APIs.", "Keep your Elasticsearch clients. Change the engine."). - S7 the one prompt gains step 5: register XERJ as an MCP server where the client takes one (llms.txt has the line per client) and save one memory note. No obligation words, no PR, under 300 chars per line. - S3 "Fund GPU hours" left the ask; Donate stays in the footer. - S4 home_counts.py holds /blog/100-contributors off the home list (HELD) until the star jump is explained. - S5 the releases tile reads "1.0 release candidates since 8 July 2026". - N1 the Copy button moved into the prompt's header bar: y=471 at 390 px (was 924), visible on load. - N2 "Agents now do a growing share of the searching" (was "most"). - N3 home meta description now matches the H1 (pagedata.py, 150 chars). - N5 block 06: 787 MB is the search node; the Query-1 node used 1.05 GB (docs/examples/usecase-cohere/RESULTS.md:9). - N6 the #traction label is now an h2 styled as a label. - Copy trimmed elsewhere to hold the word count: 1,432 words in <main> (was 1,393 before the strip and step 5). Checks: home_counts/fix_heads/fix_links/set_nav/stars_badge/build_articles/ mk_og_card/gen_sitemap/build_skills --check ok; seo_lint 0 gate violations; landing-constants-guard passed; test_verify_agent_discovery, test_llms_field_report (prompt updated) and the other seo tests OK; functions tests 12/12 and 15/15. Headless probe at 390/768/1280/1440/1920 day and 1440/390 night: one left edge per width (16/28/64/120/360), no horizontal scroll, no centred text, 1 copy button, 0 em/en dashes, 0 console errors or failed requests, 8 images shown per theme.
- Category stays: The search engine built for AI agents. - Removes the sentence about whether the XERJ Query-1 model is open; the page says nothing about model openness (engine Apache-2.0 lines kept). - No raising message; Donate stays in the footer only. - Refreshes the build-time GitHub star fallback from the API (18,497 on 2026-11-07) across 226 navs; landing/theme.js keeps refreshing it live from api.github.com (CSP already allows it). Checked with Playwright at 1440 and 390: badge shows the live count, one copy prompt, every section edge identical (120/1200 and 16/358), no horizontal scroll. SEO --check all ok, seo_lint 0 gate violations, landing guard passed.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
The home page failed a first-time-investor read: four different left edges at 1440 px (x = 64, 280, 360, 544), a centred hero over left-aligned sections, 3,166 words over 12 screens, three copy buttons, use cases squeezed into four columns, a "Looking for a CEO" ad, and a footer whose bottom bar repeated five of its own links beside twelve legacy pages.
What changed
Home page (
landing/index.html, newlanding/home.css)<main>starts at x = 120 at 1440, x = 64 at 1024, x = 16 at 390. Nothing is centred./blog/xerj-query-ai-model, each with how it was obtained), seven full-width use-case blocks (text left, real capture right), what is open, dated GitHub counts plus the three newest posts, one ask.<main>(was 3,166).Use-case images (
landing/assets/home/, 32 webp: wide and narrow, day and night, swapped by the theme toggle, each with alt text)xerj defover sled (I2);xerj autoindexover 13 public NIST/CMS PDFs, 22.9 MB, wall 4.6 s, then 0.94 ms p50 over 200 distinct queries (I3); memory, graph ego and the 24-tool MCP list (I5); the official Pythonelasticsearch8.19.3 client (I7). I4 (xerj connect) and the 787 MB figure come from the 2026-11-05 dev-build run and are tagged "Ships in rc.102".elasticsearch.helpers.bulksendsPUT /_bulk, and rc.101 (and main) answer 405. The router registers POST only (engine/crates/xerj-api/src/router.rs:311,:419). Block 07 says this.Nav and footer, site-wide (
scripts/seo/set_nav.py,scripts/seo/fix_links.py)<details>Menu (no JS). The badge count is re-read from the API (18,495, dated in its title).<footer class="footer">bar is gone from every page.scripts/seo/home_counts.pywrites the dated contributor, release and download counts (stars are left out) and the three newest posts.Removed
From the home page:
.hero-prooffootnotecorpus addaside, terminal, second link row)charts.jsandchart-data.jsFrom the nav: Cohere, Product, Answers, For agents, Dashboards, Donate, the X and Discord icons, the release button and the DAY/NIGHT row (which moved to the footer).
From the footer: the self-duplicates (Product under PRODUCT, Solutions under SOLUTIONS, Documentation under DOCS), the whole bottom bar, Architecture, Pricing, Security, Dashboards, Real-data demo, Product tour, Solutions, Industries and its 4 pages, Sales kit, Recipes, Agentic search demo, Semantic analytics, AI code-security audit, and Docs for AI.
Pages that left the site map:
/pricing/and/security/go to/#open;/solutions/,/resources/and/industries/*go to/#use-cases;/demo/and/aise-demogo to/benchmarks/elasticsearch./playgroundgate (a mock),/architecture/,/agent-search/and/brand.landing/_redirectsand the entries inurlmap.NOINDEX.Kept, with the reason:
/answers/and/compare/:build_articles.pyrequires every article hub in the footer./product,/use-casesand/for-agents: seo_lint rule 54 needs the home page to link them.docs/index.htmlnow links/docs/agents/.Screenshots
Before (live main
34a2f2441):/home/claude/ai/landing-redesign-2026-11-07/build/before-1440.png/home/claude/ai/landing-redesign-2026-11-07/build/before-390.pngAfter (preview https://redesign-preview.xerj.pages.dev):
/home/claude/ai/landing-redesign-2026-11-07/build/preview-1440.png/home/claude/ai/landing-redesign-2026-11-07/build/preview-390.png/home/claude/ai/landing-redesign-2026-11-07/build/preview-1024.png/home/claude/ai/landing-redesign-2026-11-07/build/preview-1440-night.png/home/claude/ai/landing-redesign-2026-11-07/build/preview-390-night.pngRaw run transcripts behind every image (command, date, host, binary hash):
/home/claude/ai/landing-redesign-2026-11-07/runs/I{1,2,3,4,5,7}_*.txtChecks
build_articles,fix_heads,fix_links,set_nav,stars_badge,home_counts,mk_og_card,gen_sitemapandbuild_skills--check: oklanding-constants-guard.sh: passedtest_verify_agent_discovery,test_llms_field_report(updated for the single new prompt) and the other SEO tests: OKOpen before merge
landing/llms.txtline 28 still asks agents to send a field report without asking, and to open a PR first. The hero prompt sends every agent there. That needs its own PR, because only one open PR may edit llms.txt.git@xerj.org, the only address the old page had. The owner should confirm it.🤖 Generated with Claude Code