8000 site: rewrite /product for v1.0.0-rc.98 on current main by xerj-org · Pull Request #1926 · xerj-org/xerj · GitHub
[go: up one dir, main page]
Skip to content

site: rewrite /product for v1.0.0-rc.98 on current main - #1926

Merged
xerj-org merged 3 commits into
mainfrom
site/product-page
Oct 23, 2026
Merged

site: rewrite /product for v1.0.0-rc.98 on current main#1926
xerj-org merged 3 commits into
mainfrom
site/product-page

Conversation

@xerj-org
@xerj-org xerj-org commented Oct 20, 2026 •
edited by cursor Bot
Loading
Copy link
Copy Markdown
Owner

AI disclosure: this PR was written by an AI agent (Grok Bot) working for the repo owner. A human must review it before merge. Every number was copied from a run, a file or a fetch made on 2026-10-19 (PT); nothing was estimated.

Stacked on #1920 (site/light-theme); the base is that branch so the diff shows only this change.

What changed

landing/product.html is rewritten for what ships today (v1.0.0-rc.97). The old page described an April "v0.1" product.

New structure, answer-first:

  1. A 50-word definition: a local search engine for AI agents, one Apache-2.0 Rust binary, citations with file:line, served through the CLI, MCP and the Elasticsearch API.
  2. Who it is for, with a "not a fit" note: lexical default embedder, no regex search, idle cost 8% CPU / 8.3 GB from Meta: idle cost must be near-zero on dev laptops — the reference-coding resource budget (#871 #872 #873) #874.
  3. A quickstart you can run as written, and the real output of xerj "how does lz4 find a match".
  4. The command surface, taken from xerj --help on rc.97.
  5. Corpus Hub v2:
  6. Measured: my run on psf/requests (1,303 B of xerj search output against 26,278 B of rg), next to the published benchmark table with its losses.
  7. A comparison with ripgrep, Zoekt, Elasticsearch/OpenSearch, a vector database and GitHub code search.
  8. Field reports (xerj feedback --submit, and how to opt out) and an FAQ.

scripts/seo/pagedata.py has the new title and description. The xerj-blogposts pack JSON was already stale on site/light-theme. It is regenerated here, and the bytes match what #1918 commits.

Claims removed or corrected

Old claim Why it went
"One binary. Three stores" (logs, vectors, agent memory) Not how the product is positioned or used today
"V0.1" kicker and footer; "BENCHMARKS 2026-04-14" Stale
81/91 cells, 8 losses, 2 unsupported; bool 11.5×, query_string 6.9×, wildcard 6.8×, range 3.5×, kNN 3.4×, terms 1.15×, 1.20× smaller; mixed p99 "2–19 ms vs ~60–150 ms" Replaced by the current 88-cell board: 55 W / 26 T / 4 L / 3 n/a, 176.2 vs 283.0 MB, read p99 under writes 10.3–13.6 vs 3.5–6.8 ms (a loss, now stated)
Seeded mock dashboards shown as product (model latency for OPUS 4.6 / GPT-5, an 830,000-query UMAP, service graph, attention overlay, token budget, spend heatmap) Demo seed data, not product output
"Every query is a point in 1536-dimensional space" The default embedder is lexical feature hashing; neural MiniLM is 384-d
"Drop a work email for a private binary" / "UNLOCK WITH ONE EMAIL" The binary is public Apache-2.0
"Rust static binary with no heap" False
"Hybrid planner … no RRF orchestrator" Hybrid search is RRF
Columnar/SQ8 and single-process NGINX/OTLP parser internals Not verified against main; the page now links to /architecture/
Inline lead-form / video script Removed with the email gate

Verified

  • Installed rc.97 with curl -fsSL https://xerj.org/get | sh into a scratch HOME, and ran every command on the page.
  • MCP tools/list returns 17 tools. Note: the README still says 13.
  • hub.xerj.org, the signed catalogue and /api/corpus-feedback were fetched on 2026-10-19.
  • Benchmark figures were copied from landing/benchmarks/index.html, the case study and the README, with links.
  • All site checks pass: gen_sitemap, fix_heads and fix_links --check, seo_lint (0 gate violations), build_articles --check, mk_og_card --check, landing-constants-guard, gen_posts --check, audit (LOW only), factcheck (0 ERROR), factcheck self-test, test_article_schema, test_urlmap, test_llms_field_report.
  • Took desktop and mobile screenshots.

Assumed

Not run

  • ES-YAML conformance suite: skipped, because this change is landing-only (no engine code touched). The 1,366/1,369 figure is the existing constant.
  • No Rust build or tests (no engine change).
  • No deploy preview; checked with a local static server.

What this changes, and why

PR #1926 rewrote /product for v1.0.0-rc.97 while its base was site/light-theme. That branch was already squash-merged as #1920. Merging the stack would have brought the paused site redesign in with the page. This rebuilds site/product-page on current main (5ad85724e) and retargets the base to main.

The diff is landing/product.html, the product.html title and description in scripts/seo/pagedata.py, and one sitemap row: /product <lastmod> moves from 2026-10-22 to 2026-10-23, the date of the content commit. The page uses main's nav, footer, theme.js and style.css?v=sc0706. Table rules are page-local. style.css, the blog-post JSON, and .github/workflows are unchanged.

The live page still describes v0.1. This replaces it with the product the rc.98 release binary actually serves.

Evidence

Checked 2026-10-23 against the release binary installed with XERJ_VERSION=v1.0.0-rc.98 from landing/get (xerj --version printed xerj v1.0.0-rc.98):

  • xerj --help, xerj hub --help, xerj mcp --help (proxies 20 tools), and an MCP tools/list that returned those 20 names. The generated schema on main after the rc.98 tag lists 21, including xerj_def. That tool is absent from the rc.98 binary, and the page says so.
  • xerj "how does lz4 find a match" (same command as xerj ask), scratch home, feedback disabled: started a node on port 9300, pulled lz4 0774d055 4623814 → 4 indices, 1702 definitions, first hit lz4/lib/lz4.c:939. xerj hub stop printed stopped the hub node (pid 3727).
  • Fetches the same day: catalogue.json generated_at 2026-10-22T07:11:17Z, 497 corpora; hub.xerj.org/llms.txt 514 /corpus/ links (the 17 slugs missing from the catalogue are named on the page); search/index.json 4,080 contributor rows (3,663 kind 0, 417 kind 1); GET /api/corpus-feedback prior mean 3, weight 5, three corpora scored 3.17.
  • The psf/requests block is copied from benchmarks/agent-intents-2026-10-19/ (rc.97, 2026-10-19, commit 611c6162). It was not re-run.
  • The benchmark table is copied from /benchmarks, the reference-coding case study, and demo/playbooks/SCORECARD.md, including the four read-p99 losses and the idle-daemon loss. Those cells were not re-measured.

Local gates, all exit 0: python3 scripts/seo/seo_lint.py (198 pages, 0 gate violations), fix_heads.py --check, fix_links.py --check, gen_sitemap.py --check, bash .github/scripts/landing-constants-guard.sh (footer V1.0.0-RC.98, conformance 1,366/1,369).

The first push failed Technical SEO because gen_sitemap.py --check wanted /product lastmod 2026-10-23. That is the content date of the rewrite, not a sitewide lastmod drift. Commit 1a744cb54 writes that one row.

Checks

  • ES-YAML conformance: not applicable, landing-only
  • User-visible page updated; pagedata title and description regenerated into the head by fix_heads.py
  • Operational-document audit: N/A. This does not change a runnable recipe or an operational runbook. The page quotes commands; it does not add one under docs/recipes/
  • cargo fmt, scoped release build, cargo test, new ES-YAML case: not run, no engine change

Provenance

  • Written by: AI agent (Grok 4.7 in Cursor), for the repo owner. A human should review it before merge.
  • Verified: the rc.98 help text, tools/list, the lz4 ask, the hub fetches, the file copies named above, and the local SEO gates.
  • Assumed: the published benchmark cells and the 2026-10-19 requests capture still match the files they were copied from. They were not executed again on rc.98. A one-command reproduction of the ES board was not run.
Open in Web Open in Cursor 

@cla-bot cla-bot Bot added the cla-signed label Oct 20, 2026
cursor Bot pushed a commit that referenced this pull request Oct 23, 2026
@siliconpravda @xerj-org
docs(site): rewrite /product against the rc.98 binary
a149b88 
PR #1926 was stacked on site/light-theme after that redesign had
already landed on main as #1920. Rebuilt the page on current main so
the diff is the content rewrite only. The live day/night chrome stays:
nav, footer, style.css?v=sc0706 and theme.js are main's. Table rules
are page-local. style.css, the sitemap and the blog-post JSON are
untouched.

The old page still describes v0.1. The stacked rewrite described
rc.97 (17 MCP tools, a 14-corpus catalogue). Neither matches the
release the installer serves.

Verified 2026-10-23, binary xerj v1.0.0-rc.98 from landing/get:
xerj --help, xerj hub --help, xerj mcp --help (20 tools), tools/list
(those 20 names), and one xerj ask of lz4 (pin 0774d055, 4623814
bytes, 1702 definitions, first hit lz4/lib/lz4.c:939). Hub fetches
the same day: catalogue.json 497 corpora, hub llms.txt 514 /corpus/
links, search/index.json 4080 contributor rows, corpus-feedback prior
mean 3 weight 5 with 3 scored corpora. Other numbers are copied from
the files named on the page. The psf/requests capture was not re-run.
SEO lint, fix_heads/fix_links/gen_sitemap --check and the landing
constants guard passed locally.

Assumed: the published benchmark cells were not re-measured here.

Co-authored-by: xerj-org <xerj-org@users.noreply.github.com>
@cursor
cursor Bot force-pushed the site/product-page branch from 8ef46c1 to a149b88 Compare October 23, 2026 06:48
@cursor cursor Bot changed the title site: rewrite /product against v1.0.0-rc.97 (CLI, xerj ask, MCP, Corpus Hub v2, current benchmarks) site: rewrite /product for v1.0.0-rc.98 on current main Oct 23, 2026
@cursor
cursor Bot changed the base branch from site/light-theme to main October 23, 2026 06:49
@xerj-org xerj-org mentioned this pull request Oct 23, 2026
Merged
@cla-bot
cla-bot Bot commented Oct 23, 2026
Copy link
Copy Markdown

Thanks for this contribution — we'd like to merge it.

Before we can, we need a signed Contributor License Agreement from: @siliconpravda

It is a one-time thing per contributor, not per pull request. Read CLA.md, then open a small pull request adding your GitHub username to .contributors — that pull request is your signature, since it is made from your own account.

Once it is merged, comment @cla-bot check here and this check will go green. If you believe you are already covered, comment @cla-bot check and it will re-verify.

@cla-bot cla-bot Bot removed the cla-signed label Oct 23, 2026
xerj-org and others added 3 commits October 23, 2026 08:44
@xerj-org
docs(site): rewrite /product against the rc.98 binary
7fd45a8 
PR #1926 was stacked on site/light-theme after that redesign had
already landed on main as #1920. Rebuilt the page on current main so
the diff is the content rewrite only. The live day/night chrome stays:
nav, footer, style.css?v=sc0706 and theme.js are main's. Table rules
are page-local. style.css, the sitemap and the blog-post JSON are
untouched.

The old page still describes v0.1. The stacked rewrite described
rc.97 (17 MCP tools, a 14-corpus catalogue). Neither matches the
release the installer serves.

Verified 2026-10-23, binary xerj v1.0.0-rc.98 from landing/get:
xerj --help, xerj hub --help, xerj mcp --help (20 tools), tools/list
(those 20 names), and one xerj ask of lz4 (pin 0774d055, 4623814
bytes, 1702 definitions, first hit lz4/lib/lz4.c:939). Hub fetches
the same day: catalogue.json 497 corpora, hub llms.txt 514 /corpus/
links, search/index.json 4080 contributor rows, corpus-feedback prior
mean 3 weight 5 with 3 scored corpora. Other numbers are copied from
the files named on the page. The psf/requests capture was not re-run.
SEO lint, fix_heads/fix_links/gen_sitemap --check and the landing
constants guard passed locally.

Assumed: the published benchmark cells were not re-measured here.

Co-authored-by: xerj-org <xerj-org@users.noreply.github.com>
@xerj-org
docs(site): set /product sitemap lastmod to the rewrite date
497c02b 
gen_sitemap derives lastmod from the last commit that changed the
page outside shared chrome. The rc.98 rewrite moved that date from
2026-10-22 to 2026-10-23, and the SEO gate's --check failed on that
one URL. No other sitemap row changed.

Co-authored-by: xerj-org <xerj-org@users.noreply.github.com>
@xerj-org
docs(site): count 52 field reports now on main
72dba73 
#2252 and #2259 landed after the product rewrite was drafted.
user-feedback/16-agent-field-reports holds 52 markdown files
besides the README on origin/main (214f243), dated
2026-08-18 through 2026-10-23.

Co-authored-by: xerj-org <xerj-org@users.noreply.github.com>
@cursor
cursor Bot force-pushed the site/product-page branch from 1a744cb to 72dba73 Compare October 23, 2026 08:44
@cla-bot cla-bot Bot added the cla-signed label Oct 23, 2026
@xerj-org
xerj-org merged commit ffc4e97 into main Oct 23, 2026
23 of 24 checks passed
@xerj-org
xerj-org commented Oct 23, 2026
Copy link
Copy Markdown
Owner Author

Xerj Merger made the fix: rewrites /product for v1.0.0-rc.98 on current main (landing/product.html, sitemap lastmod, pagedata entry); no redesign files. Approved by the maintainer; Build + Test passed on head 72dba73.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

cla-signed

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@xerj-org
0