docs(site): drop the count-based claims from the homepage

[os-sales]

Fixes #17351

Applies the maintainer's ruling to the homepage, in the maintainer's own words:

建议首页也是非必要不强调数字，省的一直要改。

(given 2026-09-10, after the same ruling was applied to README.md: 「数量 相关的统计都去掉」)

What changed — apps/docs/app/[lang]/page.tsx, and nothing else

The five sites the card names, in order:

1. HOME_DESCRIPTION — now "…into typed metadata that fits in a single context window." It feeds metadata.description, openGraph.description, twitter.description and the JSON-LD SoftwareSourceCode.description via homeGraph(), so all four move together, which is what the file's own comments require.
2. The hero paragraph — the same first sentence, verbatim, followed by the unchanged "Agents read it whole, reason it whole, refactor it whole." The description and the hero still say the same thing, in the same relationship they had before: the description is the hero's opening sentence.
3. The two-column prose section — the token-budget paragraph ("under 100k tokens; the UI adds just 50k more") is deleted. The ontology paragraph becomes the section's single full-width <p>; the <section> wrapper and its padding are untouched, so vertical rhythm is unchanged.
4. The count strip — 1,600+ typed schemas | 5 database drivers | 7 framework adapters | MCP built-in is deleted outright, not replaced with a number-free strip. The "Browse the full vocabulary in the docs" link that follows is the section's exit, and it already carries its own mt-8.
5. GATES[2].copy — "not a pile of generated glue".

One comment amended, because the edit falsified it: the section header above (3) read {/* ── the claim the video makes, in prose ── */} and that claim was the paragraph now deleted. It now reads {/* ── what that metadata is, in prose ── */}. The long comment above metadata and the two comments in homeGraph() describe how the description is shared — still true after the edit, so they are untouched.

The "7 framework adapters" cell was also false

packages/adapters/ on this tree holds exactly one adapter:

$ ls packages/adapters/
hono

Deleting the strip removes the false claim; it is not corrected in place, per the card.

Evidence

The card's constraint grep, before and after, over the same file:

$ grep -nE '[0-9]+k|[0-9],[0-9]{3}\+|\b[0-9]+\+' 'apps/docs/app/[lang]/page.tsx'
# BEFORE (origin/main, 706ad0f) — 5 matches, all of them count claims:
36:  '…into typed metadata: a complete CRM in under 150k tokens, one context window.';
222:            typed metadata: a complete CRM in under 150k tokens, one context window. Agents
298:              The business logic alone — every object, workflow and permission — is under 100k
299:              tokens; the UI adds just 50k more.
447:            <span>1,600+ typed schemas</span>

# AFTER — no matches (exit 1).

A wider sweep for the same claims by literal spelling — 150k|100k|50k|1,600|N database|N framework|N typed — also returns nothing.

The rendered page, not just the source. After pnpm docs:build, over apps/docs/.next/server/app:

$ grep -rl '150k tokens\|1,600+ typed schemas\|7 framework adapters\|fifty thousand lines\|50k more' apps/docs/.next/server/app
# exit 1, 0 files

$ grep -o 'ObjectStack turns the whole app[^"<]*' apps/docs/.next/server/app/en.html | sort -u
ObjectStack turns the whole app — … — into typed metadata that fits in a single context window.
ObjectStack turns the whole app — … — into typed metadata that fits in a single context window. Agents read it whole, reason it whole, refactor it whole.

The short form is the <meta> / OpenGraph / Twitter / JSON-LD description; the long form is the hero. Both moved, together.

Gates

Derived in the worktree with node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no hand-written path list) at commit dd66285, then reconciled with --ran:

Run reconciliation — 27 derived, 27 run, 0 NOT-MEASURED, 0 UNRUN.
  EXIT CODES — all 27 accounted famil(ies) carry one, so the NOT-MEASURED count above is DERIVED from them.
✓ dispatch-gates --ran: 27 derived famil(ies) accounted for — 27 run, 0 NOT-MEASURED

All 27 exited 0. Plus, beyond the derivation:

Command	Verdict
pnpm --filter @objectstack/docs typecheck	exit 0 (fumadocs-mdx && next typegen && tsc --noEmit)
pnpm lint (whole repo, not narrowed)	exit 0 — 6488 files linted, 0 errors, 0 warnings, 1m43s
pnpm docs:build under bash scripts/pm/os-verify-lock.sh -c	VERDICT command-exit 0 · held the lock 173s (2m53s) · waited 0s

pnpm lint was run over the whole population rather than the changed file, so no narrowing needs proving. For the record, the population figure comes from eslint's own resolution (--format json, 6488 entries), and this repo's eslint.config.mjs never enables type-aware linting for any file (its own comment above QUERY_OPTIONS_TEST_GLOBS states and measures this) — so a one-file diff could not have moved an untouched file's verdict in any case.

Not measured here, by design: the CI job [ci.yml · Build Docs] runs 4 steps in CI's own shell that no local family names; the 49 artifact-roster families, the 11 wide-population families and the always-runs tail are outside the derived union. CI owns those.

Changeset

None, and skip-changeset is on the PR. apps/docs/package.json is "private": true with no files[], so this diff publishes nothing from any released package — confirmed against the repo's own gates rather than assumed: pnpm check:published-files and the 26 other derived gates are green with no changeset present, and pr-automation.yml reads the skip-changeset label live.

Acceptance notes

Observed while working on this file; not changed here, and not filed:

• The GitHub repository description field still carries "a complete CRM in under 150k tokens, one context window." It is a repository setting, not a file in this repo, so no PR can reach it — it needs a hand edit in repo settings. The card already flags this for the maintainer. Carrier: the maintainer, not a PR.
• README.md on this tree still carries all four count claims (lines 10, 14, 144, 199 — 150k tokens, 100k tokens; the UI adds just 50k more, fifty thousand lines of glue, 1,600+ of them). The card describes docs(readme): rewrite root README for developer onboarding #17347 as having already landed that rewrite; on origin/main at 706ad0f it has not. That is docs(readme): rewrite root README for developer onboarding #17347's surface, not this card's, and the wording used for site (5) here is the wording that card specifies, so the two land consistent whichever order they merge in. Carrier: docs(readme): rewrite root README for developer onboarding #17347.
• Out of scope by the card, unchanged: the nav links, the hero CTA row, and the npx create-objectstack my-app spelling in STEPS[0].cmd (which differs from the npm create objectstack@latest form used elsewhere). Carrier: none — noted only.
• The card's predicted post-state for the constraint grep does not hold, in the safe direction. It expected the grep to still show "the 01/02/03 section numerals and the 90 Seconds video title". That regex cannot match either — 01 and 90 are followed by neither k nor + — so the correct post-state is zero matches, which is what we get. No match survived that the card did not name.

🤖 Generated with Claude Code

https://claude.ai/code/session_016gFSzdPX4324QMRKCjccTc