docs: adopt the new tagline across README and the docs homepage hero

[os-zhuang]

Fixes#9241

Adopts the maintainer's definitive tagline (2026-08-17) on the root README first screen and the docs-site homepage hero, and adds the business-logic split line. Both blocks are used verbatim — not one word rewritten or polished. Verified mechanically rather than by eye: the flowed text of each surface is normalized for whitespace and asserted to contain the definitive strings as substrings (README hero, docs hero, docs page metadata: all three green).

Number discipline held everywhere the copy landed: 150k is always the whole app including UI; 100k is always the business logic alone. The two numbers never swap host nouns.

Changed — 2 files, 6 spots

#	Location	Was	Now
1	README.md:7	## A complex CRM system in 170k tokens.	## Apps small enough for AI to hold whole.
2	README.md:9-13	ObjectStack compresses an entire app … refactor whole ([a complete CRM is under 170k tokens](#why-the-mistakes-dont-ship))	tagline body verbatim, then the split line as its own paragraph
3	apps/docs/app/[lang]/page.tsx:21	metadata title ObjectStack — A complete business system in 16k tokens	ObjectStack — Apps small enough for AI to hold whole.
4	apps/docs/app/[lang]/page.tsx:24	metadata description opening ObjectStack compresses an entire business app … a complete CRM is under 2,000 lines	tagline + split line verbatim
5	apps/docs/app/[lang]/page.tsx h1	A complete business system / in 16k tokens.	Apps small enough / for AI to hold whole. (gradient span kept on the second half)
6	apps/docs/app/[lang]/page.tsx lead paragraph	old compresses an entire app paragraph	tagline paragraph, split-line paragraph, then the surviving ontology/typed tail

Two shape notes, both formatting only, no words touched:

• The tagline's first sentence becomes the headline (README H2, docs h1, docs metadata title) and the remaining two sentences the body — exactly the split the README already used for the retired copy. Concatenating headline + body reproduces the definitive paragraph verbatim.
• The old anchor link [a complete CRM is under 170k tokens](#why-the-mistakes-dont-ship) is dropped, not re-pointed: the definitive copy carries no link, and inserting one would be editing it. The heading it pointed at is untouched and still reachable from the README table of contents.

The ontology / TypeScript / runtime-derives tail of both paragraphs is retained — it is separate content, not the retired tagline, and the definitive copy does not replace it.

Swept and deliberately NOT changed

Grep sweep over README*, content/docs/**, apps/docs/app/**, content/blog/** for 170k, A complex business system, compresses an entire app, hold in context, one context window, entire/whole app, and every 100k / 150k / 2,000 lines token claim. Residual old copy after this PR: zero hits.

Location	Hit	Why it stays
README.md "The other half is size"	examples/app-crm is 31 files, 1,792 lines, roughly 16k tokens — "the whole business system, in about 8% of a 200k-token context window"	A measured count of the bundled example, reproducible by the find … wc -l command printed directly beneath it. Not the retired tagline and not a 150k/100k host-noun claim. See the flag below.
content/docs/concepts/metadata-driven.mdx:219-232	same 31 / 1,792 / ~16k table + the same 8% sentence	Same measured count, same reason.
content/docs/getting-started/how-ai-development-works.mdx:29-34	same measured count in prose	Same measured count, same reason.
content/docs/concepts/metadata-driven.mdx:11-14 (Callout)	"the whole app is a few hundred lines of typed, analyzable metadata … it fits in an agent's context window"	Carries no token number and none of the retired phrases; nothing to align.
content/blog/context-window-is-the-constraint.mdx:43-52	16k (CRM) · ~14.5k (Todo) · ~101k (showcase)	Outside this card's declared surface (content/blog is neither README* nor a content/docs home/overview page), and the ~101k is a measurement of the kitchen-sink showcase app including UI — it is not the 100k business-logic host noun. Listed for sweep completeness.
content/docs/references/**	contextWindow, compression field rows	Generated schema reference; unrelated to tagline copy.
content/docs/capabilities/approvals.mdx:22, content/docs/protocol/objectui/index.mdx:511, skills/**, packages/**	100k / 100K as currency amounts, row counts, log-rotation sizes	Unrelated to token claims.
content/docs/index.mdx and the section overview pages (capabilities, build-without-code, getting-started, concepts)	—	Read in full; they carry no retired tagline, so there was nothing to align.
content/docs/releases/**	—	Repo rule: never touched in a code PR.
GitHub repo description field	—	Not in git. Per the issue, the maintainer pastes the first two sentences in the UI; out of this PR's reach.

One thing for the maintainer to decide (not fixed here). The tagline says a complete CRM is under 150k tokens, while three docs surfaces measure the bundled examples/app-crm at ~16k. Both claims read as "a complete CRM". This tension is pre-existing, not introduced here — the retired tagline said 170k while the same pages already said 16k, so the two figures have coexisted on main by prior arrangement (tagline = a full-featured CRM; the example = the minimal bundled one). Re-measuring or reconciling them would mean either inventing numbers or editing the definitive copy, so this PR does neither and surfaces it instead.

Verification

Union re-run after the final commit, at 89e946c50 (git rev-parse --short HEAD):

• node scripts/check-nul-bytes.mjs --self-test — 75 assertions green; node scripts/check-nul-bytes.mjs — OK, 6039 text files, no raw control bytes. Plus a targeted grep -naP control-byte self-scan of both changed files: clean.
• pnpm --filter @objectstack/docs types:check (fumadocs-mdx && next typegen && tsc --noEmit) — green, exit 0. The apostrophe in agent's inside JSX is the same construct the retired paragraph already used at that spot.
• node scripts/pm/dispatch-gates.mjs README.md 'apps/docs/app/[lang]/page.tsx' — re-derived against the actual changed paths: no check family names these paths and no workflow path filter schedules one for them; no model-tier path mandate. check:nul-bytes was run anyway as the every-edit family.

No changeset: both surfaces are unpublished (@objectstack/docs is private: true, the README ships in no package), so this PR releases nothing — skip-changeset applied.

---

Generated by Claude Code