docs(protocol): correct the two wrong defaultValue samples in schema.mdx

[os-help]

Fixes #7244

Docs-only. content/docs/protocol/objectql/schema.mdx is the only file touched — no schema, no engine, no driver, no gate.

What changed

# revenue (currency)
- defaultValue: { value: 0, currency: 'USD' }
+ defaultValue: 0   # currency stores a bare number — never { value, currency }

# due_datetime (datetime)
- defaultValue: "daysFromNow(7)"  # 7 days from now (CEL)
+ # A CEL default needs the explicit `{ dialect, source }` envelope. Unlike a
+ # formula field's `expression`, `defaultValue` has NO bare-string shorthand —
+ # a bare string is stored as that literal text.
+ defaultValue: { dialect: 'cel', source: 'daysFromNow(7)' }  # 7 days from now, at UTC midnight

The card's originals — confirmed wrong

valueSchemaFor(...).safeParse(...), the probe the card used:

ORIGINAL  datetime  < - "daysFromNow(7)": FAIL :: ["expected an ISO-8601 instant with explicit zone (e.g. 2026-03-15T14:30:00.000Z)"]
ORIGINAL  currency  < - { value: 0, currency: "USD" }: FAIL :: ["Invalid input: expected number, received object"]
NUMERIC_VALUE_TYPES.has("currency") = true

End-to-end insert of both current samples through the engine (memory driver) is refused, not silently stored:

CURRENT SAMPLES insert REJECTED: VALIDATION_FAILED
  bad_dt must be a valid datetime (ISO-8601); bad_cur must be a number

And the SQL DDL, better-sqlite3, real CREATE TABLE:

`cur_due_datetime` datetime default 'daysFromNow(7)',   -- the CEL source text as a physical column DEFAULT
`cur_revenue` float,                                     -- the object silently dropped: no default at all

Both halves of the card's trace reproduce exactly.

The replacements — measured, not assumed

This is what the card supplied but never probed.

REPLACE  currency  < - 0: PASS
CEL evaluate ok = true | value = Mon Aug 17 2026 00:00:00 GMT+0000 | ctor = Date
REPLACEMENTS insert OK. stored due_datetime = "2026-08-17T00:00:00.000Z"
REPLACEMENTS stored revenue = 0 | typeof number

DDL for the same two replacement declarations:

`new_due_datetime` datetime,             -- no column DEFAULT: the envelope is evaluated app-side
`new_revenue` float default '0',         -- a real literal default

So the envelope does not leak the source text into the DDL, which was the failure to rule out. daysFromNow is a genuinely registered stdlib function (packages/formula/src/stdlib.ts:117), not invented from a doc comment — an unregistered name is refused loudly:

CEL unregistered-fn probe ok = false | error = { kind: 'runtime',
  message: "found no matching overload for 'noSuchFn(int)'" }

daysFromNow(n) returns the calendar day at UTC midnight (ADR-0053 D1), so the inline comment now says so rather than the bare "7 days from now" — the old comment was true of neither the old sample nor, precisely, the new one. now() + duration("168h") was measured as the sub-day alternative (Mon Aug 17 2026 09:30:00 GMT+0000) and deliberately not used: daysFromNow is the helper the card named and the skills teach.

One thing the measurement did not confirm

The card says the bare string "stores the 14 characters into the datetime field". applyFieldDefaults does assign it, but the record validator refuses the insert immediately after, so on an ordinary field the user-visible symptom is a failed write, not silent bad data. The silent-storage reading holds only where validateRecord is skipped (readonly/system fields, isSystem writes) and for the physical column DEFAULT above, which is emitted regardless. The sample is wrong either way; the mechanism is worth stating accurately.

Population scan

The card found two defects and did not claim two was all. Every defaultValue on the page was probed:

line	field	value	verdict
112	select	draft	PASS
272	skeleton	null	n/a — null means "no default"
371	number	1	PASS
385	currency	{ value: 0, currency: 'USD' }	FAIL — fixed here
399	datetime	"daysFromNow(7)"	FAIL — fixed here
408	boolean	true	PASS
413	toggle	false	PASS
427	select	medium	PASS
998	boolean	true	PASS

Two is the complete population on this page. The look-alike at line 706 — expression: "record.first_name + ' ' + record.last_name" on a formula field — is correct and deliberately untouched: expression is typed ExpressionInputSchema, whose union accepts a bare string as shorthand for { dialect: 'cel', source }. defaultValue is z.unknown() and has no such shorthand. That asymmetry, two sections apart on one page, is precisely why the old sample looked plausible, so the new comment names it.

Also checked: the 26 YAML blocks on the page parse identically before and after (3 pre-existing non-parsing blocks are deliberate side-by-side illustrations, untouched).

Gates

gate	conclusion
check:doc-authoring	green — 374 files clean
check:docs-audit-scope	green — 179 hand-written docs in sync
check:nul-bytes	green — 6739 files, no raw control bytes
check:adr-links	green
check:quick-reference-counts	green
check:org-identifier, check:role-word	green
pnpm --filter @objectstack/docs build	green — full Next.js docs build, 391+ pages prerendered

The docs build matters here: CI's build-docs job filters on content/**, so this change triggers it.

Out of scope

Filed as #7373: a CEL defaultValue stores a raw Date into datetime/date fields, while the stored-value contract names an ISO-8601 string — the NOW() token normalizes via resolveNowDefault and the CEL branch has no counterpart. Reachable by the corrected sample this PR lands, but it is an engine defect, not a docs one, so it is filed rather than fixed here (ruling: file surface is this page and nothing else).

No changeset: docs-only, nothing user-visible ships. skip-changeset is the PM's to apply at accept time.

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
objectstack	Ignored		Aug 10, 2026 8:10am