[objectstack-fleet]

Fixes #273

Slot: Sun 2026-10-25, on the shared calendar on #267. Merge on the slot day only. Seat review. The seat found one visual defect: two cells in the three-ways diagram ran flush to their card edge. It was fixed in 3a2a799, which also fixed three more tight boxes found by a clearance scan. The seat re-viewed the diagram at native size. The measured "dry run of an update" limit stays in the post (writing rule 7). The upstream defect is filed separately in objectstack, and the post does not cite it. ACCEPT.

Slot Sun 2026-10-25 (date: 2026-10-25 in every locale, per the coordinator's calendar change during the run; the card said Tue 2026-10-27). Merge on the slot day only.

What this adds

• content/blog/how-to-import-a-spreadsheet-from-an-old-system/ — the post in all eight locales (index.mdx, index.zh-Hans.mdx, index.zh-Hant.mdx generated by pnpm gen:zh-hant, index.ja.mdx, index.de.mdx, index.es.mdx, index.fr.mdx, index.ko.mdx), a cover and two diagrams, each as an English-text asset (cover-en.svg, mapping-is-a-projection-en.svg, three-ways-a-row-fails-to-match-en.svg) and a Chinese-text pair (cover.svg, mapping-is-a-projection.svg, three-ways-a-row-fails-to-match.svg). Topic integration-data, audience business, solution crm.
• Ownable concept: the dry run is the review — a before/after count signed off before any row lands. The one number: a 3-row file reports created 3 from the dry run while the table holds 0 rows, 3 after the real run, updated 3 on the second run.
• Back-links, one sentence plus updated: 2026-10-25T10:00:00+08:00 each: how-to-move-an-internal-system-to-metadata (English only, as published), extend-existing-systems-with-ai and manufacturing-legacy-systems-ai (all seven authored locales; zh-Hant regenerated).

Proof artifact — measured on the published @objectstack/cli 17.7.0

A fresh npm create objectstack@latest crm scaffold (Create ObjectStack v17.7.0; @objectstack/cli@17.7.0, @objectstack/spec@17.7.0, @objectstack/runtime@17.7.0; npm latest = 17.7.0) carrying the CRM example's Lead object from examples/app-crm/src/objects/lead.object.ts at tag @objectstack/cli@17.7.0 (its two lookups to crm_account / crm_opportunity left out, said in the post) and a defineMapping for crm_lead registered through mappings:. Server objectstack dev --no-watch -p 49180, dev admin session, every call POST /api/v1/data/crm_lead/import. Logs under scratchpad/issue-273/proof/ (run.log, run-dryrun-update.log, run-projection.log, run-413.log, run-saved.log, build-javascript.log); the app itself under scratchpad/issue-273/app/crm/.

Step	Request	Report (quoted)	Rows in table after
Dry run	mappingName: "crm_lead_feed", dryRun: true, 3 rows	total 3, ok 3, errors 0, created 3, updated 0, skipped 0	0
Real run	same body without dryRun	created 3, an id on every row; source stored as web, partner, web	3
Same file again	same body	created 0, updated 3	3
Update mode	writeMode: "update", 2 rows (one known, one new)	updated 1, skipped 1; row 2 skipped, code: "NO_MATCH"	3
Edge rows, upsert	unknown code Trade Show / label Event / not-an-email / blank email	row 1 failed, field: "source", code: "invalid_option", "Lead Source: "Trade Show" is not a known option"; row 2 created (source event); row 3 failed, field: "email", code: "invalid_email"; row 4 created; ok 2, errors 2	5
Blank key, skipBlankMatchKey: true (dry run)	same edge file	row 4 skipped, code: "BLANK_MATCH_KEY"	5
Ambiguous	one row loaded twice with writeMode: "insert", then upsert	two records with one email; upsert row failed, code: "AMBIGUOUS_MATCH", "matchFields matched more than one crm_lead record"	7
Strict projection	a second 3-row file (Channel cells are our own labels, plus a Legacy ID column): inline mapping dry run / real run; mappingName dry run	inline: 0 of 3, each failed, code: "INVALID_FIELD", field: "Legacy ID", "Unknown field 'Legacy ID' on object 'crm_lead'" (dry run and real run alike); artifact: 3 of 3 created	7
First file through the inline rename (dry run)	the Webform / Partner Referral file, inline mapping	3 of 3 failed, code: "invalid_option", field: "source", "Lead Source: "Webform" is not a known option"	7
Ceiling	5,001 rows, dry run / 5,000 rows, dry run	413 PAYLOAD_TOO_LARGE, "Import limit is 5000 rows per request (got 5001). Use an async import job for larger files." / 200, total 5000, ok 5000, created 5000	7
Runtime-saved mapping	PUT /api/v1/meta/mapping/crm_lead_feed_saved (same shape, new name); dry run by that name; PUT to the packaged name	200, "Saved mapping 'crm_lead_feed_saved' (env-wide, state=active)"; created 3; 403 NOT_OVERRIDABLE	7
javascript transform, build time	one entry switched to transform: 'javascript', objectstack build	exit 2: "Mapping 'crm_lead_feed' uses transform 'javascript', which the import path does not execute (no server-side sandbox). Use none/constant/map/split/join/lookup, or model the logic as a flow." (restored; artifact hash identical before/after)	—
javascript transform, runtime-saved	PUT .../crm_lead_feed_js with a javascript entry, then import by name	save 200; import 400 UNSUPPORTED_TRANSFORM	7
Whole-request refusals	both mappingName and inline mapping / unknown name	400 CONFLICTING_MAPPING, "Provide either mappingName or an inline mapping, not both" / 404 MAPPING_NOT_FOUND, "No mapping artifact named "crm_lead_feed_nope" is registered"	7

Two things the card predicted came out differently and the post says what was measured: (1) a row with an unknown Channel code passes through the map unchanged, as documented, and then fails at select coercion with invalid_option naming source — it is not a row that lands unchanged; (2) the strict-projection contrast could not be shown on the card's 3-row file, because the inline route fails that file one step earlier (untranslated codes), so a second file was built whose only difference from a clean one is the junk column.

Every factual claim and its source line

objectstack at tag @objectstack/cli@17.7.0 (ba459d8d5); content/docs/data-modeling/import-mappings.mdx is byte-identical at the tag and at local origin/main (diff empty).

Claim in the post	Source
"An import mapping is a named, reusable projection from the columns of a source file onto the fields of one object."	import-mappings.mdx:7-8
first-class metadata kind; ships in a package or is saved at runtime; POST /api/v1/data/:object/import applies it by name with mappingName	:8-10
for a single ad-hoc file the wizard's inline rename is enough; two mechanisms with different semantics	:12-15
the defineMapping shape (name, label, sourceFormat, targetObject, fieldMapping[], transform: 'map' + valueMap, mode: 'upsert', upsertKey: ['email']); registered with mappings: [...]	:22-43, :48-53; the post's own mapping, compiled and served (dist/objectstack.json, GET /api/v1/meta/mapping/crm_lead_feed answers it with _provenance: "package")
defineStack() validates while you build: an undefined targetObject fails the build; a javascript transform fails the build	:56-61; measured exit 2 (above)
defineMapping() parses at import time; a misspelled key is rejected with the canonical spelling named	:63-68
two origins (package / PUT /api/v1/meta/mapping/NAME); saved rows read first, packaged registry as fallback; packaged mappings locked against tenant edits; no per-organization overlay	:74-86; measured 200 saved under a new name and 403 NOT_OVERRIDABLE under the packaged name
mode supplies writeMode when omitted; upsertKey supplies matchFields	:96-100, :181-184
the seven TransformType values; "A value with no entry in the table passes through unchanged"; lookup pass-through; javascript "Refused. There is no server-side sandbox, and silently skipping a declared transform would corrupt data"	:118-129; packages/core/src/utils/import-mapping.ts:221-226 (raw in valueMap ? valueMap[raw] : raw)
"For logic beyond these, transform the data before you post it, or model it as a flow"	:131-132
the request and the report shape (object, dryRun, writeMode, total, ok, errors, created, updated, skipped, results[])	:136-169; packages/rest/src/rest-server.ts:9347-9358
"dryRun": true returns "the same report with nothing persisted; the verdict comes from the engine's own write-path validation"	:176-177; packages/spec/src/api/export.zod.ts:274-282
runAutomations is a request key (named only, not described)	:186-187
inline mapping vs mappingName mutually exclusive, 400 CONFLICTING_MAPPING; unmapped columns "pass through to the write path under their own header" vs "Dropped. The artifact is a strict projection: only mapped targets survive"	:195-203; packages/core/src/utils/import-mapping.ts:151-152 ("Returns NEW rows containing only mapped targets")
whole-request refusals table (404 MAPPING_NOT_FOUND, 400 MAPPING_TARGET_MISMATCH, 400 UNSUPPORTED_TRANSFORM, 400 CONFLICTING_MAPPING, 413 PAYLOAD_TOO_LARGE over 5,000 rows)	:215-224; packages/rest/src/import-mapping.ts:63-96; measured 404 / 400 / 413
per-row outcomes (created, updated, skipped NO_MATCH, skipped BLANK_MATCH_KEY, failed AMBIGUOUS_MATCH, failed with a field code); "A row that fails does not stop the import"; "a row that does not match is not one behaviour but three"	:228-244; packages/core/src/utils/import-runner.ts:962-986
select labels resolve to stored values without a mapping, translated labels too; "A cell that cannot be coerced fails its own row with the offending column named"	:254-265
ceilings 5,000 synchronous / 50,000 via POST /api/v1/data/:object/import/jobs, job id, poll, cancel/undo	:271-274; measured 5,001 → 413, 5,000 → 200
the Lead object's source options web, referral, event, social, cold_outreach, partner; name, email, company fields	examples/app-crm/src/objects/lead.object.ts:19-40, 57-67 at the tag
the CRM example declares no mappings: today (the post adds one)	examples/app-crm/objectstack.config.ts:37-120 at the tag
npm create objectstack@latest installs 17.7.0	npm view @objectstack/cli dist-tags → latest: 17.7.0; scaffold banner "Create ObjectStack v17.7.0"
T4 standalone and E1 in the closing section, verbatim per locale	src/lib/positioning.ts (T4_STANDALONE, CANONICAL.e1, and the zh-Hans / ja / de / es / fr / ko strings)
the installed skills bundle does not mention defineMapping (so the CTA points at the docs page)	git grep defineMapping '@objectstack/cli@17.7.0' -- 'skills/**' → no hits

A measured limit the post states (and a finding for the ObjectStack team)

On 17.7.0, a dry run of rows that resolve to an update failed every row with code: "rule_violation" on the Lead object's lead_score_range script rule — "Validation rule 'lead_score_range' could not be evaluated (runtime: No such key: lead_score) — write rejected. The predicate reads 'lead_score', which this object does not declare" — while the real run of the same body answered updated 3, and a direct PATCH of one record with the same four fields answered 200. The object does declare lead_score; the preview evaluated the rule against the incoming cells alone. Previews of rows that are created agreed with the write in every run. The post keeps this as a version-named paragraph in "What this does not solve" (seat ruling A); the post does not cite the filing. Evidence: proof/run-dryrun-update.log A–D; server log lines "Validation rule 'lead_score_range' predicate failed to evaluate (runtime: No such key: lead_score". Contract it breaches: packages/core/src/utils/import-runner.ts:28-40 at the tag ("A dry run's contract is that it reports the verdict the real write produces"). Filed through the seat, not by this PR.

Gates (final tree, tip 3a2a799; two commits on lin/issue-273: bfc0687 the post, 3a2a799 the seat-review fix)

Gate	Result	Log
pnpm content:lint	pass, 355 files; warnings only (title/description width, as the reference post; "Date is in the future" / "Updated is in the future" for the slot)	scratchpad/issue-273/clint-5.log
pnpm content:lint --published	pass (inside the build)	build-5.log line 71
pnpm svg:lint	OK — 124 SVGs, 2,703 text boxes, no crossing; a verbose clearance scan of the six new SVGs finds no text box under 12px from its card edge on any side (smallest right clearance in the three-ways cards: 20px)	svglint-final-2.log, build-5.log line 72
pnpm check (via os-verify-lock.sh)	VERDICT command-exit 0 — 160 files, 0 errors / 0 warnings / 0 hints	check-final-3.log
pnpm build (via os-verify-lock.sh)	VERDICT command-exit 0; content lint --dist passed (494 built blog pages); CJK line breaks, quotation marks and links passed; git status clean afterwards	build-5.log
pnpm seo:smoke	passed — 910 HTML pages, 17,517 prose blocks language-checked, 99,477 internal links	seo-4.log

All descriptions ≤ 230 characters (en 224, zh-Hans 97, ja 121, de 228, es 221, fr 227, ko 133).

Browser check (actually looked at)

astro preview on 127.0.0.1:49181, Chromium via Playwright, script scratchpad/issue-273/shots.mjs, screenshots under scratchpad/www/shots-273/ (123 files, results.json). 265/265 assertions pass (shots-4.log, on the dist/ of 3a2a799): the article in en and zh-Hans at 1366×625, 1440×750, 1512×830, 1920×945, 390×844 and 360×800, light and dark, lazy images scrolled into view first — HTTP 200, no horizontal overflow outside a scroll container, no repeated image, the cover renders, every image loads; the blog list in en and zh-Hans at 1440 and 390, light and dark, links the new post; the three back-linked posts show the new link and an Updated line; every new SVG at native size on a light and a dark page, lowest text contrast 4.59:1 (first pass found 4.3:1 on the muted grey over the cream cards; lifted to 5.5:1+). Looked at: both covers light/dark, both diagrams light/dark, the article at 1440 light, 1920 dark, 390 dark, 360 light, the zh-Hans article at 1366, phone-width crops of a code block and the tables (horizontal scroll containers), the list page.

Deviations and notes for the seat

• Seat-review fix (3a2a799): in three-ways-a-row-fails-to-match (en and zh) two cell texts ran flush to the upsert column’s right border inside the gate’s 2px tolerance; shortened to skipBlankMatchKey → skipped and AMBIGUOUS_MATCH, not written. A verbose clearance scan of all six SVGs then also lifted the last line of each bottom box in mapping-is-a-projection (7px from the card edge; boxes grown to 178px) and shortened one zh record-card line (3px from the right edge). The sentence referring to the measurement being with the ObjectStack team is removed from every locale.

• Slot changed mid-run by the coordinator from 2026-10-27 to 2026-10-25; applied to every date and updated. The Publishing cadence: one post a day, switched on once seven approved posts are queued (maintainer ruling B, 2026-10-17) — charter change, shared calendar, writing capacity #267 calendar table still lists Post 4 (Post (slot Tue 2026-10-20): Agent-Operable Apps: When Every Object and Action Is an MCP Tool Under the User's Permissions #230, PR content(blog): agent-operable apps — every object and action an MCP tool under the user's permissions (post 4, published 2026-10-26) #237) on Tue 10-27 and nothing on 10-25; the seat owns that table.

• The non-English locales link the English-only how-to-move-an-internal-system-to-metadata at /en/… (the pattern the reference post uses; the content-lint cross-locale rule accepts it). de/es/fr/ja/ko link the glossary term at /en/glossary/typed-metadata/ because content/glossary/{de,es,fr,ja,ko}/typed-metadata.ts do not exist and the same-locale URL renders a noindexed fallback (lint warns on it).

• de/es/fr: the two quoted English server messages are wrapped in a span with lang=en so the SEO language check reads them as English (the first smoke run flagged the paragraph).

• ja and zh-Hans: a bold that ended on a full stop before a CJK letter rendered as literal asterisks (content-lint --dist); the full stop moved outside the bold.

• Commit trailers are the two lines the dispatcher asked for (Co-Authored-By: Claude, Claude-Session: …), not the harness reminder's model-named form.

• The worktree /home/user/www-issue-273 is kept with its node_modules for the seat to push lin/issue-273; nothing was pushed, no PR opened, nothing posted to GitHub (four gh api reads only: issue Post (slot Sun 2026-10-25): How Do You Import a Spreadsheet From an Old System Into an AI-Built App? Import Mappings, Dry Runs, and the Three Ways a Row Can Fail to Match #273, its comments, open PRs, issue Publishing cadence: one post a day, switched on once seven approved posts are queued (maintainer ruling B, 2026-10-17) — charter change, shared calendar, writing capacity #267).

---

Generated by Claude Code

Slot-ahead refresh: re-measured on @objectstack/cli 18.3.0 (2026-11-04)

npm view @objectstack/cli dist-tags on 2026-11-04 returned { next: '18.2.0', latest: '18.3.0' }. 18.3.0 was published 2026-11-03T15:16Z. The tag @objectstack/cli@18.3.0 is objectstack 83d7e4ae33 ("chore: version packages (#26517)").

Measurement setup. All scratch files are under scratchpad/ref-283/.

• Scaffold: a fresh npm create objectstack@latest app -- --skip-skills. npx os --version reports @objectstack/cli/18.3.0 linux-x64 node-v22.22.0. All @objectstack/* packages are published 18.3.0.
• Example code: examples/app-crm and content/docs/data-modeling/import-mappings.mdx were extracted from the tag with git archive into crm-src/. The /home/user/objectstack checkout was not edited.
• Objects: lead.object.ts was copied into the scaffold. Its lookup targets account.object.ts and opportunity.object.ts were copied too, because without them os validate fails with object-reference-unknown. The scaffold's app_note was dropped and the manifest namespace was set to crm, because the namespace-prefix rule demands it.
• Mapping: the post's crm_lead_feed mapping, verbatim, was registered with mappings: [LeadFeedMapping].
• Validate: os validate passed: 63 author-time rules, Data: 3 Objects 29 Fields.
• Server: booted with npx os dev --compile --no-watch --fresh -p 49800 and signed in as the banner admin.

Logs.

• rest-transcript.log covers every REST step.
• rest-dup.log covers DUPLICATE_MATCH_KEY.
• build-js.log, build-target.log and build-typo.log cover the build-time refusals.
• validate.log is the validate run.

Claim by claim

The edits went into en, ja, ko, de, es, fr and zh-Hans. zh-Hant was regenerated with pnpm gen:zh-hant.

#	Claim	Branch text (17.7.0)	18.3.0 measured	Action
1	Version label	17.7.0: description, short version, intro, alt text of the three-ways diagram, "Both halves … measured on" line, plus 6 SVGs (cover ×2 each, projection ×1 each, three-ways ×2 each)	18.3.0	relabelled everywhere
2	Inline route on file B (labels + Legacy ID)	dry and real: 0 of 3, INVALID_FIELD, field: "Legacy ID", "Unknown field 'Legacy ID' on object 'crm_lead'"	identical	—
3	mappingName dry run on file B	3 of 3 created	identical	—
4	Inline route on file A (codes)	invalid_option on source, "Lead Source: "Webform" is not a known option"	identical (rows now also carry an errors[] list, 18.3.0 #25582; not quoted in the post)	—
5	Inline mapping = "a flat rename"	flat rename	18.2 added an array form with transform (trim/uppercase/lowercase), defaultValue, required and applyOn (import-mappings.mdx:201, :209-229). The flat form is still a rename, and neither form has a code table	sentence rewritten: "a rename: its array form can trim a cell, change its case or fill a blank one, but it has no code table"
6	Runtime save reply	Saved mapping 'crm_lead_feed_saved' (env-wide, state=active)	… (env-wide, state=active) [seq=1]	quoted whole
7	Saved mapping dry run	created 3	identical	—
8	PUT to the packaged name	403 NOT_OVERRIDABLE	identical (message: "…is provided by a managed package and is sealed…")	—
9	Dry run → table → real → table → again	total 3 · created 3 · updated 0, total 0, created 3 with ids, total 3 with web/partner/web, created 0 · updated 3, 3	identical	—
10	Limit "dry run of an update fails a rule reading a field the file lacks"	3× rule_violation on lead_score_range, while the real run gave updated 3	Fixed. The dry run of the same file after the rows existed answered created 0 · updated 3, errors 0. The preview judges the stored row merged with the patch (core 18.0.0 35ef501, #22445; import-runner.ts:1511-1516 at the tag)	paragraph removed from "What this does not solve". A new paragraph in "The dry run is the review" gives the measured update preview, the merge mechanism, a "Before 18.0 …" history clause, and the warnings list on admitted rows (import-runner.ts:1148-1151; "none of the runs here produced one")
11	update file: unknown row	updated 1, skipped 1, NO_MATCH	identical	—
12	Insert twice, then upsert	AMBIGUOUS_MATCH, "matchFields matched more than one crm_lead record"	identical	—
13	Blank E-mail	upsert created; skipBlankMatchKey → skipped, BLANK_MATCH_KEY	identical	—
14	(new) Duplicate key within one file	not covered	18.2 added DUPLICATE_MATCH_KEY. A 3-row file with frances@example.com twice answered ok 1 · errors 2 on both the dry run and the real run. Message: "Rows 1, 2 of this import have the same match key (email), so none of them was written: keep one row per key and import again"	new paragraph after the blank-key paragraph. The message is in a text block in all 8 locales
15	4-row coercion file	Trade Show → invalid_option; Event → created (stored event); not-an-email → invalid_email on email; blank → created; ok 2, errors 2	identical	—
16	MAPPING_NOT_FOUND 404 text	"No mapping artifact named "crm_lead_feed_nope" is registered"	identical	—
17	CONFLICTING_MAPPING 400 text	"Provide either mappingName or an inline mapping, not both"	identical	—
18	MAPPING_TARGET_MISMATCH 400 (doc row)	documented only	also measured now: "Mapping "crm_lead_feed" targets object "crm_lead", not "crm_account""	unchanged ("three measured" is still true)
19	413 text	"Import limit is 5000 rows per request (got 5001). Use an async import job for larger files."	identical. The suffix is appended by data-action-routes.ts:581	—
20	5,000-row dry run	total 5000, ok 5000, created 5000	identical	—
21	50,000 job ceiling	doc	IMPORT_JOB_MAX_ROWS = 50_000 (spec/src/api/export.zod.ts:486); doc :317	—
22	TransformType seven values	doc + spec	mapping.zod.ts:185-193, same seven	—
23	javascript build refusal text	"Mapping 'crm_lead_feed' uses transform 'javascript', which the import path does not execute (no server-side sandbox). Use none/constant/map/split/join/lookup, or model the logic as a flow."	identical (os build exit 2)	—
24	Runtime javascript	save accepted; import 400 UNSUPPORTED_TRANSFORM	identical	—
25	Unknown targetObject fails the build	doc	measured: "Mapping 'crm_lead_feed' targets object 'crm_leads' which is not defined in objects."	—
26	Misspelled key rejected with its canonical name	doc callout	measured: fields: → "Did you mean fields → fieldMapping?"	—
27	Doc quotes (definition, "Dropped. … strict projection", dryRun sentence, "there is no server-side sandbox…", three outcomes, labels resolve)	—	all present at 18.3.0. The doc changed between 18.1 and 18.3 only by additions (applyOn, key sets, new refusal rows, DUPLICATE_MATCH_KEY, VALIDATION_ERROR)	—
28	Lead fields, source options, lead_score_range predicate	—	unchanged at the tag	—
29	date: 2026-10-25	—	not touched	—

Merge. origin/main (90e084d) was merged in with a merge commit. The only conflict was the updated: line of extend-existing-systems-with-ai in all 8 locales. The branch had 10-25 and main had 10-29. I kept main's later 10-29, because a back-link stamp has to be at least the latest revision. The seat re-dates it at switch-over anyway.

Visual fix found by the shots. On a Linux sans-serif fallback, the third card's caption in cover-en.svg, "upsertKey: ['email'] decides what "the same row" means", ran about 8px past the card's right edge. svg-text-fit measured it as inside, so it passed. I shortened it to "upsertKey: ['email'] defines "the same row"". The zh caption is short and was unaffected.

Stale-version grep over the post's mdx and svg files:

• 17.7 and 18.1: 0 hits.
• 18.2: 8 hits, one per locale. All of them are the deliberate history note "One more case arrived in 18.2" (DUPLICATE_MATCH_KEY).
• 18.0 appears only in the deliberate "Before 18.0 …" clause.

Gates

Worktree /home/user/www-ref-283, branch lin/ref-283, head fd93e9f. All gates were re-run on the final head.

• pnpm content:lint: exit 0, ✓ content lint passed (407 files, 46 glossary terms checked). The only notes on this post are the existing title/description width notes; the lengths did not change.
• pnpm check, under os-verify-lock.sh slot www-ref-283: Result (167 files): 0 errors.
• pnpm build, under the lock: exit 0.
  • gen-zh-hant ran, and content-lint --published passed.
  • svg-text-fit: OK (166 SVGs, 4061 text boxes).
  • cover-spec: OK — 0 covers of posts dated on or after 2026-11-17, 90 grandfathered. This post is dated 10-25, and its 11-06 slot is also before 11-17.
  • cover-dark: OK.
  • 972 page(s) built.
  • --dist checks passed: pages, CJK breaks (0), quotation marks, and 110,974 links.
  • Lock verdict: VERDICT command-exit 0 · held the lock 95s · waited 11s.
  • The tree was clean afterwards (0 porcelain lines).
• pnpm seo:smoke: passed (962 pages, 20,607 prose blocks language-checked).

Browser checks.

• Pages: en, zh-Hans, de and ko at 1440×750 and 390×844, light and dark, 16 loads. Every load returned HTTP 200 with scrollWidth equal to clientWidth (no overflow). There were 4 images per page and 0 broken. The one hidden image on each page is the cover variant for the other colour scheme, which has no layout box. Each page had 1 Rows 1, 2 … text block and no "17.7" text.
• Shots: 12 crops of the new update-preview paragraph and the DUPLICATE block (en@1440 light, en@390 dark, zh-Hans@390 light and @1440 dark, de@390 dark, ko@390 light), saved in www/shots-ref-283/. I looked at all of them. The one-line message scrolls horizontally inside its pre with the site's fade, which is the standard transcript behaviour.
• Word-break probe at 390/360: 7 split words across 8 locales.
  • 6 break right after a literal hyphen (AI-|Built, KI-|gebaute, Import-|Mappings, and others).
  • 1 is CSS hyphens:auto with a rendered hyphen: "AI-geschrie-|benen" on the de /blog/ first card. That card is the validation-rule-hook-or-flow post, not this one. I confirmed it in a shot.
  • Result: 0 non-hyphen breaks.
• SVGs: all 6 changed SVGs plus the generated dark cover were rendered at native viewBox size (1200 wide) and looked at.

Cleanup. Every process was killed by PID: the dev server tree (1296/1298/1310/1311/1369) and the static server (4955). No push, PR or comment was made.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FeA1nwBz1ohH65dvffUGKr

[objectstack-fleet]

席位核对(发布日 2026-11-06)

ACCEPT。

• 形态:31 个文件,+2204/−1,低于 3000 行,不涉及受管面。head 为 e26d20f,由以下提交组成:
  • fd93e9f:11-04 已在 18.3.0 上完成刷新;
  • 530acca:合并 origin/main 的提交,无冲突;
  • 发布日改日期的提交,17 个文件 +23/−23,改动只有 date: 2026-11-06,以及 9 个回链文件的 updated:。
• 前提实测:2026-11-06 02:23Z 运行 npm view @objectstack/cli dist-tags,latest = 18.3.0,与文中标注的版本一致,所以只改日期。在文章目录里 grep 旧版本号 18.[0-2].x / 17.x.x,命中 0 处。
• 门禁 check-runs:head e26d20f 上 Build and deploy = success;Post-deploy probe = skipped,PR 上按设计跳过。本地门禁:
  • content:lint exit 0;
  • check + build(在锁下运行)exit 0,共 1003 页,dist 链接、CJK 断行、引号检查均通过;
  • seo:smoke exit 0。
• 抽查读数:
  • 截图看过 8 种语言的浅色版,以及 en/zh-Hans 的深色版。zh-Hans 显示「2026年11月6日」,标题正常。
  • /en/blog/ 的 Latest Articles 第一位就是本文。
  • 390/360 宽度下,标题和正文没有词中断行。
• 接受的偏差:
  • zh-Hant 由 pnpm gen:zh-hant 重新生成,按 11-04 的术语映射改了几处用词:後設資料→中繼資料、執行時→執行環境。这是派生产物,2026-11-04 的裁决允许。
  • 表格里的代码 token(如 AMBIGUOUS_MATCH)在 390/360 下因 overflow-wrap:anywhere 词中换行,页面不出现横向滚动。这与已立卡的 fix(site): inline code in narrow table cells splits identifiers mid-token on phones (beforeDelet|e) #333 同属一类,记在那里,不在发布日处理。
• 范围外去向:表格代码换行 → fix(site): inline code in narrow table cells splits identifiers mid-token on phones (beforeDelet|e) #333。
• 落地条款:按 pm-dispatch,席位 ACCEPT 后自行落地:去掉标题里的档期日期 → ready(已是 ready)→ 合并队列 → 确认 push 部署和 post-deploy probe → 在 Post (slot Sun 2026-10-25): How Do You Import a Spreadsheet From an Old System Into an AI-Built App? Import Mappings, Dry Runs, and the Three Ways a Row Can Fail to Match #273 和 Publishing cadence: one post a day, switched on once seven approved posts are queued (maintainer ruling B, 2026-10-17) — charter change, shared calendar, writing capacity #267 上记录落地,并更新日历评论。

---

Generated by Claude Code