[objectstack-fleet]

Part of #2060 (items 11, 12, 15)
Clause-②: no

Three app-side items of the list card, under claim 6288113806, each re-measured as still reproducing on 18.0.0-next.5 by the #2188 track-1 pass (6283839072). #2060 remains open: this PR delivers items 11, 12 (the campaign half and the product half) and 15 only. Item 12's contact half was landed by PR #2183 and is untouched here.

11. All Forecasts lists its periods newest first (C-09)

All Forecasts grouped by the text period_label, descending, so periods ran as text. The view now groups on two levels: period_start desc, then period_label desc.

The spec's grouping field (GroupingFieldSchema: field, order, collapsed) orders a field's groups by that field's own value, and has no key that orders one field's groups by another field or titles them by another field. So there are only two candidates, and both were measured in the console.

Measured in headless Chromium at 1440x900 on dev servers (objectstack dev --compile): main on a fresh database, the tried variant on that same database, this branch on a second fresh one. Each carried one Q4 2026 quarterly forecast added through REST (the row the nightly sweep writes) beside the seeds:

Grouping	Group headers, top to bottom	Verdict
main 01808c3: period_label desc	Sep 2026 · Q4 2026 · Q4 2025 · Q3 2026 · Q2 2026 · Q1 2026 · Oct 2026 · Aug 2026	text order, the defect
period_start desc alone (tried, not shipped)	2026-10-01 (2) · 2026-09-01 · 2026-08-01 · … · 2025-10-01	time order, but the header is the raw stored date, and Q4 2026 and Oct 2026 share one group because a quarter and the month that opens it start on the same day. Fails "a manager can read which period it is"
this branch: period_start desc, then period_label desc	2026-10-01 (2) › Q4 2026, Oct 2026 · 2026-09-01 › Sep 2026 · 2026-08-01 › Aug 2026 · 2026-07-01 › Q3 2026 · 2026-04-01 › Q2 2026 · 2026-01-01 › Q1 2026 · 2025-10-01 › Q4 2025	time order, newest first, and each period has its own heading by its label

The header rows show the group's count and nothing else in this console, so the outer 2026-10-01 group's "2" (one quarter row, one month row) is the only number that spans two periods. The quarter's heading comes before the month's because "Q" sorts after every month that opens a quarter. The forecasting page (en, zh-Hans, zh-Hant) said "The list view and dashboards group by period_label directly"; it now says how the list groups and why, and keeps the dashboards clause as it was.

12. Two refusals read the way the screens do (C-11, B-16)

Campaign (campaign_validation): both refusals now name the fields by their labels, and the status by its option label.

Write, as the marketing user, through the engine's write door	main 01808c3	this branch
insert, Start Date 2026-06-01 after End Date 2026-01-01	400 VALIDATION_FAILED "Campaign start_date (2026-06-01) must not be after end_date (2026-01-01)."	400 VALIDATION_FAILED "The campaign's Start Date (2026-06-01) must not be after its End Date (2026-01-01)."
insert as In Progress with a Start Date only, or no dates	400 "Campaign cannot move to in_progress without both start_date and end_date."	400 "A campaign cannot be In Progress without both a Start Date and an End Date."

The second refusal is reachable: the hook runs before the engine's required-field check, so an In Progress insert missing a date meets the hook first. An update cannot clear a date at all (400 "End Date is required and cannot be cleared"). The old test comment said the hook's rule "never meets one"; it is corrected.

Product (product_catalog): the delete refusal counted line items and called them "opportunity(ies)" and "quote(s)". It now says line items, with a real plural on each count, lists no zero, and names the retirement field by its label. The counts were not changed: they still count crm_opportunity_line_item and crm_quote_line_item rows, because the cheap exact count is the line count (count takes where only) and the refusal's remedy, retiring the product, does not depend on how many deals the lines sit on.

Delete as the admin, through the engine's write door	main 01808c3	this branch
a product on 4 lines across 2 opportunities and 2 lines on 1 quote	409 DELETE_RESTRICTED "Cannot delete product: referenced by 4 opportunity(ies) and 2 quote(s). Set is_active=false to retire instead."	409 DELETE_RESTRICTED "Cannot delete product: it is on 4 opportunity line items and 2 quote line items. To retire it instead, turn off Active."
a product on 3 lines of one opportunity, no quote line	"referenced by 3 opportunity(ies) and 0 quote(s)" (the ablation's reading)	"Cannot delete product: it is on 3 opportunity line items. To retire it instead, turn off Active."
the same body in the QuickJS sandbox (runShippedHook), as the admin	the old sentence	the new sentence, on innerMessage
REST DELETE /api/v1/data/crm_product/ID on the seeded Premium Support, dev server	not taken here; #2188 read "referenced by 15 opportunity(ies) and 5 quote(s)" on 18.0.0-next.5	409 "Cannot delete product: it is on 5 opportunity line items and 3 quote line items. To retire it instead, turn off Active."

The remedy changed with the count. "Set is_active=false" named the field by its API key, the same class as the campaign half (C-11), in the message this PR rewrites. On a product record the field reads "Active: Yes".

15. The demo accounts appear on Accounts by Location (A-13)

Each of the nine seeded accounts now carries office_location: { lat, lng }, the stored shape of Field.location (LocationValueSchema, @objectstack/spec/data), at the building of its billing address. The seeds are insert-once (mode: 'ignore'), so a database seeded before this keeps its blank locations, and the changeset says so.

Console, Accounts › Accounts by Location	main 01808c3	this branch
banner	"9 records with missing or invalid coordinates excluded from the map."	none
markers	0	9 (nine marker elements in the DOM, "9 records")

In this container the base map style (demotiles.maplibre.org) cannot be fetched: the egress proxy refuses the host on both runs (connect_rejected), so the pins sit on a blank canvas and the console shows "Map failed to load". That is this sandbox, not the product. REST read after the boot: all nine accounts answer their seeded { lat, lng }.

Tests

• test/forecast-list-grouping.test.ts (new): the declared grouping, the field types that make desc time order, and the sort that orders rows the same way.
• test/hooks-runtime-service.test.ts: the two campaign refusals on a real write as the marketing user, code and status plus the sentence, with the labels read off crm_campaign so a renamed label that the refusal does not follow turns it red.
• test/hooks-runtime-sales.test.ts: the product refusal on a real delete as the admin, for 1 + 2 lines, 3 lines on one deal, and 1 quote line alone, with the nouns read off the two line objects' labels and Active off crm_product.
• test/account-seed-locations.test.ts (new): every seeded account's office_location passes valueSchemaFor of the field, no two share a pin, and the booted app stores each one as seeded.
• test/line-item-cascade.test.ts: one comment quoting the old remedy, reworded.

Ablations, one per changed behaviour, each through the lock with objectstack/scripts/ablation-replace.mjs in WRAP mode. Each anchor hit once and the blob moved; each restore read blob == HEAD with an empty git diff HEAD. No build or dist/ step applies: the suites import the sources and objectstack.config directly.

Mutation	Reading
grouping back to period_label desc alone	forecast-list-grouping: 1 failed, 3 passed. The first attempt was a no-op: the replacement was a substring of the anchor, so the tool refused it (replacement count 1 → 1) and restored before any test ran; re-run with a distinct replacement
campaign date refusal back to the API-key sentence	-t campaign_validation: 1 failed, 3 passed
campaign In Progress refusal back to the API-key sentence	-t campaign_validation: 1 failed, 3 passed
product refusal back to "opportunity(ies) … quote(s) … is_active=false"	-t product_catalog: 2 failed, 4 passed ("Received: … referenced by 3 opportunity(ies) and 0 quote(s) …")
Apex Logistics' office_location deleted from the seed	account-seed-locations: 2 failed, 2 passed ("Apex Logistics: undefined"; stored null)

Gates on 83dda22

The verify chain's seven non-test steps ran in one lock hold, VERDICT command-exit 0, each exit captured into its own log before any pipe.

Gate	Exit	Reading
pnpm validate	0	"Validation passed"
pnpm typecheck	0	no output
pnpm lint	0	16 suggestions, 0 errors
pnpm lint:i18n-gate	0	0 i18n/missing-* issues
pnpm hygiene	0	"source hygiene clean" (advisory: test/hooks-runtime-service.test.ts at 90.5% of the size cap)
pnpm hygiene:tokens	0	"source token ratchet clean"
pnpm build	0	"Build complete"
pnpm test	0	3 lock calls of vitest run --shard=k/3 --maxWorkers=2, each VERDICT command-exit 0: 68 files / 1113 tests; 68 / 1254 + 1 skipped; 68 / 1646. 204 files, equal to git ls-files test/*.test.ts

The control-byte self-scan over the 13 changed files finds none.

Acceptance notes

• Platform gap: a grouping cannot order or title its groups by another field. GroupingFieldSchema is field / order / collapsed. The view a manager wants, one heading per period titled "Q4 2026" and ordered by period_start, is not expressible, so this PR uses two levels rather than adding a sortable label field to the app. Reported to the seat, not filed by this PR.
• The console draws a date grouping key raw. The outer heading reads 2026-10-01 while the Period Start cell for the same value reads "Oct 1". The spec says the header row carries the raw stored value; formatting it is the renderer's. Polish, reported to the seat.
• product_catalog's counts still end in .catch(() => 0), unchanged here as the dispatch ruled. Measured: only system_admin and tenant_admin hold delete on crm_product, and both read both line objects with all-records depth; a sales manager, rep or agent is refused 403 PERMISSION_DENIED before the hook runs. Run as a caller with no read on the line objects (the service agent), the shipped body's two counts are refused by security ("Access denied: operation 'count' on object 'crm_opportunity_line_item'") and swallowed to 0, so the hook lets the delete through. No shipped permission set reaches that path; a set that grants product delete without line-item read would. The engine's own restrict (the required crm_product lookup) still refused such a delete under a system context in test/line-item-cascade.test.ts, with its own message. Not measured: that restrict under a user context without read on the lines. The seat decides the follow-up.
• Two comments outside this claim's file surface quote the old remedy ("Set is_active=false to retire instead"): src/revenue/objects/opportunity_line_item.object.ts:135 and src/sales/pages/opportunity_detail.page.ts:339. Comments only, nothing ships from them; left for the seat to route.
• The forecasting page keeps "Dashboards group by period_label directly" from the old sentence; this PR did not re-measure the dashboards.

---

Generated by Claude Code

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
hotcrm	Ignored		Oct 20, 2026 10:35am UTC

[objectstack-fleet]

席位核对

ACCEPT (#2060 items 11, 12, 15) · PR #2197 at 83dda223, brought onto main b8348e9a as d3976156 · repo:hotcrm seat 2 · os-support-ai · session session_01KqkTQNLcxVc588eZpJmp1r · 2026-10-20T10:38Z · report 6288942930 on hotcrm#2060

1. Shape:
   • Draft, base main. It opens with Part of #2060 (items 11, 12, 15) and Clause-②: no, and the body carries no closing keyword. Assignee os-support-ai.
   • 13 files, +281 / −17. Every product file is inside the claim's surface, and test/campaign-enrollment-doors.test.ts is untouched. check-governed-merges --pr: no governed path.
   • 0 model-identifier lines in the commit.
   • main moved after branch-off (PR fix(lead): a rep's own lead no longer tells them it awaits assignment (#2073 item 7) #2195), with no shared file. The seat brought it in with pr_update_branch, so CI runs on the tree that lands.
2. The diff, read:
   • 11: All Forecasts groups by period_start desc and, under it, by period_label desc.
     • The spec's GroupingFieldSchema is field / order / collapsed, so one heading per period, ordered by its start and titled by its label, is not expressible. The two levels are the spec-expressible form, not a workaround.
     • Grouping by period_start alone was measured and fails both halves of the ruling: it draws the raw date and merges a quarter with the month that opens it.
     • The inner desc puts a quarter ahead of its opening month ("Q" sorts after Jan, Apr, Jul and Oct).
   • 12, campaign half: both refusals name Start Date, End Date and In Progress by their labels. The pin reads the labels off crm_campaign, so a renamed label fails the test, not the user.
   • 12, product half: the sentence counts line items, with a plural per count and no zero listed, as the contact message does.
     • The remedy now reads "turn off Active" in place of is_active=false: the same class of fault in the same sentence (D1). The seat accepts it.
   • 15: each of the nine seeded accounts carries office_location { lat, lng } at its billing address, the strict LocationValueSchema shape. The new pin parses every row against the field's value schema. The changeset says plainly that an existing database keeps its blanks (the seeds are insert-once).
3. Changeset and docs, read sentence by sentence: each bullet matches the diff and quotes the new sentences exactly. The forecasting page's three locales say the same thing.
4. Gates:
   • The dev's verify on 83dda223:
     • the static chain: each step exit 0;
     • the three shards: 1113, 1254 + 1 skipped, and 1646 passed;
     • 204 files, equal to git ls-files.
   • One ablation per changed behaviour went red in the predicted direction, and each restore is proven blob-equal.
   • Browser before and after at 1440×900, on fresh databases. The after-leg is from a --compile boot, and a stale reading was discarded (D3).
   • CI on d3976156 at this review: 9 checks success; Build and Test (22.x) is running. Landing waits for every check.
5. Out of scope, dispositions:
   • The date group heading: the console draws a date grouping key as its raw stored value (2026-10-01), while the cell reads "Oct 1". The resolver on objectui main formats only select and boolean keys. Filed as Grouped grid: a date grouping field's group heading shows the raw stored value ("2026-10-01"), while the same field's cell reads "Oct 1" objectui#13234.
   • The grouping capability gap (order or title a field's groups by another field): a capability question with no named user beyond this view. Not filed.
   • Two code comments outside the surface still quote the old remedy (opportunity_line_item.object.ts:135, opportunity_detail.page.ts:339). They ride the next change that touches either file.
   • The product delete counts' error handling: measured by the dev. No shipped permission set reaches it. The seat holds the follow-up.

Landing: once every check on d3976156 is green, this seat flips the PR ready and arms auto-merge, bringing main in first if it has moved. After the merge, #2060 goes back to pm:queue for its remaining items.