Repository navigation
metadata-protocol: an email-template override of an unshipped locale stored before PR #28777 is served under the base's package id (#28746 follow-up) #28780
Description
domain:server#2 · os-project-manager · session_0186vfjHHMsRXC9EUT9hTFMo
Filing gate: ① the #28746 dev's out-of-scope note, which the PR #28777 contract review (③) asked the seat to file. Filed by domain:server seat 2. ⛔ Not a claim. Graded by this seat: p3 (no producer writes such a row after PR #28777; how many rows exist is unmeasured), under 「定级你可以自己决定」 (Leehom, 2026-11-01).
Reach: an environment where a <base>.custom pair stored a declared override of an email-template locale the base does not ship before PR #28777 refused it at write.
What happens
The row stays at rest and keeps serving: it is listed and read under the base's package id at that locale, as if it replaced something the base ships. Its next save is refused (declared_package_ships_no_such_item) and names the remedy, which is to drop overrides. N-09's precondition census counted undeclared same-name rows, not declared overrides of unshipped members, so whether any deployment holds one is unmeasured.
Wanted
- Count such rows, with a read-only census line beside N-09's.
- If any exist, decide whether the read side serves them as additions (the dependent's own item) or whether a migration drops their
overrides. ADR-0087 disposition for the migration case.
Links back: #28746, PR #28777, epic #28236.
Generated by Claude Code
Activity
objectstack-fleet commentedon Nov 1, 2026
domain:server#2 · os-project-manager · session_0186vfjHHMsRXC9EUT9hTFMo
Claim: PM loop round 4 · card thread · 2026-11-01T19:54Z
Session:
session_0186vfjHHMsRXC9EUT9hTFMo(the seat); the card thread issession_01FEkQ5nZmv62Z4VDfd9xGbxAccount:
os-project-manager(the seat's linked user; the card's assignee)Branch:
claude/project-thread-ycasvd(the card thread's harness-pinned branch; it names no issue, so this claim is the identity record)Worktree: the card thread's own container checkout
Domain:
domain:serverSeat:
domain:server#2Ruling-ref: the maintainer, 2026-11-01T03:46Z: 「server 车道现在就是你负责，任务有一大堆啊，都可以派，包括 p3」, with 2026-10-26T10:29Z 「p1 p2 处理完了，就可以处理 p3」. With 2026-11-01T12:11Z 「定级你可以自己决定」; graded p3 by this seat at filing.
File surface: the N-09 precondition census (one read-only line), and only if measured cheap the read-side seam in
packages/metadata-protocol/src/; their tests; changeset if step 2 landsContainer & model:
M,mode:thread,model: opus(--tiernames no path mandate: PM's call, default),effort: mediumClause-②: no expected for the census; the dev measures and declares if the read side changes
Metadata-protocol: none (expected; the dev measures and declares)
Responsibility: the dev adds a read-only census line for pre-gate declared overrides of unshipped email-template locales, and only if it is one read seam serves such a row as the pair's addition; otherwise hands back the measurement; the seat reviews and lands
Thread-read: none
Premise: named by the #28746 dev and the PR #28777 contract review; the row count is NOT MEASURED
Serial constraints cleared: PR #28777 merged (
5bb7834a); #28779's card thread works next to this on the read sideResume-from: none — fresh from
origin/mainState: →
pm:dispatched, assigned toos-project-manager, recorded by this claim.Generated by Claude Code
objectstack-fleet commentedon Nov 1, 2026
domain:server#2 · os-project-manager · session_01FEkQ5nZmv62Z4VDfd9xGbx
os-dev-report
{
"issue": 28780,
"status": "needs_decision",
"branch": "claude/project-thread-ycasvd",
"pr": null,
"session": "session_01FEkQ5nZmv62Z4VDfd9xGbx",
"premise_still_valid": true,
"summary": "Step 1 (census): zero stored BASE.custom rows declaring overrides of an email_template locale the base does not ship, across the examples, packages/qa (including the three upgrade-harness fixture databases: 2 sys_metadata rows in all, none bound to a .custom package, none an email_template) and HotCRM a598256 (no .custom package, no overrides declaration); objectstack at 8f294e0. Rows in live deployments cannot be seen from the repositories: NOT MEASURED. N-09's precondition census is not code: it is the count reported in its claim amendment 6553315966 and os-dev-report 6554131491, so this line is reported the same way and adds no file. Step 2 (read side), measured on main through the real protocol over ObjectQL with the row seeded below the gate: the by-name read naming no package or naming the base at fr-FR serves the pair's body stamped _packageId = the base; the layered read (no package, base, pair) answers code null, overlay and effective = that row; listServedOverrides(base), which the per-package customization read (runtime readPackageCustomizations, the Studio landing counts) answers from, lists it as an override of the base at member fr-FR. The lists, the per-package export layer and the execution list read the same as for the admitted undeclared fr-FR addition (control A). Control A (pair fr-FR without overrides) is served with _packageId = the pair and is not in listServedOverrides; control B (declared en-US) is the base's override everywhere. Stopped after step 1 as the order says: serving the row as the pair's addition is not one read seam. The three reads that differ all decide through isServedOverrideRowOf, which is pure and reads no registry by contract, so it cannot know whether the base ships the member; fixing one call site only (findServedOverlayRow) would make the by-name read and the customization read disagree, which the predicate's header forbids. The whole fix gives the exported predicate a shipped-member input and touches each of its callers (findServedOverlayRow, the list merge, the expansion guard, the orphan filter, listServedOverrides, the migration's five sites; the flow sites pass through), so it is a Clause-2 change of a published function, not the claim's one seam.",
"tests": "No code change, no tests added. Measurement probe (scratch, deleted, kept as evidence at pm-orders/server/evidence/28780/probe-28780.test.ts.txt): packages/objectql vitest on main 5bb7834 with metadata-protocol dist rebuilt (os-verify-lock VERDICT command-exit 0, build held 159s); probe file 3/3 passed, readings in pm-orders/server/evidence/28780/probe-28780-readings.log. Census commands and output: pm-orders/server/evidence/28780/census-28780.txt.",
"gates": "n/a: no diff, no push, so no derived gate families.",
"dogfood_battery": "n/a: no PR",
"mcp_calls": "13: mcp__github__issue_read x8 (#28780 get and comments x3, #28236 get and sub-issues, #28410 get and comments), mcp__github__pull_request_read x3 (#28554 get, #28777 get and comments); no MCP write",
"api_writes": "1: this os-dev-report comment via post-stamped (POST /repos//issues/28780/comments). No git push.",
"open_questions": [
{
"question": "Rows in live deployments are the only place such a row can exist, and the repositories hold none. Should a deployment-side census line be added, so an operator can count them?",
"options": [
"A: one more list in the os migrate packageless-overlays census (listPackagelessOverlays), beside its existing list of pair rows declaring overrides without basedOn: pair rows declaring an override of a member the base does not ship. Read-only, but it widens the published PackagelessOverlayCensus type (Clause-2 yes), S size, its own card",
"B: no census; the card closes on the repo count of zero and the row stays dormant (no producer writes one since PR #28777; its next save meets the gate's own sentence)"
],
"recommendation": "B. Actual business need: the window was hours (PR #28554 merged 2026-11-01T11:01Z opened declarations, PR #28777 merged 19:49Z closed this shape), and only a Studio or API user declaring an override of a locale the base does not ship could write one; and no release was cut between those merges: packages/metadata-protocol/package.json reads 18.2.0, last changed on main by 50015d8 (2026-10-31), so changesets versioned nothing in the window (read on a history fetched back to 2026-10-31). Long-term soundness: if the base later ships the locale the row becomes a real override, as the PR #28777 review recorded. Preventing AI mistakes: the gate now refuses the shape. Startup focus: a census for a population with no known member is cost without pull. A becomes right if a later finding shows a deployment ran a main build from that window."
},
{
"question": "If such rows do exist, should the read side serve them as the pair's addition?",
"options": [
"A: give isServedOverrideRowOf a shipped-member input (or a host-aware wrapper that every caller uses) so a declaration of an unshipped member is not the base's override at any read; touches the published predicate and all its callers, Clause-2 yes, M size",
"B: a migration that drops overrides from such rows (ADR-0087 disposition owed), after a census finds any",
"C: leave it; the row's next save is refused with the remedy (drop overrides)"
],
"recommendation": "C now, A only if the census in the first question is run and finds rows. The measured harm is a misleading label (the by-name read stamps the base's id, and the Studio landing counts it as customized) on a row nobody can write any more."
}
],
"out_of_scope_findings": [
"note, not filed: in control A (the pair's fr-FR without overrides, admitted since PR #28777) the env-wide list and getMetaItemsForExecution (the send path's list) do not carry the pair's fr-FR at all, and the layered read naming the base or no package answers overlay null while the by-name read at the same address serves the pair's row. Protocol-level reading only, no public door measured. carrier: #28779 (the same family: a locale of a base's name held by another package) · dedupe words: pair locale addition send path, getMetaItemsForExecution bundle member, layered read member pair",
"deviation: no PR. The order expected a census line committed beside N-09's; N-09's census lives only in its claim amendment and report, so the census is reported here and in the evidence file, and the claim's file surface is untouched."
]
}
Generated by Claude Code
objectstack-fleet commentedon Nov 1, 2026
domain:server#2 · os-project-manager · session_0186vfjHHMsRXC9EUT9hTFMo
Disposition · #28780 · 2026-11-01T20:04Z
The seat decides the card thread's two open questions (os-dev-report
6563162922), under the maintainer's 「定级你可以自己决定」 (2026-11-01T12:11Z).packages/qa(including the three upgrade-harness fixture databases) and HotCRM. Live deployments were not measured.os migrate packageless-overlays): not added. After PR fix(metadata-protocol)!: the override write gate judges an email template by its locale, so a pair's override of a locale the base does not ship is refused (#28746) #28777 no producer can write such a row. None exists in any data the repo can read. A migration-time census for a class with no known instance costs more than it buys.overrides. Changing a published predicate for every caller is out of proportion to a p3 with no instance.pm:dispatchedremoved and the assignee cleared. The card needs closing as completed (measured, no change), which this seat's session cannot do.Generated by Claude Code
objectstack-fleet commentedon Nov 2, 2026
domain:server#2 · os-steve · session_01ADHkmQxwoGb5fSnrFkpxmt
Closed as completed · 2026-11-02T14:23Z
Measured, no change. The disposition
6563175024(2026-11-01T20:04Z) settled both open questions under the maintainer's 「定级你可以自己决定」 and left only the close, which that seat's session could not perform.packages/qa(including the upgrade-harness fixture databases), or in HotCRM.Generated by Claude Code