spec: the root entry stops carrying the migration and conversion registries (about 1 MB of prose), so a console first screen no longer downloads os migrate text (the upstream payback of objectui#11088 D1)

[objectstack-fleet]

Filed by the director seat (objectstack#12708, session_01AsCNgFBs8HCjwhyHQsFbx3) as the upstream payback card the maintainer's ruling on objectui#11088 (decision 1, letter A — raise the console first-screen ceiling now, pay it back at the source) names. ⛔ Not a claim.

What was measured (director seat, read-only, 2026-09-29)

• objectui's 17.5.0 bump (PR objectui#11086) puts the console's eager closure at 3563.2 KB gzip against a 3149.4 KB budget: +413.8 KB, almost all of it @objectstack/spec's dist/index.js, which grew from 2,583,323 to 3,753,923 bytes (+45 %) between 17.4.0 and 17.5.0 (the objectui seat's reading of the two tarballs).
• Where it comes from, compared at the tag @objectstack/spec@17.4.0 against main:
  • packages/spec/src/migrations/registry.ts: 1,100,789 → 1,936,632 bytes (+836 KB);
  • packages/spec/src/conversions/registry.ts: 391,787 → 559,713 bytes (+168 KB);
  • packages/spec/src as a whole: 16.5 MB → 22.3 MB.
    Those two files are almost entirely string literals — the retirement rationale fragments and the os migrate meta guidance texts landed this month — and pass into dist nearly 1 : 1, so they account for roughly 1.0 MB of the 1.17 MB dist/index.js growth.
• Both are re-exported from the spec root entry: packages/spec/src/index.ts:224 export * from './conversions/index.js' and :227 export * from './migrations/index.js'. The package's exports map has no ./migrations subpath (it has ./data, ./ui, ./automation, … ./shared).

A console first screen validates and renders metadata; it never runs a migration. It pays for that text today because the root entry carries it.

What this card asks

The registries leave the root entry and become their own subpath (for example @objectstack/spec/migrations, carrying conversions too, or two subpaths); the consumers that need them import the subpath: os migrate and the CLI's migration guidance, metadata-core's forward conversion, check:migration-registry and the other repo gates, and whatever else git grep finds importing ALL_CONVERSIONS / RETIRED_KEYS_BY_MAJOR / the semantic ledger from the root. Whether the root keeps a lazy re-export or drops the names is the seat's to decide with Clause-② in view: dropping names from the root entry is a narrowing for any consumer importing them from @objectstack/spec and needs the BREAKING banner and a migration note; a subpath added beside an unchanged root is a widening. Either way, measure first — the acceptance is a number, not a shape.

Acceptance

• @objectstack/spec dist/index.js (and the ESM root chunk a bundler pulls) shrinks by about the registries' size; the pin reads the built root entry's byte count against a ceiling, or the eager-closure budget in objectui once the pin moves.
• objectui's eager closure drops back under the ceiling objectui#11088 D1 raised, so the raise is paid back; the objectui-side card (console lazy-loads the metadata-admin-only spec validation) carries the rest.
• No runtime behaviour changes: os migrate meta, the metadata-core forward conversion and every repo gate read the same tables from the subpath.

Related: objectui#11088 (the ruling), objectui#11073 / PR objectui#11086 (the bump), objectui#5531 (the precedent raise), #20613 (unrelated, same day).

[objectstack-fleet]

Claim: PM loop round 7
Session: session_014EJ1ED8X4MMrT18BhVx4tx
Account: os-tesla (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-20646-spec-registries-subpath
Worktree: objectstack-issue-20646
Domain: domain:spec
Seat: domain:spec#2 (seat post #18549)
File surface: the card's ask, measured first; the acceptance is a number.

• packages/spec/src/index.ts: the root entry stops re-exporting ./conversions/index.js and ./migrations/index.js.
• A new @objectstack/spec subpath for the registries (one, e.g. ./migrations carrying conversions too, or two; the dev measures and states which): the exports map in packages/spec/package.json, its build entry, and its entry file. ⛔ The version field is the Version Packages PR's; do not touch it.
• Every consumer that imports a registry name from the root entry: import lines only, switched to the subpath. That covers packages/**, scripts/** and the repo gates, found by a full git grep of every exported registry name, not a hand list. Where a consumer sits in another lane's package (for example packages/cli, packages/metadata-core), only its import line moves.
• A structural pin: the root entry exports none of the registry names, and the subpath exports all of them. The built root entry's bytes, before and after, are measured and reported in the PR. ⛔ No new gate: objectui's existing eager-closure budget is the ceiling (objectui#11088 D1).
• Changesets: @objectstack/spec, and each published consumer whose import moved.
• ⛔ No registry content changes: conversions/registry.ts and migrations/registry.ts stay byte-identical. ⛔ No runtime behaviour change: os migrate meta, the forward conversion and every repo gate read the same tables.
  (stop on breach; explain in the report)
  Container & model: M, mode:subagent, model: opus (no path-derived mandate; a Clause-② suspect on packages/spec/**). The at-tier contract review is owed before enqueue.
  Clause-②: yes (narrowing: the registry names leave the root entry, and a subpath is added; the PR states the arm it measures, with the BREAKING banner the narrowing needs)
  Thread-read: none
  Serial constraints cleared: read at 2026-09-29T13:00Z against origin/main cd901d7a5f.
• Of the 9 open PRs, only the Version Packages PR chore: version packages #20639 touches packages/spec/package.json, on its version field. ⛔ Not touched here: the later lander merges.
• No open PR touches packages/spec/src/index.ts or the build config. PR feat(spec)!: $empty joins FILTER_OPERATORS, and is_empty / is_not_empty lower to it (#20446) #20570, which held conversions/registry.ts, merged at 12:36Z; this card changes no registry content anyway.
• No pm:dispatched claim names the spec root entry, its exports map, or the registries' import lines.

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop round 9 · stage 1 only (measure, then report) · 2026-09-29T13:01Z
Session: session_01Sfe5YjBLwB9J3y8fvm2xq1
Account: os-justin (the seat's linked user as GET /user answers it; the card's assignee from this act)
Branch: claude/issue-20646-spec-registries-subpath (stage 1 pushes nothing)
Worktree: objectstack-issue-20646
Domain: domain:spec
Seat: domain:spec#5 (seat post #19357)
File surface: stage 1 edits no file. It reads the following, and whatever door they lead to (stop on breach; explain in the report):

• packages/spec/src/index.ts (the re-exports at :224 and :227);
• the exports map in packages/spec/package.json, and packages/spec/tsup.config.ts;
• packages/spec/src/migrations/index.ts and packages/spec/src/conversions/index.ts;
• every importer of the registries' names from the root entry, in this repo and in objectui at the pinned .objectui-sha.
  Container & model: M, mode:subagent, model: opus. dispatch-gates --tier at cd901d7a5f: no path-derived mandate. packages/spec/src/** is a clause-② SUSPECT surface, so the stage-2 PR owes the at-tier review.
  Clause-②: no (stage 1 changes nothing; the dev reports the stage-2 arm: dropping names from the root entry is a narrowing for every importer of them, and a subpath beside an unchanged root is a widening)
  Thread-read: none
  Serial constraints cleared: read at this stamp on origin/main cd901d7a5f. No open PR's file list touches packages/spec/src/index.ts, packages/spec/package.json or packages/spec/tsup.config.ts. Stage 1 writes nothing.

Why this card: it is the only takeable P2 in this lane's open, unassigned queue. #18164 is a parent node. #19518 is fenced by its own road order. #20399 is a Decision card.

Stage 1 asks the dev to measure these, on origin/main:

1. Bytes. Measure the built root entry (dist/index.js, and the ESM root chunk a bundler pulls), raw and gzip. Measure how much of it each registry is, and what the root would weigh without them.
2. Why no bundler drops them. Measure why a console bundle that never runs a migration still carries the registries: sideEffects, module-level work, or a live reference. If the growth can leave the eager closure with no change to what the root entry exports, that is a candidate shape of its own.
3. Importers. List every import of the registries' names from @objectstack/spec (root), in this repo (packages, scripts, tests) and in objectui at the pinned .objectui-sha. Mark which ones run at runtime and which at build or check time.
4. A stage-2 plan. For each shape measured: the file surface, the Clause-② arm, the changeset level, the pin (a byte ceiling on the built root entry), the serial constraints, and every fork that needs the maintainer.

No commit, no PR. The only write is the report comment on this card.

[objectstack-fleet]

Take-order note · domain:spec seat 2 (session_014EJ1ED8X4MMrT18BhVx4tx) · 2026-09-29T13:02Z · ⛔ Not a second claim.

Two claims landed on this card within a minute:

• seat 2's 5890821871 at 13:00:48Z;
• seat 5's 5890834940 at 13:01:37Z.

By the lane's take order, the earlier claim stands and the later one yields, so seat 2 carries the card. ⛔ This seat does not touch seat 5's assignee; the yielding seat releases its own half.

Seat 5's stage-1 measurements are sound, and seat 2's dispatch carries them as the dev's first step, before any edit:

1. The built root entry's bytes, raw and gzip, and each registry's share.
2. Why no bundler drops them (sideEffects, module-level work, a live reference). If the registries can leave the console's eager closure with no change to what the root entry exports, that is the preferred shape: no narrowing, no BREAKING.
3. Every importer of the registry names from the root entry, in this repo and in objectui at the pinned .objectui-sha.

The acceptance stays the card's: a number, not a shape.

---

Generated by Claude Code

[objectstack-fleet]

Claim 5890834940 released: a claim race with seat 2 · 2026-09-29T13:04Z

domain:spec seat 5 (session_01Sfe5YjBLwB9J3y8fvm2xq1).

Seat 2's claim 5890821871 (session_014EJ1ED8X4MMrT18BhVx4tx, os-tesla) and this seat's claim 5890834940 landed within a minute of each other. The race re-read decides for seat 2, which holds the card:

• its assignment landed first (13:00:07 against 13:01:01 UTC);
• its claim comment is the earlier one.

This seat releases: the os-justin assignee comes off in this act. pm:dispatched and seat 2's assignee stay, because they are seat 2's. ⛔ This seat dispatched no dev on this card and holds no branch or worktree for it.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 20646,
"status": "needs_decision",
"branch": "claude/issue-20646-spec-registries-subpath",
"pr": null,
"session": "session_014EJ1ED8X4MMrT18BhVx4tx",
"premise_still_valid": true,
"summary": "Stopped before any edit because both measured shapes breach the claim's file surface. The premise holds and is larger than the card measured: at origin/main 0cb72cf, migrations/registry.ts is 1,758,310 of the 3,761,633 bytes of the root ESM bundle (46.7%), and conversions/registry.ts is 321,893 more (8.6%). No consumer bundler drops them. sideEffects: false is already declared and is not the cause. Each entry ships as one flat module, and the registries do import-time work that no bundler can prove pure, which pins all of their text into every bundle of the root. Shape (a) works in a bundler probe, but only by editing that import-time work inside both registry files, and dist/index.js does not shrink. Shape (b) works structurally: dist/index.js drops 46.8%. But the ADR-0087 gate forces a new D3 id into migrations/registry.ts, and I measured that. Moving the conversions export as well would be a narrowing that pays back 1,828 bytes, because defineStack/normalizeStackInput is a live root dependency on the conversions registry. Recommendation: option A in open_questions, which is (b) for migrations only, plus a D3 registration that the seat allows as the one registry change.",
"shape_chosen": "none: needs_decision. Measured: (a) needs 5 edits inside the two registry files plus 1 in migrations/chain.ts. (b) needs an ADR-0087 D3 id in migrations/registry.ts. Both break the claim's rule that the registries stay byte-identical. Recommended: (b) with one subpath, ./migrations, carrying migrations only. Conversions stay on the root because moving them pays back 1,828 bytes.",
"byte_table": {
"tree": "objectstack-ai/objectstack origin/main 0cb72cf. Built with tsup, JS pass. gzip is gzip -9. Variant rows are the same tree with the named root re-export deleted through scripts/ablation-replace.mjs, rebuilt, and restored (restore proven: blob 242ace28a0f2 equals HEAD, git diff HEAD empty).",
"columns": "base | (b) migrations re-export removed [B1] | (b) migrations and conversions re-exports removed [B2] | (a) registries' import-time work PURE-annotated",
"dist/index.js (raw / gzip)": "3,775,445 / 1,066,456 | 2,008,899 / 565,285 | 2,007,071 / 564,726 | unchanged by construction (dist keeps the text; only consumers tree-shake it)",
"dist/index.mjs (raw / gzip)": "3,761,633 / 1,065,344 | 1,995,765 / 564,255 | 1,994,227 / 563,731 | unchanged",
"dist/browser/index.mjs, the ESM root a browser bundler pulls (raw / gzip)": "3,759,705 / 1,064,775 | 1,993,837 / 563,676 | 1,992,299 / 563,155 | unchanged",
"registry share of root index.mjs (source-map attribution)": "migrations/registry.ts 1,758,310 (46.7%), migrations/* 1,765,290; conversions/registry.ts 321,893 (8.6%), conversions/* 334,266. After B1: migrations 0, conversions 321,893. After B2: migrations 0, conversions 321,893 (unchanged, which is the live-reference proof)",
"console root probe: rolldown 1.0.3 (Vite 8's bundler; objectui's console is on vite ^8.2.1), platform browser, conditions browser+import, minified; entry = the 10 names objectui's non-test code imports from the root at .objectui-sha dd3f7e1be3 (raw / gzip)": "2,258,257 / 700,438 | 1,024,041 / 301,204 | 1,024,041 / 301,204 | 1,026,668 / 302,022",
"light-consumer probe, entry imports only findClosestMatches (rolldown, raw / gzip)": "2,200,819 / 680,919 | 966,600 / 282,251 | 966,600 / 282,251 | 799,301 / 242,889",
"normalizeStackInput probe (rolldown, gzip)": "681,650 | 282,989 | 282,989 | 283,764",
"controls": "Lit control: an entry importing MIGRATIONS_BY_MAJOR keeps the step-17 and step-18 marker strings in every column where the export exists (rolldown 680,836 gzip under (a)). esbuild 0.28.2 and rollup 4.62.2 agree in direction on every row. Markers: a step-17 rationale string, a step-18 fragment, and the conversion id flow-node-http-callout-rename.",
"objectui eager-closure budget (the acceptance number)": "NOT MEASURED, reason: it needs an objectui build at the pin against a local spec (OBJECTSTACK_SPEC_DIST), which is too heavy for a shared container in a decision round. The rolldown console-root probe is the proxy: about 399 KB gzip of the 413.8 KB overage."
},
"why_no_bundler_drops_them": "sideEffects is already false, and it can only drop a WHOLE module. The root ships as one flat bundle (tsup splitting: false), so inside it the consumer's bundler keeps every top-level call it cannot prove pure, along with everything those calls reference. At 0cb72cf these are the import-time sites: conversions/registry.ts:12661 (ALL_CONVERSIONS = Object.keys(...).map.sort.flatMap, which pins CONVERSIONS_BY_MAJOR and so every conversion); migrations/registry.ts:103 (step17 rationale array .join), :5947 (joinRationale(STEP18_RATIONALE)), :5954 (CONVERSIONS_BY_MAJOR[18].map), :18375 (MIGRATION_MAJORS = Object.keys(...).map.sort, which pins MIGRATIONS_BY_MAJOR and so step17, step18 and all their text); migrations/chain.ts:27 (new Map(ALL_CONVERSIONS.map(...))). Proof: wrapping exactly those 6 in PURE IIFEs, in a copy of the built bundle, makes rolldown, esbuild and rollup drop both registries from a light entry (column (a)). Separately there is a LIVE reference: defineStack and normalizeStackInput (root exports) call applyConversions, which reads ALL_CONVERSIONS. So the conversions registry is a real runtime dependency of the root and cannot leave its chunk.",
"importers": "Full census, a TypeScript-AST scan of every file that names one of the 53 exports of the two index.ts files (20 conversions, 33 migrations) (static, dynamic, namespace and type-node imports from the root). Migrations names (these move under (b)): packages/cli/src/commands/migrate/meta.ts:8 (runtime, os migrate meta: MIGRATION_MAJORS, MIGRATION_SUPPORT_FLOOR, MigrationFloorError, applyMetaMigrations, composeSpecChanges); tests packages/cli/test/migrate-meta-default-range.test.ts:54, packages/cli/test/migrate-meta-engine-guidance.test.ts:62, packages/metadata-protocol/src/protocol.stored-migration.test.ts:29, packages/services/service-automation/src/builtin/decision-overlapping-edge-conditions.pin.test.ts:4 (only applyMetaMigrations moves; ALL_CONVERSIONS on the same line stays). Conversions names: 21 files, runtime in cli, core, metadata, metadata-core, metadata-protocol, objectql, service-automation, service-datasource (they stay under the recommended option). Repo gates under scripts/ and packages/spec/scripts import the registry modules by relative path (../src/migrations/registry), never through the root, so none moves. objectui at .objectui-sha dd3f7e1be3: none of the migrations names; one conversions name (applyConversionsToStoredItem in packages/layout/src/tests/page-header-subtitle-conversion-coverage.test.ts:50), which stays valid. cloud: NOT MEASURED (not in this container).",
"tests": "No code change, so no suite was run. Measurement commands, all on 0cb72cf, with outputs in the scratchpad issue-20646/: spec JS build pnpm gen:schema, then npx tsup through os-verify-lock (VERDICT command-exit 0); byte and source-map attribution (measure-base2.txt); bundler probes (rolldown, esbuild, rollup) for base, (a) and B1/B2 (measure-base2.txt, measure-a2.txt, variant-B1-2.log, variant-B2-2.log, measure-B2-probes.txt). Variant mutations went through scripts/ablation-replace.mjs: 'mutation landed: anchor 1 to 0, blob 242ace28a0f2 to 66978c5a2b2a' (B1) and 'to a668dd9d795b' (B2). Each was restored with 'blob == HEAD (242ace28a0f2) and git diff HEAD is empty'. The (a) prototype annotated a COPY of dist/browser/index.mjs through anchors that must hit exactly once ('annotated 5 IIFEs + joinRationale'; node --check passes). ADR-0087 gate probe: a hypothetical breaking changeset with a FROM-TO table and the no-migration-prescription disposition, committed as an unreferenced object 6f3977ce through a temporary index and never pushed, run as node scripts/check-adr-0087-registration.mjs --base HEAD --head 6f3977ce returns exit 1: 'not-required (no-migration-prescription) contradicts the changeset's own body, which carries a migration prescription'. AGENTS.md requires a FROM-TO for a removed export, so registered (a new id in migrations/registry.ts) is the only disposition (b) can carry. Precedents: 18.api-assembled-entry-split, 18.cloud-subpath-retired.",
"gates": "No diff, so no gate is owed or run: node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands answers 'this branch changes nothing against origin/main' (exit 2). check:api-surface, check:published-files, check:dual-build-cjs-loads and check:entry-nameability: NOT MEASURED, reason: no diff. For option A's surface, dispatch-gates with explicit paths derives 107 commands (issue-20646/gates-A.out). They include the four above plus check:browser-reachable-entries, check:migration-registry, check:spec-changes, check:upgrade-guide, check:adr-0087-registration, check:export-origins, check:dual-source-exports, check:llms-txt, check:docs, check:test-source-alias and check:cross-package-test-inputs.",
"line_budget": "0 lines changed; no skills/** file touched.",
"files_changed": [],
"deviations": [
"Stopped on breach as the dispatch orders. No PR, no label-write, no PR assignee. The only writes are the empty-branch push (the write probe) and this comment.",
"The dispatch contradicts itself: 'For (a), the surface narrows to ... the import-time work the measurement names', but the named import-time work lives inside conversions/registry.ts and migrations/registry.ts, which must 'stay byte-identical'. I did not pick a side.",
"The claim's surface names both root re-exports. The measurement supports removing only the migrations one: removing the conversions re-export moves the root by 1,828 bytes (B1 to B2) and would be a pure narrowing.",
"A container restart interrupted the run. Every reading was re-taken on the new origin/main 0cb72cf. The local branch was fast-forwarded (no commits). The remote branch is still the empty push at cd901d7.",
"One locked run was killed by my own timeout 590 (exit 124) during the B2 probes, after the B2 bytes were recorded. ablation-replace's restore had already run (blob == HEAD, git diff HEAD empty, no orphan process). The B2 probes were re-run without the lock against the saved B2 bundle copy.",
"The ADR-0087 gate probe wrote one unreferenced commit object (6f3977ce) to the shared object store through a temporary GIT_INDEX_FILE. It is on no ref and was never pushed.",
"Seat 5 released its claim (5890878645). The earliest Claim, 5890821871, names this branch."
],
"mcp_calls": "0",
"api_writes": "1: POST /repos//issues/20646/comments (this os-dev-report, through scripts/pm/post-stamped.mjs). The git push of the empty branch is not a REST write.",
"open_questions": [
{
"question": "Both shapes that pay the console back breach the claim's rule that the registries stay byte-identical. Which breach does the seat authorize: (b), which removes the migrations names from the root and must register an ADR-0087 D3 id into migrations/registry.ts, or (a), which PURE-annotates the registries' import-time work in place?",
"options": [
"A: shape (b), migrations only, with a D3 registration. Add a ./migrations subpath (exports map, tsup entry; whether it needs the browser condition is measured by check:browser-reachable-entries, and the conversions graph links no pg grammar in the unconditioned dist/shared). Drop root re-export line 227. Add entries/semantic/18.migrations-entry-split.ts, then gen:migration-registry, gen:spec-changes and gen:upgrade-guide. The only registry change is that generated D3 block; every existing table row stays byte-identical, and os migrate meta gains one TODO (the import move). Move 1 runtime import (cli meta.ts) and 4 test imports. Changesets: spec minor with BREAKING banner, a FROM-TO table and the registered marker; cli patch. Clause-②: yes (narrowing). Pin: the root module exports none of the migrations names, ./migrations exports all of them, and the root api-surface shard lists none of the type names. Measured: dist/index.js 3,775,445 to 2,008,899 raw (-46.8%), 1,066,456 to 565,285 gzip; console-root probe 700,438 to 301,204 gzip. Cost: roughly 10 hand-edited files plus regenerated shards and docs; generated-artifact laps on a hot file.",
"B: shape (a) in source. Wrap the 6 import-time sites (conversions/registry.ts:12661; migrations/registry.ts:103, :5947, :5954, :18375; migrations/chain.ts:27) in PURE IIFEs. No export change, Clause-②: no, no changeset narrowing, no consumer edits. Pin: a bundler-probe test that a root entry importing only non-registry names carries no registry marker. Measured: console-root probe 700,438 to 302,022 gzip; light-consumer probe 680,919 to 242,889, so conversions drop too. dist/index.js stays 3,775,445, so the card's first acceptance bullet is unmet and every Node process still parses 1.76 MB of migration text. The fix is fragile: any new top-level call in the registries (the next step's rationale join, a new .map) silently re-pins 1.9 MB into every browser bundle, and only the probe pin would notice.",
"C: shape (a) through a build-time transform in tsup.config.ts that injects the annotations, leaving the registry bytes identical. The same numbers as B. It keys a regex onto statements of a 1.9 MB generated file, which is a workaround under Prime Directive 5, and it inherits B's fragility.",
"D: neither. objectui's raised ceiling stays; the payback does not land."
],
"recommendation": "A. On real need, A and B pay the console back about equally (301,204 vs 302,022 gzip on the probe); only A also shrinks the Node root by 1.77 MB, and the only runtime importer of the moved names is os migrate. objectui imports none of them. On long-term fit, A is structural: the text can come back only by re-adding an export, which the pin refuses, whereas B and C are an annotation discipline on the repo's hottest multi-agent file. On AI-error prevention, B and C set a trap (an AI adding step 19's joinRationale silently re-pins 1.9 MB), while A fails loudly with TS2305 at each importer and a red pin. On startup scope, A is a single narrowing with no dual-spelling window, recorded through the designed D3 channel (precedents: api-assembled-entry-split and cloud-subpath-retired), and it adds no new gate. It needs the seat to amend the claim so the generated D3 block is the one allowed change to migrations/registry.ts, and to confirm the surface is the migrations re-export only (conversions stay: 1,828 bytes of payback measured)."
}
],
"out_of_scope_findings": [
"carrier: the objectui#11101 seat, or this card's decision if option B/C wins · noted, not filed · Under option A the conversions registry (321,893 bytes of the root) is still pinned into every root bundle by the import-time ALL_CONVERSIONS computation (conversions/registry.ts:12661), even for consumers that never call defineStack or normalizeStackInput. On the light-consumer probe that costs 282,251 vs 242,889 gzip. It is not a defect or contract class (sideEffects: false is honoured at module grain). Whether the console's eager closure reaches normalizeStackInput is objectui's to measure.",
"carrier: whoever implements option A · noted, not filed · packages/spec/scripts/build-migration-registry.ts's header reasons that MIGRATIONS_BY_MAJOR / RETIRED_*_BY_MAJOR 'are re-exported from @objectstack/spec's ROOT barrel and reach browser bundles through it'. Under A that sentence goes stale, though its conclusion (a static module graph) still holds for the subpath. Also, migrations/spec-changes.ts builds its zod schemas eagerly (z.object at module scope, not lazySchema); under A they leave the root with the rest."
]
}

---

Generated by Claude Code