@kudzujs/core 0.16.25 → 0.16.26

package/docs/next-architecture/application-capability-release-plan.md

@@ -1,2445 +0,0 @@
-# Application Capability Release Plan
-
-This is the authoritative execution plan from `0.10.0` through `1.0.0`. It
-replaces the provisional `0.10` normalize, `0.11` diagnostics, and `0.12`
-adapter ordering recorded in the completed 0.9 packet. Those tools remain in
-scope, but application capability evidence now determines their order.
-
-The plan assigns one application-capability section to each minor release and
-one independently releasable evidence packet to each patch release. Planned
-versions are not completed releases. `package.json`, release notes, tags, and
-registry metadata change only after the matching packet passes every gate.
-
-## Product Outcome
-
-Kudzu must support complete modern application journeys with no React runtime,
-VDOM, hydration, retained browser component tree, generic component renderer,
-or hook dispatcher. Static routes may still emit zero JavaScript. Interactive
-routes may emit every route- or feature-specific capability required for
-correct behavior.
-
-The target is zero unused runtime, not zero runtime.
-
-## Version Map
-
-| Minor | Capability section | Required outcome |
-|---|---|---|
-| `0.10.x` | Local and shared state | The greenfield application shell and state ownership baseline are executable. |
-| `0.11.x` | Async and server data | Loading, errors, stale work, mutation, and evidence-backed shared coordination are complete. |
-| `0.12.x` | Routing and application lifetime | Authenticated route, layout, history, focus, and failure journeys are complete. |
-| `0.13.x` | Forms | Production-shaped validation, drafts, conditional fields, autosave, and upload boundaries are complete. |
-| `0.14.x` | Lists, tables, and virtualization | Large keyed data behavior and a measured bounded-DOM decision are complete. |
-| `0.15.x` | Overlay and layer ownership | Accessible dialog, popover/menu, toast, and any proven layer ownership are complete. |
-| `0.16.x` | External complex UI | Real editor, chart/map, or grid packages use bounded mount/update/dispose ownership. |
-| `0.17.x` | Long-lived applications | Realtime, repeated navigation, cleanup, and long-running memory gates pass. |
-| `0.18.x` | Lazy loading and code splitting | Large feature dependencies load only when their owner requires them. |
-| `0.19.x` | React ecosystem migration | Apache Answer and Memos pass connected browser journeys through the compatibility boundary. |
-| `0.20.x` | AI tooling and delivery cost | Structured tools and a fair tracked AI benchmark reduce cost per successful task. |
-| `0.21.x` | Production-scale proof | Functional parity, build scale, browser performance, memory, and release evidence pass together. |
-| `1.0.0` | Stable application model | Semantic IR v1 is stable and the maintained AI delivery gate passes. |
-
-## Execution Rules
-
-1. Work on exactly one patch packet at a time. Start the next packet only after
-   the previous packet is released or this document explicitly replans it.
-2. Begin behavior work with a real failing greenfield or React-derived fixture.
-3. Reduce the failure through native platform behavior, existing Kudzu
-   semantics, normalization, or an internal adapter, in that order.
-4. Two unrelated real fixtures register a repeated application limitation.
-   Three unrelated real fixtures are required before architecture review may
-   authorize a new semantic primitive.
-5. A patch may complete without a new primitive, compiler pass, public API, or
-   runtime when existing composition satisfies its acceptance contract.
-6. Package-specific knowledge ends in compatibility analysis, normalization,
-   or an internal adapter. It must not enter generic IR, codegen, or runtime.
-7. Every static sibling remains JavaScript-free. Every interactive route emits
-   only the capabilities selected by its validated route record.
-8. Every patch records semantic primitives, core compiler/pass LOC, runtime
-   concepts, browser bytes, fixtures, output hashes, and benchmark deltas.
-9. Do not publish a planned version merely because its source compiles. The
-   exact release commit must pass behavior, accessibility, browser, output,
-   package, and applicable performance gates.
-10. Never silently combine patch packets. Revise this plan first when an
-    inseparable producer/consumer boundary is proven.
-
-## Shared Testbeds
-
-### Greenfield Project Application
-
-The maintained greenfield testbed is a project-management SaaS with this route
-shape:
-
-```text
-/login
-/app/projects
-/app/projects/:projectId
-/app/issues
-/app/issues/:issueId
-/app/settings
-/help
-```
-
-It grows one accepted user journey at a time. `/help` is the static zero-JS
-control. The application uses a deterministic local HTTP server until a packet
-explicitly requires a different transport.
-
-### Apache Answer
-
-The connected acceptance target is:
-
-```text
-login
--> question list
--> search/filter/pagination
--> detail
--> create/edit
--> server validation/error
--> mutation and list refresh
--> logout
-```
-
-### Memos
-
-The connected acceptance target is:
-
-```text
-login
--> feed
--> pagination
--> create memo
--> reaction
--> realtime refresh
--> logout
-```
-
-Actual Budget remains deferred until a complete pinned workspace and executable
-dependency graph are available.
-
-## `0.10.x`: Local And Shared State
-
-### `0.10.0`: Application Contract And State Baseline
-
-- **Purpose:** create the durable greenfield testbed and capability inventory.
-- **Current limitation:** state, forms, lists, and navigation are proven mostly
-  in isolated fixtures rather than one maintained application.
-- **Expected files:** a dedicated project-application fixture, browser
-  acceptance entry, package script, and capability manifest; no production
-  compiler or runtime file should change.
-- **Acceptance:** `/app/projects` renders deterministic project state, one
-  interaction updates direct DOM, `/help` emits zero JavaScript, required
-  Chrome runs in CI, and route/artifact baselines are recorded.
-- **AI effect:** one application and one manifest replace repeated repository
-  exploration when selecting later capability work.
-- **Stop condition:** the fixture becomes a syntax showcase instead of a user
-  journey, or requires a new primitive before the baseline exists.
-- **Done condition:** every later packet can add behavior to the same testbed
-  and reuse its framework-neutral acceptance harness.
-
-**Completion evidence:** the tracked project application emits `/login`,
-`/app/projects`, and `/help`. The project route reuses one existing primitive
-SignalIR state and one command handler; it adds no production compiler LOC,
-pass, semantic primitive, Handler ESM, or runtime concept. Its existing command
-runtime graph is 768 raw / 441 aggregate gzip bytes, while `/help` emits 0 B
-JavaScript. The complete fixture emits 4 files, 1,970 raw / 1,265 aggregate gzip
-bytes, with deploy digest
-`94ce7d01cac2e80823a36279294015d2f22328acaed571abb9996f8ca6171121`.
-The focused Chrome journey, `npm run check`, 273 non-browser tests, 273
-required-Chrome tests, and package smoke pass. No production performance claim
-is made because compiler and runtime sources are unchanged.
-
-### `0.10.1`: Local, Object, Array, And Derived State Scale
-
-- **Purpose:** prove stable dependency and ownership behavior with many ordinary
-  states and nested JSON-safe application records.
-- **Expected boundary:** existing SignalIR, DerivedIR, bindings, conditions, and
-  keyed ownership; no normalized-state runtime.
-- **Acceptance:** deterministic updates across project summary, filters, issue
-  counts, derived labels, conditional owners, and nested lists; unrelated DOM
-  does not change; removal releases owned state.
-- **Performance:** record commit latency and runtime metadata for increasing
-  state/dependency counts.
-- **Stop condition:** proposed support depends on arbitrary object mutation or a
-  generic rerender loop.
-- **Done condition:** the recorded scale remains within the material regression
-  threshold and nearby unsupported dynamic paths fail at authored locations.
-
-**Completion evidence:** the project journey composes five ordinary route
-states: one object, two arrays, and two primitives. Six existing reactive
-bindings, one conditional owner, three keyed lists, one nested issue list, and
-one keyed row-local state cover summary replacement, filtering, derived issue
-counts, saved-filter append, conditional release/remount, keyed identity, and
-row-state release/reset. An unrelated control records zero DOM mutations. The
-route adds no semantic primitive, pass, browser runtime concept, normalized
-state, or generic rerender path; the only supported source adjustment is the
-application's use of a separate project array state at the existing keyed-list
-boundary. Direct object-field nested-list replacement remains outside this
-packet after its isolated experiment failed browser replacement, while the
-existing dynamic object-property diagnostic remains source-located.
-
-The application deploy contains 12 files, 47,546 raw / 16,285 aggregate gzip
-bytes, with SHA-256
-`d50c9ff6dccba338559dd9e9141c4607ec62da30385658e617a86f7af3f52e0f`.
-`/app/projects` uses 39,178 raw / 14,143 aggregate gzip JavaScript bytes;
-`/help` remains 0 B. The generated 1/8/32-state scale benchmark emits identical
-deploy output under `v0.10.0` and the candidate. Seven fresh Chrome processes
-record commit medians of 0.5/0.6/0.9 ms as the derived binding dependency edges
-increase from 0 to 8 to 32. Runtime JavaScript is 768/10,935/11,341 raw bytes
-and 441/4,789/4,841 aggregate gzip bytes. No material regression is present.
-Release commit `f04041b`, tag `v0.10.1`, both CI jobs, the GitHub release,
-`@kudzujs/core@0.10.1`, `create-kudzu@0.1.105`, registry metadata, and a fresh
-scaffold install/check are verified.
-
-### `0.10.2`: Shared Layout And Cross-Route State
-
-- **Purpose:** keep workspace/session state across an explicit application
-  navigation group while resetting route-owned state.
-- **Expected boundary:** existing SharedStateIR, layout state IDs, route release,
-  and native fallback.
-- **Acceptance:** workspace selection survives project/list/detail transitions;
-  route drafts reset; reload behavior is explicit; a separate route group does
-  not receive the shared state.
-- **Stop condition:** implementation requires a document-global store or SPA
-  router.
-- **Done condition:** state lifetime and disposal are exact in browser tests and
-  static routes receive no shared-state runtime.
-
-**Completion evidence:** the greenfield application adds one exact Alpha detail
-route and one explicit navigation group around list/detail routes. A relative
-Context hook lowers through the existing `SharedStateIR`: both routes own the
-same `workspace` layout state while list state and the detail `draft` remain
-route-owned. Required Chrome proves retained layout DOM identity, workspace
-persistence in both directions, detail DOM release, fresh draft state on
-re-entry, and initial layout/draft values in a fresh document. `/help` remains
-outside the group, retains its native anchor path, and emits complete HTML with
-0 B JavaScript.
-
-This packet adds no semantic primitive, ModuleIR kind, compiler pass, production
-compiler LOC, runtime concept, normalization rule, adapter rule, store, or SPA
-router. It adds one application layout/context module, one route, one config,
-and one positive browser journey; existing navigation-group diagnostics retain
-the nearby rejected configuration boundary. The fixture emits 14 files, 57,293
-raw / 19,985 aggregate gzip bytes, with deploy SHA-256
-`1c867ee7119f7eb23b935c4506c2a46639acb75c13f0916b29c255421b0d123d`.
-The list/detail session uses 10 unique JavaScript files totaling 46,418 raw /
-17,045 aggregate gzip bytes. Seven fresh Chrome profiles record list-to-detail
-completion at `[2.3, 6.5, 1.4, 1.3, 5.1, 1.3, 4.0]` ms, median 2.3 ms.
-
-Release commit `76a45c4`, tag `v0.10.2`, both CI jobs, the GitHub release,
-`@kudzujs/core@0.10.2`, `create-kudzu@0.1.106`, registry metadata, and a fresh
-registry-backed scaffold install/check passed on 2026-08-22. Publish workflow
-`32566976581` completed every version, package, publication, and registry gate.
-
-### `0.10.3`: Persistence Recipe And State Release
-
-- **Purpose:** prove URL, storage, and server-backed persistence without adding a
-  persistence framework.
-- **Expected boundary:** application-owned effects and Web Storage.
-- **Acceptance:** guarded restore, schema/version failure fallback, write after
-  state changes, logout clear, conditional/keyed release, and fresh remount.
-- **Stop condition:** a generic persistence API is proposed without another
-  independent application requirement.
-- **Done condition:** deterministic static fallback and browser recovery pass
-  with no storage runtime in unused routes.
-
-**Completion evidence:** the greenfield application layout now owns one guarded
-`localStorage` recipe composed entirely from ordinary state and two existing
-effects. Static HTML starts with `Primary`; a mount effect accepts only version
-1 records whose workspace is `Primary` or `Secondary`, while absent, malformed,
-wrong-schema, and wrong-version records fall back safely. A readiness state
-prevents the dependency effect from overwriting storage before restore.
-Workspace changes write the versioned record, and logout disables persistence,
-removes the record, and resets visible state in one synchronous handler.
-
-Required Chrome proves valid restore, every fallback, post-commit persistence,
-logout removal, conditional and keyed release/fresh remount, route draft
-release/fresh revisit, and a fresh direct document. `/help` remains outside the
-navigation group with complete HTML and 0 B JavaScript. The fixture emits 19
-files, 65,376 raw / 23,471 aggregate gzip bytes, with deploy SHA-256
-`bb58826876bdf090a85d2c1f8a1edcbbc9abd35d5335eb3d1e7fb1fafe6bed72`.
-The two-route session uses 15 unique JavaScript files totaling 53,562 raw /
-20,323 aggregate gzip bytes.
-
-This packet adds zero semantic primitives, ModuleIR kinds, compiler passes,
-production compiler lines, runtime concepts, normalization rules, adapter
-rules, public APIs, stores, or persistence runtimes. One existing positive
-application fixture gains valid, absent, malformed, wrong-schema, wrong-version,
-write, and clear browser cases. A matched 21-profile measurement records the
-authored capability cost: the 0.10.2 list/detail median of 1.9 ms and 46,418 raw /
-17,045 gzip session JavaScript becomes 1.5 ms and 53,562 / 20,323 bytes. The
-timing ranges overlap with no material loss; the 7,144 raw / 3,278 gzip byte
-increase is the explicit cost of route-owned effect/native entries, not a
-compiler or shared-runtime change.
-
-Release commit `4420ccc`, tag `v0.10.3`, both CI jobs, the GitHub release,
-`@kudzujs/core@0.10.3`, `create-kudzu@0.1.107`, registry metadata, and a fresh
-registry-backed scaffold install/check passed on 2026-08-22. Publish workflow
-`32568752477` completed every version, package, publication, and registry gate.
-
-## `0.11.x`: Async And Server Data
-
-### `0.11.0`: Owned Fetch Lifecycle
-
-- **Purpose:** establish the greenfield loading/error/data/refetch baseline.
-- **Expected boundary:** existing EffectIR, ordinary state, and native `fetch`.
-- **Acceptance:** delayed stale response loses state authority, HTTP failure is
-  accessible, explicit refetch recovers, route removal blocks late writes, and
-  authored cancellation is tested where required.
-- **Stop condition:** a query cache is added to implement one request.
-- **Done condition:** the project list journey passes and `/help` stays zero-JS.
-
-**Completion evidence:** the project list now composes one route-owned native
-`fetch` effect with ordinary request, status, error, project, and summary state.
-The deterministic server drives initial loading, a fast explicit refetch, an
-HTTP 500 alert, and recovery. A delayed prior continuation attempts its writes
-after dependency replacement and cannot replace the newer keyed project data.
-The effect owns an `AbortController`; dependency replacement and route removal
-invoke authored abort cleanup, and late route work cannot mutate the detail
-route or recreated list ownership. Existing project filtering, nested keyed
-identity, row state, shared layout state, persistence, and logout journeys still
-pass. `/help` remains complete HTML with 0 B JavaScript.
-
-This packet adds zero semantic primitives, ModuleIR kinds, compiler passes,
-production compiler lines, runtime concepts, normalization rules, adapter
-rules, public APIs, caches, Providers, retry schedulers, or query runtimes. One
-existing positive application fixture gains loading, success, stale completion,
-HTTP failure, recovery, cancellation, and route-release cases. The fixture
-still emits 19 files and now totals 67,886 raw / 24,077 aggregate gzip bytes,
-with deploy SHA-256
-`9c06d2a5d47b4ea2ae54cebbc0448bf161907691ee37a02a5dae0e7c37779ae4`.
-The project route uses 52,130 raw / 19,502 aggregate gzip JavaScript bytes; the
-two-route session uses 54,770 / 20,763 bytes, an authored capability increase of
-1,208 raw / 440 gzip bytes over `0.10.3`. Twenty-one fresh Chrome profiles record
-0.8 ms median list-to-detail completion with a 0.6/1.3 ms range. The changed
-fetch-completion prerequisite permits no latency improvement claim; no material
-regression is present.
-
-Clean-worktree `npm run check`, 273/273 standard tests, the required project
-Chrome journey, the focused required-Chrome retry for the one full-suite browser
-timeout, and packed-package smoke pass on 2026-08-22.
-
-Release commit `8065d86`, tag `v0.11.0`, both CI jobs, the GitHub release,
-`@kudzujs/core@0.11.0`, `create-kudzu@0.1.108`, registry metadata, and a fresh
-registry-backed scaffold install/check passed on 2026-08-22. Publish workflow
-`32571667242` completed every version, package, publication, and registry gate.
-
-### `0.11.1`: List And Detail Data Consistency Evidence
-
-- **Purpose:** expose the first real multi-consumer coordination limit.
-- **Expected boundary:** layout shared state plus existing effects first.
-- **Acceptance:** project list and detail observe one mutation, duplicate
-  requests are counted, route changes and refresh are deterministic, and all
-  current limitations are machine-recorded.
-- **Stop condition:** the fixture invents package-shaped query semantics not
-  required by the user journey.
-- **Done condition:** existing composition is accepted or the smallest repeated
-  missing coordination contract is isolated.
-
-**Completion evidence:** one layout-owned project record composes from two
-primitive shared signals and one existing mount effect. The list and detail
-routes observe the same server-confirmed rename without another read. The
-deterministic journey records one initial GET, zero additional GETs across two
-enhanced route changes, one mutation POST, and exactly one new GET after full
-document reload. The refreshed document reconstructs revision 1 from server
-truth while route-owned detail draft state resets. `/help` remains complete HTML
-with 0 B JavaScript.
-
-Current composition is accepted. Shared data lasts for one document and retained
-layout owner; independently authored fetch owners are not deduplicated, and no
-request registry, invalidation graph, subscriber set, or retained query cache
-exists. This one fixture does not authorize `0.11.2`: it needs no coordination
-primitive after ownership is placed correctly, while the packet requires two
-additional independent fixtures to demonstrate the same missing contract.
-
-This packet adds zero semantic primitives, ModuleIR kinds, compiler passes,
-production compiler lines, runtime concepts, normalization rules, adapter rules,
-public APIs, caches, Providers, or query runtimes. One existing positive fixture
-gains list/detail consistency, counted domain requests, mutation, navigation,
-and reload cases. It emits 20 files totaling 69,501 raw / 24,649 aggregate gzip
-bytes with deploy SHA-256
-`bc6335131b7867da2857621211919fad43ebfb9414aac711c45f0c8ed3803022`.
-The two-route session uses 16 JavaScript files totaling 55,736 raw / 21,208
-aggregate gzip bytes, an authored application increase of 966 raw / 445 gzip
-bytes over `0.11.0`. Twenty-one fresh Chrome profiles record a 0.9 ms median
-list-to-detail completion with a 0.7/1.2 ms range, overlapping the prior
-0.6/1.3 ms range; no latency change is claimed.
-
-Clean-worktree `npm run check`, 273/273 required-Chrome tests, and packed-package
-smoke passed on 2026-08-22. Release commit `5bab3fe`, tag `v0.11.1`, feature CI
-`32573961250`, the GitHub release, `@kudzujs/core@0.11.1`,
-`create-kudzu@0.1.109`, registry metadata, and a fresh registry-backed scaffold
-install/check passed. Publish workflow `32574182988` completed every version,
-package, publication, and registry gate.
-
-### `0.11.2`: Shared Request Coordination
-
-- **Purpose:** implement only the deduplication, subscriber, and invalidation
-  semantics authorized by `0.11.1` and two additional independent fixtures.
-- **Expected boundary:** package-neutral application semantics; no TanStack,
-  SWR, Provider, retry scheduler, or public query API by default.
-- **Acceptance:** one in-flight request per key, exact subscriber ownership,
-  invalidation after mutation, route/layout disposal, bounded retention, and
-  static exclusion.
-- **Stop condition:** three unrelated fixtures do not require the same semantic
-  primitive, or ordinary shared state remains sufficient.
-- **Done condition:** architecture review records the reused or new primitive,
-  runtime bytes, cache bound, and cleanup behavior.
-
-**Completion evidence:** architecture review covers three unrelated executable
-fixtures: the project application, TanStack Query-shaped product loading, and
-Apache Answer/SWR-shaped question loading. Every fixture has one explicit data
-owner, ordinary loading/error/data state, one native fetch effect, dependency or
-owner invalidation, exact cleanup, and static exclusion. None has two independent
-owners requesting one key, independently releasable subscribers, mutation-key
-invalidation, retained results after owner release, or a cache bound.
-
-The stop condition is met and this packet closes with no new primitive. The
-project application records `closed-no-new-primitive`, the three reviewed
-fixtures, an empty repeated-missing-contract list, and the reused semantics in
-its machine-readable capability contract. Existing fixture journeys remain the
-executable behavioral evidence. Absence of generic deduplication is not itself
-fixture pressure.
-
-This packet adds zero semantic primitives, ModuleIR kinds, compiler passes,
-production compiler lines, runtime concepts, normalization rules, adapter rules,
-public APIs, request registries, subscriber sets, invalidation graphs, caches, or
-browser bytes. One existing machine-readable contract and assertion change;
-all three existing positive fixtures remain behaviorally unchanged. The project
-application retains its 20-file, 69,501 raw / 24,649 aggregate gzip deploy and
-55,736 raw / 21,208 aggregate gzip two-route JavaScript session. Cache bound is
-not applicable because no cache exists; cleanup remains existing effect and
-route/layout owner release.
-
-Clean-worktree `npm run check`, 273/273 required-Chrome tests, and packed-package
-smoke passed on 2026-08-22. Release commit `0ae4d5d`, tag `v0.11.2`, CI
-`32575316981`, the GitHub release, `@kudzujs/core@0.11.2`,
-`create-kudzu@0.1.110`, registry metadata, and a fresh registry-backed scaffold
-install/check passed. Publish workflow `32575541567` completed every version,
-package, publication, and registry gate.
-
-### `0.11.3`: Mutation, Optimistic Update, And Rollback
-
-- **Purpose:** complete issue create/edit/delete behavior under failure.
-- **Expected boundary:** immediate state setters and existing shared data
-  coordination; a transaction primitive is evidence-gated.
-- **Acceptance:** optimistic list/detail update, server rejection, exact
-  rollback, duplicate-submit prevention, retained row identity, and accessible
-  error state.
-- **Stop condition:** generic transactions are proposed for a single-owner
-  handler.
-- **Done condition:** data integrity survives success, failure, navigation, and
-  retry.
-
-**Completion evidence:** one layout-owned async rename handler snapshots the
-current project name/revision, writes an immediate optimistic value, disables its
-button after the synchronous-turn commit, and sends one native POST. The
-deterministic server rejects the first attempt. The handler restores the exact
-snapshot, exposes `HTTP 500` through an accessible alert, and leaves the retained
-keyed row identity unchanged. A retry from detail writes the optimistic value,
-clears the alert, commits the confirmed revision, survives list/detail
-navigation, and restores from server truth after full document reload. Counts
-record two attempts, one accepted mutation, and no mutation-triggered GET.
-
-Existing immediate setters, native async handlers, shared layout signals, and
-owner authority are sufficient. This packet adds zero semantic primitives,
-ModuleIR kinds, compiler passes, production compiler lines, runtime concepts,
-normalization rules, adapter rules, public APIs, transaction primitives, caches,
-or schedulers. One positive application fixture gains optimistic success,
-failure, rollback, disabled duplicate submission, navigation, retry, and reload
-cases. It emits 19 files totaling 73,202 raw / 25,057 aggregate gzip bytes with
-deploy SHA-256
-`40c68fef4eeba5ea4c1a92e9a8003b74e6a45b7a10671ee2f3e4952010117b26`.
-The two-route session uses 15 JavaScript files totaling 56,318 raw / 21,211
-aggregate gzip bytes, an authored application increase of 582 raw / 3 gzip bytes
-over `0.11.2`. Twenty-one fresh Chrome profiles record a 0.8 ms median with a
-0.7/1.0 ms range; no latency change is claimed.
-
-Clean-worktree `npm run check`, 273/273 required-Chrome tests, and packed-package
-smoke passed on 2026-08-22. Release commit `ba6f97a`, tag `v0.11.3`, CI
-`32576861865`, the GitHub release, `@kudzujs/core@0.11.3`,
-`create-kudzu@0.1.111`, registry metadata, and a fresh registry-backed scaffold
-install/check passed. Publish workflow `32577090955` completed every version,
-package, publication, and registry gate.
-
-### `0.11.4`: Pagination, Refresh, And Polling Policy
-
-- **Purpose:** complete bounded server-data loading before infinite data work.
-- **Expected boundary:** URL signals, effects, native timers, and visibility
-  events.
-- **Acceptance:** page/filter synchronization, back/forward, refresh, optional
-  polling cleanup, no duplicate work, and bounded retained results.
-- **Stop condition:** background policy becomes a universal scheduler without
-  repeated evidence.
-- **Done condition:** the exact owner and network behavior are measured and
-  documented.
-
-**Completion evidence:** the project list reads page and filter from two
-existing React Router query signals and restarts its existing abortable fetch
-effect when either changes. Static updater handlers push page 2/all and page
-1/active URLs. Two browser back operations restore the prior URL, signal values,
-and bounded server results. An explicit refresh increments the same route-owned
-request signal. Optional polling owns one native interval and one
-`visibilitychange` listener; hidden events perform no work, one visible event
-performs one refresh, and disabling polling clears both owners before a later
-event. The deterministic journey records 11 total list requests across initial
-stale replacement, HTTP recovery, page/filter history, refresh, polling, and
-route remount. Every response retains at most two rows.
-
-Existing query signals, primitive state, dependency effects, abort invalidation,
-native timers/listeners, and route cleanup are sufficient. This packet adds zero
-semantic primitives, ModuleIR kinds, compiler passes, production compiler lines,
-runtime concepts, normalization rules, adapter rules, public APIs, caches,
-schedulers, or retained result stores. One positive application fixture gains
-page/filter URL ownership, history, refresh, visibility-aware polling, exact
-cleanup, request counts, and bounded-result cases. It emits 20 files totaling
-77,263 raw / 26,156 aggregate gzip bytes with deploy SHA-256
-`a3864a5155d658e86378fce80de94e082ef0894c5db5f5793d91b4de7f359b46`.
-The two-route session uses 16 JavaScript files totaling 58,970 raw / 22,124
-aggregate gzip bytes, an authored application increase of 2,652 raw / 913 gzip
-bytes over `0.11.3`. Twenty-one fresh Chrome profiles record a 1.3 ms median
-with a 1.2/1.6 ms range. The prior range does not overlap, so the 0.5 ms median
-increase is disclosed rather than presented as a no-regression result. `/help`
-remains 0 B JavaScript.
-
-## `0.12.x`: Routing And Application Lifetime
-
-### `0.12.0`: Project Route Shell And Runtime Parameters
-
-- **Purpose:** prove directly addressable project and issue routes.
-- **Expected boundary:** file routes, runtime parameters, complete HTML, and
-  native navigation.
-- **Acceptance:** direct entry, reload, invalid parameter rejection, project to
-  issue navigation, and standalone fallback.
-- **Stop condition:** an SPA route registry is introduced.
-- **Done condition:** all target routes are addressable without a browser router.
-
-**Completion evidence:** the maintained project application adds one bracket
-project page and one nested bracket issue page, both using the existing aliased
-React Router `useParams()` migration path and the existing shared application
-layout. Each fallback emits complete document structure plus only pathname
-parameter, binding, native-handler, and owned-layout capability ESM. The project
-page owns a native parameterized anchor to its first issue. The issue page owns a
-native parameterized project return. Neither standalone artifact includes the
-enhanced-navigation entry.
-
-The fixture emits the compiler-provided ordered rewrite records through its
-existing `afterBuild` host boundary. A deterministic static server checks exact
-files first, then applies those records for direct project and issue entry.
-Required Chrome proves project entry,
-issue entry, and issue reload; the generated parameter module rejects an encoded
-slash before mounting. Static HTML and artifact assertions prove the project to
-issue anchor remains native, both routes share one signature-keyed standalone
-runtime family, and `/help` remains 0 B JavaScript. Deployment still must install
-the generated rewrites; Kudzu does not add a server or browser route registry.
-
-Existing file routes, runtime pathname parameters, layout ownership, route
-bindings, native anchors, host rewrites, and browser reload are sufficient. This
-packet adds zero semantic primitives, ModuleIR kinds, compiler passes,
-production compiler lines, runtime concepts, normalization rules, adapter rules,
-or public APIs. One positive application fixture gains two route files, direct
-entry, reload, invalid-parameter rejection, native project-to-issue linkage, and
-host fallback cases. It emits 36 files totaling 105,577 raw / 38,067 aggregate
-gzip bytes with deploy SHA-256
-`6a6c5985cf2a7e9649b0bbd70506357051dac995a74225b0022c5f1ecedaf2b8`.
-The runtime project route uses 17,905 raw / 8,455 gzip JavaScript bytes and the
-issue route uses 18,002 raw / 8,474 gzip bytes. The existing two-route enhanced
-session remains byte-identical at 58,970 raw / 22,124 gzip bytes. Twenty-one
-fresh Chrome profiles record a 1.3 ms median with a 1.1/1.6 ms range; the range
-overlaps `0.11.4`, so no latency change is claimed.
-
-### `0.12.1`: Shared Layout, History, Focus, And Scroll
-
-- **Purpose:** complete the accepted same-document application group behavior.
-- **Expected boundary:** existing navigation runtime and layout/route lifetime.
-- **Acceptance:** back/forward, retained layout state, route cleanup, title/live
-  announcement, focus destination, hash behavior, and explicit scroll policy.
-- **Stop condition:** browser-native navigation is weakened for routes outside
-  the approved group.
-- **Done condition:** keyboard and browser-history journeys pass with native
-  recovery.
-
-**Completion evidence:** the project application adds route titles, one native
-hash anchor, and one intrinsic hash destination. Required Chrome activates the
-focused anchor, retains the layout and workspace state through push, back, and
-forward navigation, releases and recreates dirty route state, updates the title
-and polite live status, focuses hash and heading destinations, applies explicit
-hash and top scroll policies, and leaves `/help` browser-native outside the
-approved route group.
-
-The journey exposed one existing JavaScript value bug: an absent hash produced
-the empty string and prevented the nullish heading/main focus fallback. The
-navigation runtime now produces `null` for that case. This packet adds zero
-semantic primitives, ModuleIR kinds, compiler passes, core compiler lines,
-runtime concepts, normalization rules, adapter rules, or public APIs. Runtime
-source remains 351 lines. One positive application fixture gains the complete
-acceptance journey; `/help` remains 0 B JavaScript.
-
-The fixture emits 36 files totaling 106,096 raw / 38,201 aggregate gzip bytes
-with deploy SHA-256
-`4f1ce541af0794fcb17458ad37db7b32beb434f3a83cb7f949f13b3e4536fe1c`.
-The two-route session uses 16 JavaScript files totaling 58,974 raw / 22,129
-aggregate gzip bytes, 4 raw / 5 gzip bytes above `0.12.0`. Seven fresh Chrome
-profiles record a 1.3 ms navigation median with a 1.3/1.5 ms range. The range
-overlaps `0.12.0`, so no latency change is claimed. This correctness packet has
-no measured AI-delivery cost effect.
-
-### `0.12.2`: Authentication And Permission Boundary
-
-- **Purpose:** complete login, restore, 401, logout, and permission-aware UI.
-- **Expected boundary:** shared layout state, owned fetch, storage, and server or
-  host authorization.
-- **Acceptance:** invalid and valid login, reload restoration, direct protected
-  entry, role-aware controls, server rejection, token clear, and logout.
-- **Stop condition:** client conditional rendering is presented as a security
-  boundary.
-- **Done condition:** server authorization and client UX responsibilities are
-  executable and explicit.
-
-**Completion evidence:** the project application replaces its placeholder login
-route with a native constrained form and one async handler. The shared layout
-restores a stored token, validates the current user, exposes authenticated
-identity and role state, and clears local state before replacement navigation on
-rejection or logout. Required Chrome proves invalid and valid login, reload
-restoration, anonymous and rejected direct protected entry, member/admin control
-visibility, server 401/403 enforcement, token clearing, and logout. Client
-conditions are recorded as presentation only; the deterministic API server
-authorizes every protected read and mutation.
-
-Existing native forms, state, layout ownership, effects, storage, fetch, and
-native navigation are sufficient. This packet adds zero semantic primitives,
-ModuleIR kinds, compiler passes, production compiler/runtime lines, runtime
-concepts, normalization rules, adapter rules, public APIs, auth runtimes, or
-servers. The fixture emits 43 files totaling 123,476 raw / 45,305 aggregate gzip
-bytes with deploy SHA-256
-`81d0b3c5e0d1d1f72f29d219647601463f5317f215274e14a45fe5ba92eb033e`.
-The login route uses 12,093 raw / 5,795 gzip JavaScript bytes. The two-route
-session uses 16 JavaScript files totaling 60,767 raw / 22,561 aggregate gzip
-bytes. Seven fresh Chrome profiles record a 1.4 ms navigation median with a
-1.2/1.5 ms range; the range overlaps `0.12.1`, so no latency change is claimed.
-`/help` remains 0 B JavaScript.
-
-### `0.12.3`: Route Failure And Restoration Policy
-
-- **Purpose:** decide application-visible handling of navigation/data failures.
-- **Expected boundary:** native fallback and application-owned error UI first.
-- **Acceptance:** invalid document, asset failure, offline/error state, retry,
-  focus restoration, and no half-replaced route for those precommit failures.
-- **Stop condition:** a generic React-style error-boundary renderer is required.
-- **Done condition:** each classified transport, document-validation,
-  capability-loading, or stylesheet-loading failure leaves a valid document
-  and recoverable navigation path.
-
-**Completion evidence:** required Chrome injects deterministic fetch and
-response-body transport rejections through a pending prefetch and click retry,
-then proves the current URL, route DOM, and retained layout remain unchanged.
-The test moves focus away while the request is pending; the existing polite
-navigation status announces the failure and restores the initiating link.
-Activating that link again completes list-to-detail replacement and focuses the
-destination heading.
-Separate invalid-identity, missing-capability-module, and missing-stylesheet
-responses fail before route disposal and recover through the existing native
-document fallback. The maintained HTTP 500 project read still reports an
-application-owned alert and succeeds on explicit retry.
-
-Only `navigation-runtime.js` changes: fetch and response-body transport failures
-receive one private error classification, while document validation and asset
-failures keep their existing fallback. The source grows from 351 to 374 lines and adds no
-semantic primitive, ModuleIR kind, compiler pass, core compiler line, runtime
-concept, public API, error-boundary renderer, route registry, or additional
-browser file. The application emits 43 files totaling 123,835 raw / 45,379
-aggregate gzip bytes with deploy SHA-256
-`fb9d24cc01791bf80b67b659738ba62beded6ff6ce14a86a278fa4b34d088acf`.
-The two-route session remains 16 JavaScript files totaling 61,126 raw / 22,663
-aggregate gzip bytes, 359 raw bytes above `0.12.2`; the unpaired,
-environment-sensitive 102-byte gzip difference is not attributed to the patch.
-`/help` remains 0 B JavaScript. Seven fresh Linux x64 Chrome 142 profiles record
-a 1.8 ms median with a 1.5/2.4 ms range. No latency comparison is claimed
-against the macOS arm64 Chrome 151 `0.12.2` samples.
-
-### `0.12.4`: Nested Layout Evidence Decision
-
-- **Purpose:** determine whether multiple independently retained layout owners
-  are required.
-- **Acceptance:** at least three unrelated application routes demonstrate the
-  same lifetime that one shared layout plus static composition cannot express.
-- **Stop condition:** route nesting is only a source-organization preference.
-- **Done condition:** either existing layout composition is accepted or a
-  minimal layout-owner chain is approved with explicit disposal order.
-
-**Completion evidence:** architecture review covers the project application's
-list, static detail, runtime project, and runtime issue routes; Apache Answer's
-questions, tags, admin, and legal route-shell reduction; its independently
-reduced authentication shell; the Context and Zustand cart reductions; and the
-two disjoint navigation-group shells. All executable evidence fits one retained
-layout owner around one replaceable route owner, followed by existing
-conditional, keyed, effect, and DOM ownership. The project runtime routes use
-native document navigation, the Answer shell is static composition, and the
-independent navigation groups require disjoint top-level owners rather than a
-nested owner chain.
-
-No reviewed route proves an intermediate layout whose DOM, state, or effects
-must survive descendant route replacement and then dispose on subtree exit.
-The qualifying evidence count is zero, below the required three unrelated
-routes. The stop condition is met: current nesting is source organization or
-static composition, so this packet accepts the existing single-layout model and
-does not authorize a layout-owner chain. A future review must first prove the
-same otherwise-inexpressible intermediate lifetime in three unrelated
-executable application routes, including leaf-before-intermediate and
-deepest-first disposal requirements.
-
-The project application records `closed-by-stop-condition`, the reviewed route
-sets, an empty qualifying-route list, and the reused ownership semantics in its
-machine-readable capability contract. This packet adds zero semantic
-primitives, ModuleIR or RouteIR kinds, compiler passes, production compiler or
-runtime lines, runtime concepts, normalization or adapter rules, public APIs,
-layout registries, disposal paths, browser files, or browser bytes. Existing
-fixtures and output remain unchanged: 43 deploy files totaling 123,835 raw /
-45,379 aggregate gzip bytes with the same deploy digest, a 61,126 raw / 22,663
-aggregate gzip two-route JavaScript session, and 0 B JavaScript on `/help`. No
-benchmark rerun or latency claim applies to a machine-readable evidence decision
-with no generated-output change.
-
-## `0.13.x`: Forms
-
-### `0.13.0`: Production Form And Server Validation
-
-- **Purpose:** complete issue creation with native constraints and server errors.
-- **Expected boundary:** native form, FormData, state, and async handler.
-- **Acceptance:** keyboard submit, native constraints, pending state, field and
-  form errors, retry, focus/ARIA linkage, and retained valid input.
-- **Stop condition:** React Hook Form is reproduced.
-- **Done condition:** complete accessible create/edit behavior passes without a
-  form runtime.
-
-**Completion evidence:** the maintained Alpha project detail now authors one
-ordinary uncontrolled issue form with native `required` and `minLength`
-constraints, `FormData`, three route-owned primitive states, one object ref, and
-one async submit handler. Required Chrome sends real Enter key events through
-the browser input domain and proves empty and short titles make zero requests.
-For valid input, the deterministic server returns a delayed `422` field error,
-then a `503` form error, then a `201` success. The journey verifies pending
-disable/label state, title focus, `aria-invalid` and `aria-describedby` linkage,
-alert/status roles, exact attempt/create counts, retry without re-entry, and
-retention of the valid title and description through every response.
-
-The existing native-form, state, ref, conditional, binding, and handler
-semantics cover the complete packet. Production compiler and runtime source do
-not change. The application route grows from 20 to 70 authored lines and emits
-one additional route handler file; no form package, registration API, proxy
-metadata graph, schema adapter, semantic primitive, IR kind, compiler pass,
-runtime concept, public API, or browser runtime file is added. `/help` remains
-complete HTML with 0 B JavaScript.
-
-The project application emits 44 files totaling 127,343 raw / 46,261 aggregate
-gzip bytes with deploy SHA-256
-`183f5a7ca081f99ae38de6974e61fdcb32ca49de699359aac053d0154ae38036`.
-The two-route session uses 17 JavaScript files totaling 62,118 raw / 23,096
-aggregate gzip bytes, 992 raw / 433 gzip bytes above `0.12.4`; the detail route
-itself adds the same 992 raw bytes and 440 gzip bytes. Seven fresh Linux x64
-Chrome 142 profiles record list-to-detail samples of
-`[3.8, 5.6, 4.3, 2.6, 2.9, 3.6, 3.1]` ms, a 3.6 ms median and 2.6/5.6 ms range.
-The range does not overlap `0.12.4`; this is the measured cost of adding the
-form DOM and route-specific handler to the destination application route, not a
-same-content compiler/runtime comparison or a latency improvement claim.
-
-### `0.13.1`: Nested Fields, Field Arrays, Dirty, And Touched
-
-- **Purpose:** determine which form metadata needs framework support.
-- **Expected boundary:** object/array state and keyed rows first.
-- **Acceptance:** dynamic assignee/checklist rows, reorder/remove, conditional
-  fields, dirty/touched display, reset, and exact row identity.
-- **Stop condition:** a registration/proxy runtime is added before three
-  independent forms require the same metadata graph.
-- **Done condition:** application composition is accepted or the smallest
-  repeated metadata semantic is isolated.
-
-**Completion evidence:** the maintained Alpha issue form now owns nested
-assignee metadata as one ordinary object state and a dynamic checklist as one
-ordinary array state with stable string keys. Native `input`, `blur`, `change`,
-and `reset` events drive explicit dirty/touched display, conditional assignee
-fields, immutable checklist add/reorder/remove operations, and reset to the
-authored baseline. Required Chrome proves retained checklist keys preserve exact
-DOM identity through reorder and reset, removed and reset-added keys disconnect,
-conditional ownership releases the assignee controls, and native reset clears
-uncontrolled title/body values plus application-owned metadata. The existing
-keyboard constraints and delayed `422`, `503`, and `201` server journey continue
-to pass while nested assignee/checklist values survive field and form failures.
-
-The first failing fixture was the `0.13.1` project-application contract against
-the released `0.13.0` Alpha form. Native controls and existing object/array
-state, condition, keyed-list, binding, and handler semantics close it without a
-normalization rule, adapter, semantic primitive, IR kind, compiler pass,
-production compiler/runtime line, runtime concept, public API, registration
-function, proxy metadata graph, schema adapter, or form runtime. The
-machine-readable decision is `closed-by-application-composition`; dynamic field
-names, schema-driven registration, generic watchers, and proxy dirty/touched
-graphs remain unsupported until independent real forms prove a repeated gap.
-
-The application emits 44 files totaling 138,064 raw / 47,964 aggregate gzip
-bytes with deploy SHA-256
-`f0d96cfd5dee2cf3fa67f4ebe25d8b32cbb1bf66e72ed5ca1c29bd06fdcc0a33`.
-The maintained two-route session remains 17 JavaScript files and grows by 3,286
-raw / 803 aggregate gzip bytes to 65,404 / 23,899 bytes. This is authored
-route-specific form behavior, not a shared form runtime; `/help` remains 0 B
-JavaScript. Seven fresh Linux x64 Chrome 142 profiles record list-to-detail
-samples of `[3.5, 3.3, 4.4, 3.3, 3.2, 4.7, 3.3]` ms, a 3.3 ms median and
-3.2/4.7 ms range. The range overlaps `0.13.0` and the destination content
-changed, so no latency regression or improvement is claimed.
-
-### `0.13.2`: Multistep Draft And Autosave
-
-- **Purpose:** prove draft persistence across steps and failures.
-- **Expected boundary:** routes or conditionals, state, dependency effects,
-  debounce cleanup, and storage/server persistence.
-- **Acceptance:** step navigation, validation gate, debounced save, stale-save
-  rejection, reload restore, reset, and conflict/error state.
-- **Stop condition:** a wizard or autosave scheduler is added for one form.
-- **Done condition:** no data loss occurs across navigation, reload, or failure.
-
-Development evidence is complete. The Alpha application now owns a two-step
-project setup draft with native first-step constraints, parent-owned values,
-versioned local storage, one mount restore effect, and one dependency effect
-whose direct timeout cleanup debounces an owned server write. Required Chrome
-proves step navigation, validation, two overlapping versions, server rejection
-of the older save, ignored stale completion, enhanced-navigation restoration,
-reload restoration, retained conflict input, and reset of state, storage, and
-pending work.
-
-The first failing fixture was the `0.13.2` project-application contract against
-the released `0.13.1` Alpha route. Existing ordinary state, conditional
-ownership, dependency invalidation, effect cleanup, owned fetch, late-write
-rejection, and guarded versioned storage close it without a semantic primitive,
-IR kind, compiler pass, production compiler/runtime line, normalization rule,
-adapter, runtime concept, public API, wizard scheduler, or autosave scheduler.
-The machine-readable decision is `closed-by-application-composition`.
-
-The application emits 44 files totaling 144,633 raw / 49,217 aggregate gzip
-bytes with deploy SHA-256
-`06a35341c5e70bfec48c46ad767ac351326b39ba701f5c1f1d03992a65ec8b9e`.
-The maintained two-route session remains 17 JavaScript files and grows by 3,260
-raw / 848 aggregate gzip bytes to 68,664 / 24,747 bytes; `/help` remains 0 B
-JavaScript. Seven fresh Linux x64 Chrome 142 profiles record list-to-detail
-samples of `[10.7, 7.4, 3.7, 7.9, 6.1, 4.0, 5.4]` ms, a 6.1 ms median and
-3.7/10.7 ms range. The destination content changed and the range is noisy, so
-no same-content latency regression or improvement is claimed.
-
-### `0.13.3`: File Upload Boundary
-
-- **Purpose:** move beyond local file reading to a real upload lifecycle.
-- **Expected boundary:** native input, FormData/fetch or a proven browser API.
-- **Acceptance:** type/size validation, progress when the selected transport can
-  expose it, cancellation, failure/retry, route departure, and successful
-  attachment mutation.
-- **Stop condition:** chunking/resume/background upload is added without a real
-  application requirement.
-- **Done condition:** the supported transport and unsupported progress/resume
-  boundary are explicit and tested.
-
-Development evidence is complete. The Alpha application accepts one plain-text
-attachment up to 1 KiB, rejects invalid type and size before any request, reads
-the selected file into bounded route state, and reconstructs it as a `Blob` in
-one dependency effect that owns `FormData`, fetch, and `AbortController`.
-Required Chrome proves explicit user cancellation, delayed `503` feedback,
-retry without reselection, successful keyed attachment mutation, and exact
-abort when enhanced navigation releases the route.
-
-The first failing fixture was the `0.13.3` project-application contract against
-the released `0.13.2` Alpha route. Existing native file handlers, ordinary
-state, dependency invalidation, effect cleanup, owned fetch, route release, and
-keyed list identity close it without a semantic primitive, IR kind, compiler
-pass, production compiler/runtime line, normalization rule, adapter, runtime
-concept, public API, upload scheduler, or transfer runtime. The machine-readable
-decision is `closed-by-application-composition`.
-
-Fetch does not expose upload progress, so the application renders that boundary
-explicitly and does not emit a fake progress element. Chunking, resume,
-background sync, binary formats, multiple concurrent files, and files above the
-authored 1 KiB in-memory text limit remain unsupported. Add a different
-transport only when a real application requires one of those capabilities.
-
-The application emits 44 files totaling 150,023 raw / 50,240 aggregate gzip
-bytes with deploy SHA-256
-`a16fd5dbc08086aa6a7c0ec0aba3a38fbe2a63b96005a81e8747a7f4e45e214e`.
-The maintained two-route session remains 17 JavaScript files and grows by 2,367
-raw / 557 aggregate gzip bytes to 71,031 / 25,304 bytes; `/help` remains 0 B
-JavaScript. Seven fresh Linux x64 Chrome 142 profiles record list-to-detail
-samples of `[3.7, 5.0, 3.5, 3.7, 3.8, 3.8, 4.0]` ms, a 3.8 ms median and
-3.5/5.0 ms range. The destination content changed, so no same-content latency
-regression or improvement is claimed.
-
-## `0.14.x`: Lists, Tables, And Virtualization
-
-### `0.14.0`: Project Table CRUD And Identity
-
-- **Purpose:** establish the production table baseline.
-- **Expected boundary:** existing keyed list runtime and valid table structure.
-- **Acceptance:** insert, update, delete, reorder, sort, filter, selection,
-  row-local edit state, keyboard access, and retained DOM identity.
-- **Stop condition:** a data-grid runtime is introduced for ordinary tables.
-- **Done condition:** complete CRUD behavior passes with measured update latency.
-
-Evidence complete. The first failing fixture was the
-`0.14.0` project-application contract against the previous article list. It now
-uses a native table, ordinary array/object state, the existing pure collection
-selector, three existing keyed row-state slots, and native button/input keyboard
-behavior for insert, update, delete, reverse, sort, filter, selection, and edit.
-The browser journey proves retained row and draft-input identity through reorder
-and sort, fresh identity/state after filter removal and restoration, and focused
-keyboard reachability. Native `hidden` keeps the editor mounted rather than
-introducing a conditional callback-scope adapter or data-grid runtime.
-
-This slice adds zero semantic primitives, zero compiler passes, zero core LOC,
-and zero runtime concepts. It changes one real fixture, its contract, the
-machine-readable capability record, and the existing project benchmark. The
-application remains 44 files and changes from the `0.13.3` baseline by +9,235
-raw / +1,387 aggregate gzip bytes to 159,258 / 51,627 bytes, with deploy SHA-256
-`ebb3358e7a03612459e723ae765c39d105498db9d293415b924f1066cddf4793`.
-The maintained two-route session remains 17 JavaScript files and grows by 2,273
-raw / 626 aggregate gzip bytes to 73,304 / 25,930 bytes; `/help` remains 0 B
-JavaScript.
-
-Seven fresh Linux x64 Chrome 142 profiles record table Save-click-to-committed-
-DOM samples of `[0.7, 0.8, 1.0, 0.7, 1.1, 0.7, 0.8]` ms, a 0.8 ms median and
-0.7/1.1 ms range. This is the first equivalent table-update measurement, so no
-same-content latency delta is claimed. The same runs record list-to-detail
-samples of `[3.9, 3.9, 5.5, 4.2, 6.1, 3.5, 3.4]` ms, a 3.9 ms median and
-3.4/6.1 ms range. No compiler/runtime production logic changed; `0.14.0`
-publishes the accepted application-capability packet and `create-kudzu@0.1.122`
-generates projects on `@kudzujs/core@^0.14.0`.
-
-### `0.14.1`: Nested And Object-State Collections
-
-- **Purpose:** prove project/group/issue/checklist nesting from ordinary object
-  state without authored field state.
-- **Expected boundary:** existing object-property links and nested KeyedBlockIR.
-- **Acceptance:** replacement, nested insertion/removal/reorder, latest handlers,
-  row cleanup, invalid-shape rejection, and static sibling exclusion.
-- **Stop condition:** dynamic property paths or mutation require generic runtime
-  object observation.
-- **Done condition:** direct immutable paths cover the application journey and
-  nearby dynamic forms fail clearly.
-
-Release evidence complete. The first failure replaced the project route's
-separate array state with one direct `projectData.projects` field and failed at
-build time because nested lists required their source ownership signal itself to
-contain an array. The compiler already emitted a binding-backed root
-`KeyedBlockIR`, its selector states, and the nested issue child block. Build-time
-list validation now allows an object-valued signal only for a compiler-proven
-`ownerField`; root lists still require arrays and nested non-array fields still
-fail. Browser mounting initializes binding-backed roots with children from their
-serialized state before the evaluator module loads, preserving the existing
-parent prototype and ownership path.
-
-Required Chrome proves immutable whole-object replacement, nested issue insert,
-update, reorder, removal, re-addition, retained project/issue identity, latest
-issue handlers, descendant state release, and fresh remount. Existing dynamic
-object-property, alias, mutation, mutating-sort, and non-array diagnostics retain
-the negative boundary, while `/help` remains complete HTML with 0 B JavaScript.
-The slice adds zero semantic primitives, IR kinds, compiler passes,
-normalization rules, adapters, runtime concepts, or public APIs. It changes one
-build-time validation line and adds one production list-runtime line.
-
-The release emits 44 files totaling 164,305 raw / 52,185 aggregate gzip bytes
-with deploy SHA-256
-`0a3b7e10b3447a76a9c04ef34ad7c2bfa30400ea53c9d16fc85b1c7cbf2ea203`.
-The two-route session remains 17 JavaScript files totaling 74,826 raw / 26,204
-aggregate gzip bytes. Seven fresh Chrome 142 profiles record table
-update samples of `[0.8, 0.8, 0.7, 0.8, 0.9, 0.8, 0.8]` ms, a 0.8 ms median and
-0.7/0.9 ms range, plus navigation samples of
-`[3.7, 4.1, 4.5, 3.9, 4.2, 3.9, 4.3]` ms, a 4.1 ms median and 3.7/4.5 ms range.
-`0.14.1` publishes this accepted packet, and `create-kudzu@0.1.123` generates
-projects on `@kudzujs/core@^0.14.1`.
-
-### `0.14.2`: Infinite Loading Composition
-
-- **Purpose:** prove sentinel-driven incremental loading before virtualization.
-- **Expected boundary:** IntersectionObserver, owned fetch, append, keyed list.
-- **Acceptance:** cursor progression, duplicate suppression, end/error/retry,
-  route cleanup, retained identity, and bounded result policy.
-- **Stop condition:** infinite-query or observer runtime is added for one list.
-- **Done condition:** the full journey passes with explicit network and memory
-  bounds.
-
-Published release evidence complete. The maintained project application adds one
-intrinsic sentinel, an owned native `IntersectionObserver`, cursor and request
-primitive state, an owned fetch with `AbortController`, immutable keyed append,
-duplicate ID suppression, and explicit loading/error/retry/end UI. The authored
-policy permits two successful pages of two records and retains at most six unique
-projects. It adds no infinite-query runtime, observer scheduler, cache, retained
-component tree, or virtualization machinery.
-
-Required Chrome proves duplicate sentinel triggers issue one request, duplicate
-records do not remount retained rows, cursor progression reaches an authored
-`503`, retry reaches terminal state, post-terminal triggers issue no request, and
-enhanced navigation disconnects the observer and aborts the in-flight fetch
-signal. Exact request counts and retained DOM identity pass. `/help` remains
-complete HTML with 0 B JavaScript.
-
-The slice adds zero semantic primitives, IR kinds, compiler passes, production
-compiler/runtime lines, normalization rules, adapters, runtime concepts, or
-public APIs. The application emits 44 files totaling 168,037 raw / 52,955
-aggregate gzip bytes with deploy SHA-256
-`230dea098442275e42a966ca595743a4da9c832327ba52d4d09de663c68f8f1a`.
-The maintained two-route session remains 17 JavaScript files totaling 76,889 raw
-/ 26,750 aggregate gzip bytes.
-
-Seven fresh macOS arm64 Chrome 151 profiles record table update samples of
-`[0.4, 0.5, 0.5, 0.3, 0.5, 0.4, 0.6]` ms, a 0.5 ms median and 0.3/0.6 ms range,
-plus navigation samples of `[2.3, 2.3, 2.2, 2.6, 2.5, 2.5, 2.3]` ms, a 2.3 ms
-median and 2.2/2.6 ms range. The environment differs from `0.14.1`, so no timing
-delta is claimed. `0.14.2` publishes this accepted packet, and
-`create-kudzu@0.1.124` generates projects on `@kudzujs/core@^0.14.2`.
-
-### `0.14.3`: 10,000-Item Browser Decision
-
-- **Purpose:** measure direct DOM, pagination, and windowing alternatives.
-- **Acceptance:** equivalent behavior, keyboard access, edit identity, scroll
-  latency, DOM count, heap, and update latency across alternatives.
-- **Stop condition:** a virtualization result is claimed without equivalent
-  behavior or repeated samples.
-- **Done condition:** the selected strategy and thresholds are recorded without
-  assuming direct 10,000-row rendering is acceptable.
-
-Decision evidence is complete. A tracked three-route fixture and
-`npm run benchmark:project-list-decision` now compare one static 10,000-row
-table with pagination and a scroll-driven 100-row window over the same
-deterministic records and native edit controls. Seven rotating fresh Chrome 142
-profiles show direct DOM
-at an 806.6 ms load median and 90,023 DOM nodes, versus 165.4/161.2 ms and a
-1,470/1,471-node median for pagination and the scroll window. Direct JavaScript
-heap is lower because its route is static; DOM size and load remain materially
-worse.
-
-Pagination is selected. Its 12.1 ms range median is lower than the window's 16.5
-ms, it uses 100,300 fewer median JavaScript-heap bytes, and it preserves native
-page, keyboard, focus, and variable-row-height behavior without a scroll listener
-or measurement policy. Both bounded routes release an edited off-range row and
-restore fresh native input identity. The experiment adds no semantic primitive,
-IR kind, compiler pass, production compiler/runtime line, runtime concept, or
-public API. `0.14.4` remains blocked because this one fixture cannot authorize
-virtual range ownership.
-
-### `0.14.4`: Virtual Range Ownership
-
-- **Purpose:** implement bounded DOM only if `0.14.3` and three independent
-  fixtures authorize it.
-- **Expected boundary:** keyed ownership extended by a minimal visible-range
-  owner; no general component renderer.
-- **Acceptance:** spacer/measurement behavior, retained key state, keyboard and
-  focus, dynamic row height policy, cleanup, and bounded heap.
-- **Stop condition:** evidence favors pagination or native containment.
-- **Done condition:** range semantics are package-neutral, optional, measured,
-  and absent from non-virtual routes.
-
-Closed by stop condition. The `0.14.3` decision selected pagination over both
-direct 10,000-row DOM and the authored fixed-height scroll window. No independent
-application fixture proves pagination or application-owned bounded ranges
-insufficient, against the required three. Kudzu therefore adds no visible-range
-owner, measurement policy, range primitive, or virtualization runtime and moves
-to the next evidence-ready application capability.
-
-## `0.15.x`: Overlay And Layer Ownership
-
-### `0.15.0`: Native Dialog Baseline
-
-- **Purpose:** use browser top-layer semantics for destructive issue actions.
-- **Acceptance:** initial focus, Escape/cancel, confirm, trigger focus restore,
-  inert background, route cleanup, and static exclusion.
-- **Stop condition:** a custom modal/focus-trap runtime replaces `<dialog>`.
-- **Done condition:** project delete and edit confirmation pass accessibly.
-
-Validation evidence is complete and the packet closed without consuming a release
-because no production compiler or runtime changed. The maintained
-project application replaces immediate keyed project deletion with one shared
-native `<dialog>`, one route-owned object ref, and one pending project ID state.
-Existing row callback specialization opens the browser top layer; native cancel,
-explicit cancel, and confirmation handlers close it without a Portal, focus trap,
-overlay runtime, or retained component tree.
-
-Required Chrome proves modal top-layer behavior, native initial focus, explicit
-cancel, Escape-equivalent `requestClose()`, trigger-focus restoration, confirmed
-keyed deletion with fallback focus, retained sibling identity, and exact top-layer
-release on enhanced route navigation. The static `/help` sibling remains 0 B
-JavaScript. The slice adds no semantic primitive, IR kind, compiler pass,
-production compiler/runtime line, normalization rule, adapter, runtime concept,
-or public API. The application emits 44 files totaling 169,778 raw / 53,294
-aggregate gzip bytes with deploy SHA-256
-`91c5517e69c341bbd2051f07b3f0ad7df627098ba8e1984561268ea8666f3407`.
-The maintained two-route session remains 17 JavaScript files totaling 77,453 raw
-/ 26,905 aggregate gzip bytes.
-
-Seven fresh macOS arm64 Chrome 151 profiles record table update samples of
-`[0.5, 0.6, 0.6, 0.6, 0.6, 0.4, 0.5]` ms, a 0.6 ms median and 0.4/0.6 ms range,
-plus navigation samples of `[2.3, 2.6, 2.9, 2.6, 2.5, 2.7, 2.5]` ms, a 2.6 ms
-median and 2.3/2.9 ms range. These retained-path ranges overlap the previous
-same-environment evidence, so no timing improvement or regression is claimed.
-
-### `0.15.1`: Popover, Dropdown, And Menu Behavior
-
-- **Purpose:** establish the native-first non-modal overlay boundary.
-- **Expected boundary:** native Popover where available, ordinary state/events,
-  and owned listeners.
-- **Acceptance:** positioning policy, outside click, Escape, trigger restore,
-  roving keyboard behavior where menu semantics apply, and route cleanup.
-- **Stop condition:** package Portal/Slot reconciliation enters the runtime.
-- **Done condition:** supported native and custom boundaries are explicit.
-
-Completed with the maintained project application as the qualifying fixture. A
-keyed project row now authors one top-level `useId()`, a native
-`popovertarget`, and a native `popover="auto"` action group. The compiler scopes
-the generated ID by the existing keyed ownership path and the list runtime
-rewrites only intrinsic `id`, `for`, static ARIA IDREF, `form`, `headers`,
-`list`, and `popovertarget` attributes. Unsupported ID escapes receive a source
-diagnostic. Native Chrome behavior supplies the top layer, Escape, light
-dismiss, and trigger focus restoration; ordinary keyed ownership preserves IDs
-across reorder, scopes inserted rows, recreates deterministic IDs after
-remove/re-add, and releases an open popover during enhanced navigation. The
-authored overlay uses static anchor positioning and `role="group"`, so menu
-roving focus is not applicable.
-
-The slice adds no semantic primitive, IR kind, compiler pass, runtime concept,
-public API, Portal/Slot reconciliation, focus manager, positioning engine, or
-retained overlay tree. It adds 52 and removes 24 production lines across the
-source compiler, RouteIR validation, capability projection, build renderer, and
-existing list runtime. The maintained application emits 44 files totaling
-171,202 raw / 53,629 aggregate gzip bytes with deploy SHA-256
-`46ba45b4a423b9627c6ede5f8dd3a667f593cb5401454ae42ecc66207b07b323`.
-The two-route session remains 17 JavaScript files totaling 77,779 raw / 27,050
-aggregate gzip bytes, a 326 raw / 145 gzip byte increase from `0.15.0`; the
-static `/help` sibling remains 0 B JavaScript. Familiar React-shaped `useId()`
-authoring remains intact, so no application adapter or migration-only source
-structure is added.
-
-Seven fresh macOS arm64 Chrome 151 profiles record table update samples of
-`[0.4, 0.4, 0.5, 0.4, 0.5, 0.5, 0.5]` ms, a 0.5 ms median and 0.4/0.5 ms range,
-plus navigation samples of `[4, 2.4, 2.6, 2.5, 2.5, 2.8, 2.4]` ms, a 2.5 ms
-median and 2.4/4 ms range. These retained-path ranges overlap prior evidence,
-so no timing improvement or regression is claimed.
-
-Release transaction complete at commit
-`2dd5c409d3553cdf0b04080f1ba7c39b4541c523` and immutable tag `v0.15.1`.
-CI run `33069675599` passed both jobs. The GitHub release is
-[`v0.15.1`](https://github.com/kudzujs/kudzu/releases/tag/v0.15.1), and publish
-workflow run `33071439012` published and verified `@kudzujs/core@0.15.1` and
-`create-kudzu@0.1.126`. A fresh registry install imported `useId()` from core
-and confirmed both exact package versions. The first publish attempt stopped at
-the existing concurrent Chrome timing gate before any publish step; the exact
-same immutable tag passed on the release-event retry.
-
-### `0.15.2`: Toast And Notification Ownership
-
-- **Purpose:** provide application feedback across route mutations.
-- **Expected boundary:** layout state, keyed list, timer effects, and live region.
-- **Acceptance:** queue, deduplication, timeout, pause/focus behavior if required,
-  dismissal, route persistence, announcement, and disposal.
-- **Stop condition:** a global scheduler is added before ordinary composition is
-  shown insufficient.
-- **Done condition:** notifications remain owner-bounded and absent from unused
-  route output.
-
-**Result:** closed by ordinary layout composition. The maintained project
-application now owns one deduplicated notification queue in layout state, renders
-keyed rows in a native polite live region, and uses one cleanup-owned sequential
-timeout plus explicit dismissal. Required Chrome preserves the live-region and
-row identities through enhanced route replacement, proves timeout removal and
-disposal, and keeps `/help` at zero JavaScript. The 45-file deploy totals 193,770
-raw / 60,227 aggregate gzip bytes. No scheduler, primitive, runtime concept, or
-production source changed, so no release was consumed.
-
-### `0.15.3`: Layer Owner Decision
-
-- **Purpose:** decide whether logical ownership of DOM rendered outside its
-  authored range is necessary.
-- **Acceptance:** three independent overlay fixtures fail native top-layer and
-  in-place composition for the same reason.
-- **Stop condition:** the request is React Portal source compatibility alone.
-- **Done condition:** no new concept is added, or a minimal owner edge with exact
-  cleanup/focus order is approved.
-
-**Result:** closed with no owner edge. Native dialog and Popover already satisfy
-modal and non-modal top-layer behavior, while maintained dropdown, tooltip,
-notification, and menu fixtures compose in their authored ranges. Zero fixtures
-fail because DOM must render outside its authored range, below the required three
-independent failures. Portal source compatibility alone remains insufficient;
-no production source or release changed.
-
-## `0.16.x`: External Complex UI
-
-### `0.16.0`: Bounded External UI Contract
-
-- **Purpose:** define and prove `mount`, `update`, `dispose`, owner, lifetime, and
-  route-specific assets without a React island.
-- **Expected boundary:** DOM ref plus owned effect first.
-- **Acceptance:** a small imperative package creates and disposes one real DOM or
-  canvas instance; static siblings exclude it.
-- **Stop condition:** the package requires React reconciliation.
-- **Done condition:** package ownership remains inside effect/handler ESM and no
-  generic widget runtime is added.
-
-**Result:** closed through existing effect ownership. A real Typed.js instance
-creates DOM from one attached host ref, replaces on application-state dependency
-change, disposes on conditional/document release, and remounts fresh. The package
-is absent from the complete zero-JavaScript static sibling. No compiler, IR,
-runtime concept, or production source changed, so no release was consumed.
-
-### `0.16.1`: Real Editor Lifecycle
-
-- **Purpose:** integrate one real CodeMirror-class editor.
-- **Acceptance:** mount, initial value, application-state update, editor-to-state
-  update, dependency changes, route disposal, remount, errors, and accessibility.
-- **Stop condition:** preserving the React wrapper requires a React island.
-- **Done condition:** the native package owns its DOM and Kudzu owns only its
-  bounded lifecycle.
-
-**Result:** complete. The Apache Answer-derived CodeMirror fixture exposes the
-first real retained-instance failure: an unattached `useRef(null)` shared by one
-mount/cleanup effect and a later dependency update effect was incorrectly treated
-as a DOM ref. Kudzu now lowers exactly one proven retained handle to existing
-owner-scoped object state. Required Chrome proves initial value, retained
-application updates, editor-to-state updates, package-error recovery, accessible
-textbox naming, conditional disposal, fresh remount, and document disposal. The
-fixture emits 214,968 raw / 71,607 aggregate gzip JavaScript bytes while its
-static sibling emits zero. One existing normalization pass grows; semantic
-concepts, IR kinds, runtime concepts, runtime files, and public APIs remain
-unchanged.
-
-### `0.16.2`: Chart And Map Lifecycle
-
-- **Purpose:** validate the same contract against unrelated canvas/SVG/map DOM.
-- **Acceptance:** resize/data update, listener/source cleanup, route transition,
-  retained instance where required, and asset exclusion.
-- **Stop condition:** package-specific concepts enter generic IR/runtime.
-- **Done condition:** one lifecycle model covers editor and chart/map evidence.
-
-**Result:** closed through the `0.16.1` retained-instance contract. A reduced
-Mattermost Chart.js doughnut chart first reproduced the existing diagnostic when
-its dependency effect both created and updated the retained instance. Moving
-acquisition, resize-listener registration, and exact destroy/reset cleanup into
-one mount effect leaves later data effects read-only and lowers the handle to the
-same owner-scoped object state used by CodeMirror. Required Chrome proves initial
-drawing, retained data and resize updates, listener removal, enhanced route
-disposal, fresh remount, accessibility, and package exclusion from the static
-sibling. The interactive graph is 208,686 raw / 73,312 aggregate gzip bytes
-across six JavaScript files; the static sibling emits zero. Semantic primitives,
-IR kinds, compiler passes, core/compiler LOC, normalization rules, runtime
-concepts, runtime files, and public APIs change by zero. One positive fixture and
-one browser contract were added. The maintained Worker/effect benchmark remains
-907 raw / 477 gzip bytes for the Worker graph and 14,456 raw / 6,159 gzip bytes
-for the window graph; its seven-run Linux median was 2,464.9 ms, recorded without
-a timing comparison because the host differs from the published baseline. No
-production source changed, so no release was consumed.
-
-### `0.16.3`: Grid And Drag/Drop Decision
-
-- **Purpose:** test external data-grid or drag engine ownership against keyed
-  Kudzu state.
-- **Acceptance:** keyboard equivalent, reorder/mutation, state synchronization,
-  package disposal, and failure recovery.
-- **Stop condition:** two independent renderers attempt to own the same DOM.
-- **Done condition:** either bounded ownership is proven or the package is
-  explicitly unsupported with a migration path.
-
-**Result:** complete through existing keyed state and effect ownership. A
-THRM-derived SortableJS fixture reproduced one compiler failure because the
-browser `HTMLElement` constructor was classified as an effect capture and then
-evaluated during the Node build. Adding that platform constructor to the
-existing browser-global set fixes the shared root cause while lexical shadowing
-remains covered. During a gesture SortableJS may move DOM temporarily, but its
-`onEnd` callback restores authored order before one immutable state reorder;
-Kudzu's keyed reconciler remains the durable DOM owner and preserves retained
-row and input identity. The same state operation provides keyboard movement and
-reset, while invalid package indexes restore DOM without changing state.
-Required Chrome proves drag and keyboard reorder, state synchronization,
-identity, invalid-input recovery, conditional/document disposal, and fresh
-remount. The interactive graph is 72,251 raw / 26,603 aggregate gzip bytes
-across ten JavaScript files; the static sibling emits zero. Its normalized
-deploy archive SHA-256 is
-`a55c18e14ba29df3f1f8ccdc7bcae936092508a01d05abdb887417e6fffb2ecf`.
-Semantic primitives, IR kinds, compiler passes, normalization rules, runtime
-concepts, runtime files, drag/drop abstractions, and public APIs change by zero;
-one existing compiler lookup gains one token. One positive fixture, one browser
-contract, and one browser-global shadowing case were added. The maintained
-Worker/effect benchmark remains 907 raw / 477 gzip bytes for the Worker graph
-and 14,456 raw / 6,159 gzip bytes for the window graph; its seven-run Linux
-median was 999.7 ms, recorded without a timing comparison because the host
-differs from the published baseline.
-
-### `0.16.4`: GSAP Animation Lifecycle
-
-- **Purpose:** validate one real GSAP landing-page animation against existing
-  DOM-ref and effect ownership.
-- **Expected boundary:** one owned effect and native reduced-motion behavior;
-  no animation IR or scheduler runtime.
-- **Acceptance:** mount, deterministic static fallback, scoped targets,
-  prefers-reduced-motion behavior, dependency update where required, route and
-  conditional disposal, fresh remount, and static sibling asset exclusion.
-- **Stop condition:** package timelines compete with Kudzu for durable DOM
-  structure or require a retained component renderer.
-- **Done condition:** GSAP owns only bounded presentation changes and exact
-  cleanup while Kudzu retains document structure and lifetime ownership.
-
-**Result:** complete. A reduction of the MIT-licensed
-Magic Modal landing page at revision
-`ab3df864de5b01d604d26345165ea5f7900eae14` reproduced one compiler failure:
-GSAP's root-scoped `context()` receives an inline callback whose direct package
-calls were rejected because package-reference validation stopped at the nearest
-function. The existing ancestry check now accepts only an unbroken chain of
-inline call arguments back to the owned effect; named, local, and imported helper
-indirection remains rejected. Required Chrome proves deterministic visible HTML,
-root-scoped presentation, native reduced-motion fallback without residual inline
-opacity/transform, dependency replacement with retained DOM identity,
-conditional and enhanced-route disposal, and fresh remount. GSAP owns only
-bounded opacity and transform changes; Kudzu retains structural DOM and lifetime
-ownership. The interactive graph is 94,110 raw / 37,819 aggregate gzip bytes
-across eight JavaScript files; the static sibling emits zero. Its normalized
-deploy archive SHA-256 is
-`2b6467f04452c51878ea81d5e80db21c60286e191e1e49ac482556dbf7c713c3`.
-Semantic primitives, IR kinds, compiler passes, normalization rules, runtime
-concepts, runtime files, animation abstractions, schedulers, and public APIs
-change by zero; one existing compiler ancestry helper shrinks by two LOC. One
-positive fixture and one two-scenario browser contract were added. The maintained
-Worker/effect benchmark remains 907 raw / 477 gzip bytes for the Worker graph and
-14,456 raw / 6,159 gzip bytes for the window graph; its seven-run Linux median was
-625.9 ms, recorded without a timing comparison because the host differs from the
-published baseline.
-
-### `0.16.5`: Endurance Harness
-
-- **Purpose:** make long-session correctness locally reproducible.
-- **Expected files:** a dedicated browser soak harness and machine-readable heap,
-  listener, resource, DOM, and navigation counters.
-- **Acceptance:** repeated route, conditional, keyed, dialog, and editor cycles;
-  raw traces are retained.
-- **Stop condition:** only a short synthetic timing loop is measured.
-- **Done condition:** failures identify the owner that retained memory or work.
-
-**Result:** closed through existing document, route, conditional, keyed, and
-effect ownership. A dedicated application fixture repeatedly opens and closes a
-native dialog, removes and recreates keyed row state, conditionally disposes and
-remounts a real CodeMirror editor, and navigates to a canonical released route.
-The standalone Chrome harness keeps one browser profile alive, runs five warm-up
-and 30 measured cycles by default, forces GC only before five-cycle checkpoints,
-and retains machine-readable `samples.jsonl`, `summary.json`, and
-`environment.json` artifacts under the ignored `test-results/endurance/` path.
-The required default run completed 71 enhanced navigations with 71 editor mounts
-and 71 disposals. Every released checkpoint had zero active editors, dialogs,
-and keyed rows, 29 route elements, six documents, 141 DOM nodes, and seven event
-listeners. Forced-GC heap moved from 1,820,132 to 2,005,572 bytes, below the
-larger of the declared 2 MiB or 15% observation alarm. A deterministic failure
-reports `route:/plain` and the mismatched counters; browser exceptions and failed
-requests are retained with the same owner. The interactive route graph is
-242,928 raw / 81,792 aggregate gzip bytes across 11 JavaScript files. Semantic
-primitives, IR kinds, compiler passes, core/compiler LOC, runtime concepts,
-runtime files, and public APIs change by zero; one test fixture and one browser
-harness were added. This evidence changed no production source by itself and was
-published with the later stale-prefetch fix in the actual `0.16.5` patch.
-
-Release transaction complete at commit
-`f23db7403060df7615b684f3a4a39781e600ed23` and immutable tag `v0.16.5`.
-CI run `33235364419` passed Node 22, required-Chrome Node 24, and the Cloudflare
-deployment. The GitHub release is
-[`v0.16.5`](https://github.com/kudzujs/kudzu/releases/tag/v0.16.5), and publish
-workflow run `33235705347` published and verified `@kudzujs/core@0.16.5` and
-`create-kudzu@0.1.130`. A fresh registry-only install imported core, confirmed
-both exact package versions, generated a two-route application, and passed its
-TypeScript and production build check.
-
-## `0.17.x`: Long-Lived Applications
-
-### `0.17.0`: WebSocket Application Journey
-
-- **Purpose:** move from isolated fake socket ownership to project notifications.
-- **Acceptance:** connect, message, keyed update, reconnect, stale socket
-  rejection, route/layout lifetime, offline/error state, and exact cleanup.
-- **Stop condition:** a shared transport runtime is added for one consumer.
-- **Done condition:** repeated navigation balances every listener, timer, and
-  socket handle.
-
-**Result:** closed through existing layout state, `EffectIR`, keyed-list, and
-enhanced-navigation ownership. The first project-application browser extension
-failed at `websocket-connect`; one inline layout effect now owns a native
-WebSocket, four socket listeners, native online/offline listeners, generation
-and socket-identity stale guards, and one 50 ms reconnect timer. Incoming
-snapshots are validated before replacing the existing notification array, so a
-same-key server update retains its DOM node and live-region ownership. Required
-Chrome proves connect, message, keyed update, visible error, reconnect, stale
-message/error rejection, offline cancellation, online recovery, and eight
-list/detail enhanced transitions with one retained layout socket. The journey
-creates three sockets total; while connected it has one active socket, four
-socket listeners, and two online/offline listeners. Of two reconnect timers,
-one fires and one is cleared. Final document disposal leaves zero active sockets,
-socket listeners, window listeners, and reconnect timers, and repeated disposal
-changes no counter. The project deploy grows by 2,908 raw / 652 aggregate gzip
-bytes; the list-route JavaScript graph grows by 1,755 raw / 460 aggregate gzip
-bytes, while `/help` remains zero JavaScript. Semantic primitives, IR kinds,
-compiler passes, core/compiler LOC, runtime concepts, runtime files, transport
-abstractions, and public APIs change by zero; one existing application fixture
-and one browser scenario grow. The maintained Worker/effect graph remains 907
-raw / 477 gzip bytes for Worker and 14,456 raw / 6,159 gzip bytes for window;
-its seven-run Linux median was 605.6 ms. No production source changed, so no
-release was consumed.
-
-### `0.17.1`: Shared SSE/Transport Decision
-
-- **Purpose:** test multiple independently mounted consumers, replay, and
-  reconnect against existing layout effects/shared state.
-- **Acceptance:** subscription add/remove, one transport, replay ordering,
-  reconnect, consumer removal, final close, and bounded history.
-- **Stop condition:** fewer than three independent applications require the same
-  shared transport semantics.
-- **Done condition:** existing composition passes or a minimal transport owner is
-  architecture-approved.
-
-**Result:** closed by the stop condition with zero qualifying applications.
-Mattermost and Twenty remain external research candidates, but neither has a
-repository-owned reduction proving the complete shared-consumer, ordered replay,
-reconnect, independent removal, final close, and bounded-history contract. The
-project application has one layout consumer; all Lupin reductions belong to one
-application and each owns its socket in one page effect; the route WebSocket,
-navigation telemetry, Worker dashboard, Memos audit, and development EventSource
-also lack the complete shared transport shape. Existing layout state and
-independent `EffectIR` owners remain authoritative. Semantic primitives, IR
-kinds, compiler passes, core/compiler LOC, runtime concepts, runtime files,
-fixtures, browser bytes, and public APIs change by zero. No production source
-changed, so no release was consumed.
-
-### `0.17.2`: Memory And Disposal Gate
-
-- **Purpose:** close document/layout/route/range/external-instance memory risks.
-- **Acceptance:** long repeated journeys show bounded heap, DOM, prefetch cache,
-  resource handles, listener counts, and state maps.
-- **Stop condition:** undocumented browser GC timing is presented as exact proof.
-- **Done condition:** retained growth has a declared bound and repeated samples
-  pass the agreed threshold.
-
-**Result:** complete with one navigation-cache ownership fix. A deterministic
-race holds unique prefetched documents, starts navigation to each, commits a
-canonical route that prunes them, and then releases the stale responses. Before
-the fix, five races left `browserState.size` at zero but increased CDP documents
-from 8 to 16 under owner `navigation:documents`: stale `navigate()` calls wrote
-their parsed documents back into the cache before checking the navigation
-revision. `navigation-runtime.js` now checks revision before the existing cache
-write; no cache type, LRU, cancellation registry, or public instrumentation was
-added. The default gate batches ten races within browser connection limits, then
-runs 30 route, conditional, keyed, dialog, and editor cycles in one Chrome
-profile. Across 101 navigations, baseline/final documents remain 8, browser state
-entries 0, DOM nodes 174, listeners 7, and editor mounts/disposals 71/71. Forced-
-GC heap moves from 1,826,244 to 2,022,160 bytes, below the declared larger-of-2-
-MiB-or-15% alarm. Semantic primitives, IR kinds, compiler passes, runtime
-concepts, runtime files, cache abstractions, and public APIs change by zero; one
-production runtime reorders one existing statement with zero LOC growth. The
-maintained Worker graph remains 907 raw / 477 gzip bytes and the window graph
-remains 14,456 raw bytes with gzip moving from 6,159 to 6,160 bytes; its seven-
-run Linux median was 795.3 ms, recorded without a timing comparison because the
-host load differs. One endurance fixture and browser gate grow, while static
-routes remain zero JavaScript. This packet ships as the actual `0.16.5` patch to
-retain the public `0.16.x` version line.
-
-## `0.18.x`: Lazy Loading And Code Splitting
-
-### `0.18.0`: Route And Feature Artifact Baseline
-
-- **Purpose:** record current route-specific ESM, shared chunks, preload, Worker,
-  CSS, and navigation prefetch behavior.
-- **Expected boundary:** existing RouteBuildRecord and artifact report.
-- **Acceptance:** exact source-to-artifact ownership and bytes for the greenfield
-  app, editor, Answer, and Memos routes.
-- **Stop condition:** a new loader is designed before the current graph is
-  measured.
-- **Done condition:** every eager dependency has a structural reason.
-
-**Result:** closed by the existing RouteBuildRecord and artifact report. A
-test-only collector builds the greenfield project, real CodeMirror editor, four
-separate Apache Answer reductions, and Memos outline reduction: seven fixtures
-and 25 routes total. It joins each route's source page to HTML, runtime entries
-and requirements, handler entries/chunks, Worker entries/chunks, and styles,
-then records exact raw/gzip bytes and SHA-256 values. Every generated JavaScript
-file has at least one route owner, every direct module script has one matching
-`modulepreload`, and every static control owns zero JavaScript and zero preload.
-No measured fixture emits a shared chunk.
-
-The greenfield deploy is 45 files and 196,678 raw / 60,880 aggregate gzip
-bytes. Its project routes own 38,652-69,178 raw JavaScript bytes; `/login` owns
-12,093 raw bytes and `/help` owns zero. The editor route owns 214,968 raw /
-71,607 gzip bytes, including its 200,260 raw / 64,949 gzip CodeMirror handler;
-its static sibling owns zero. The Answer question reduction owns 32,022 raw /
-12,408 gzip bytes. Its route-shell static pages own zero while the two runtime
-parameter pages own 2,134 and 2,078 raw bytes. Answer auth routes own 25,684 and
-26,818 raw bytes with a zero-byte public sibling; authoring/admin routes own
-12,251 and 20,919 raw bytes with a zero-byte legal sibling. The Memos outline
-owns 14,701 raw / 7,044 gzip bytes and its static sibling owns zero.
-
-Current navigation prefetch remains structurally bounded: eligible same-origin
-group links prefetch on hover, focus, viewport proximity, or idle fallback;
-promises deduplicate, failures leave the cache, current/eligible URLs define the
-retained set, and the `0.16.5` revision check rejects stale navigation results
-before cache insertion. The report intentionally records emitted closure rather
-than esbuild input-byte attribution. Answer and Memos remain disconnected
-reductions, not whole-application claims. Semantic primitives, IR kinds,
-compiler passes, production compiler/runtime LOC, runtime concepts/files, and
-public APIs change by zero. No release is consumed.
-
-### `0.18.1`: Bounded Lazy Capability Import
-
-- **Purpose:** permit one statically analyzable feature dependency to load only
-  when its owner requests it.
-- **Expected boundary:** validated static dynamic-import edge, route artifact
-  closure, owner token, and native ESM import.
-- **Acceptance:** exact specifier, no arbitrary graph escape, deduplicated load,
-  error/retry, owner release, and static exclusion.
-- **Stop condition:** arbitrary dynamic imports or lazy component rendering are
-  required.
-- **Done condition:** the feature loader is capability-specific, not a component
-  runtime.
-
-**Completion evidence:** one direct literal bare-package `import()` may remain
-inside an inline synchronous effect when its first statement is the exact
-dependency-state guard `if (!state) return`. The source graph, descriptor
-session, ModuleIR validation, build-module erasure, esbuild split graph, and
-route artifact report preserve that edge without adding a loader. Relative,
-URL, render-time, event, indirect, async-effect, unguarded, and multiple dynamic
-imports remain rejected. Failed native ESM evaluation remains visible through
-application-owned error state; same-document retry of the failed URL is not
-claimed because the browser module map cannot be reset without adding the
-loader forbidden by this packet.
-
-The CodeMirror-derived fixture initially requests no deferred chunk. Activation
-requests one chunk, mounts the editor, and owner release destroys it. Remount
-uses the browser module map without another request, and document disposal
-balances the second mount. The initial eager graph is 12,982 raw / 6,153 gzip
-bytes versus 214,968 raw / 71,607 gzip bytes for the existing eager editor;
-250,086 raw / 80,890 gzip bytes move to the deferred graph. Initial gzip falls
-by 65,454 bytes (91.4%), while the static sibling remains 0 B JavaScript.
-
-The packet adds no semantic primitive, compiler pass, public API, browser
-runtime concept/file, React, VDOM, hydration, retained component tree, or retry
-cache. It adds one validated `ImportIR` kind and additive
-`handlers.lazyChunks` reporting. Production compiler sources change by 76 added
-and 11 removed lines. The artifact collector now covers eight fixtures and 27
-routes with no ownerless JavaScript or preload mismatch. `npm run check`, 284
-required-Chrome tests, package smoke, and the maintained benchmark pass. The
-Worker graph remains 907 raw / 477 gzip bytes; the window graph remains 14,456
-raw / 6,160 gzip bytes, with a seven-build median of 222.8 ms. This packet ships
-as the actual `0.16.6` patch to retain the public `0.16.x` version line.
-
-### `0.18.2`: Lazy Editor Journey
-
-- **Purpose:** apply `0.18.1` to the real editor.
-- **Acceptance:** editor package absent before activation, loaded once on demand,
-  mount/update/dispose behavior unchanged, failure recoverable, and revisit
-  semantics explicit.
-- **Stop condition:** package evaluation must occur during build rendering.
-- **Done condition:** initial route bytes fall by the exact deferred package
-  graph without losing behavior.
-
-**Completion evidence:** a real retained CodeMirror editor now begins closed and
-uses the `0.18.1` guarded literal package edge for its mount. The existing
-effect-private-ref normalization recognizes its one direct `.then()` assignment
-as the retained setup write, lowers the handle to owner-scoped state, and leaves
-later value effects read-only. Indirect callbacks, repeated writes, attached
-refs, update-effect writes, missing null-reset cleanup, and broader dynamic
-imports retain diagnostics. Package evaluation remains entirely in browser
-effect ESM.
-
-Required Chrome records zero chunk requests before activation and one afterward.
-It proves initial content and accessible name, application-to-editor and
-editor-to-application updates, retained DOM identity, authored update failure
-and recovery, exact conditional disposal, fresh `Alpha` state on remount, no
-second package request, and final document disposal. Native ESM evaluation
-failure retry remains outside this packet as recorded in `0.18.1`.
-
-The eager editor owns 214,968 raw / 71,607 gzip bytes. The lazy retained editor
-initially owns 15,896 raw / 7,230 gzip bytes and defers 250,086 raw / 80,890 gzip
-bytes, reducing initial gzip by 64,377 bytes (89.9%). Its static sibling remains
-0 B JavaScript. The artifact collector covers nine fixtures and 29 routes with
-no ownerless JavaScript or preload mismatch.
-
-The packet adds no semantic primitive, IR kind, compiler pass, public API,
-browser runtime concept/file, loader, resource registry, or widget runtime. One
-existing normalization pass changes by 15 added and four removed production
-lines. `npm run check`, 285 required-Chrome tests, package smoke, and the
-maintained benchmark pass. Worker and window graphs remain 907 raw / 477 gzip
-and 14,456 raw / 6,160 gzip bytes; the seven-build median is 230.5 ms. This
-packet ships as the actual `0.16.7` patch to retain the public `0.16.x` line.
-
-### `0.18.3`: Shared Chunks And Prefetch Policy
-
-- **Purpose:** avoid duplicate large feature graphs and define optional preload.
-- **Acceptance:** shared chunk ownership, no duplicate fetch, route/interaction
-  prefetch policy, cache bound, cancellation limits, and output report.
-- **Stop condition:** all features are globally prefetched.
-- **Done condition:** policy minimizes unused transfer while meeting accepted
-  interaction latency.
-
-**Completion evidence:** two enhanced-navigation route owners now use one
-deferred CodeMirror graph. Existing multi-entry esbuild splitting emits one
-250,086 raw / 80,890 gzip byte chunk, and the existing artifact report records
-that exact lazy path as shared by `/` and `/second`. Their eager graphs remain
-21,599 raw / 9,466 gzip and 21,609 raw / 9,468 gzip bytes. Neither route HTML
-preloads the deferred chunk, navigation document prefetch does not request it,
-and the static sibling remains complete HTML with 0 B JavaScript.
-
-Required Chrome proves zero requests before interaction, one request when the
-first owner activates, exact disposal, no route-triggered request, and no second
-request when the second owner activates. The selected policy is therefore no
-optional package preload: interaction calls native `import()` immediately, the
-browser module map deduplicates it for the document lifetime, and Kudzu owns no
-additional cache. Native ESM imports cannot be cancelled after request; owner
-cleanup invalidates the callback and prevents a late mount rather than claiming
-network cancellation.
-
-The released lazy-import gate is also hardened before closing the packet.
-`ownedLazyPackageImport()` now resolves the callback callee through the existing
-lexical binding index and accepts only named `useEffect` imports from React or
-Kudzu. A local function with the same text can no longer bypass the ordinary
-dynamic-import diagnostic. The packet adds no semantic primitive, IR kind,
-compiler pass, public API, runtime concept/file, loader, prefetch registry, or
-cache. Four existing compiler files change by 17 added and eight removed lines.
-The artifact collector covers ten fixtures and 32 routes. `npm run check`, 288
-required-Chrome tests, package smoke, and the maintained benchmark pass. Worker
-and window graphs remain 907 raw / 477 gzip and 14,456 raw / 6,160 gzip bytes;
-the loaded Linux host records a 696.6 ms seven-build median without a timing
-comparison. This packet ships as the actual `0.16.8` patch to retain the public
-`0.16.x` line.
-
-## `0.19.x`: React Ecosystem Migration
-
-### `0.19.0`: Compatibility Boundary And Inventory
-
-- **Purpose:** centralize package recognition and compatibility provenance without
-  changing semantic IR or browser output.
-- **Expected files:** focused compatibility metadata/registry, current passes,
-  project report; no public plugin API.
-- **Acceptance:** Native, Compiled, Normalized, Adapter, Owned External UI,
-  Partial, and Unsupported sites are deterministic and source-located.
-- **Stop condition:** package origin enters ModuleIR, RouteIR, codegen, or runtime.
-- **Done condition:** existing React Router, Bootstrap, Zustand, and browser
-  normalizations preserve output and diagnostics.
-
-Implementation and release-candidate evidence complete. One internal registry
-now owns package identities and classifies original reachable import, re-export,
-and literal dynamic-import sites before normalization. The deterministic
-`.kudzu/kudzu-compatibility.json` report records package/API identity, stable
-rule, Native, Compiled, Normalized, Adapter, Owned External UI, Partial, or
-Unsupported treatment, and exact authored line/column range. Type-only,
-relative, Node, Kudzu-first-party, and unreachable source stays outside the
-inventory.
-
-React Router, React Bootstrap, Zustand, React migration, browser-signal, `clsx`,
-and i18next recognition reuse the registry's package identities. Focused tests
-cover all seven classes, aliases, exact ranges, deterministic ordering, and a
-real build where an unreachable unsupported module is excluded. Existing
-project output baselines and package diagnostics remain green. Package origin
-does not enter SourceResult, ComponentAnalysis, ModuleIR, RouteIR, codegen,
-runtime files, `afterBuild()`, or deploy output.
-
-The slice adds zero semantic primitives, IR kinds, compiler passes,
-normalization entries, runtime concepts/files, browser bytes, or public APIs.
-Production compiler/orchestration changes are 161 added and 13 removed lines,
-including the 141-line registry, for +148 net compatibility/build lines and zero
-core semantic LOC. `npm run check`, 290 required-Chrome tests, package smoke,
-and the maintained benchmark pass. Worker and window graphs remain 907 raw /
-477 gzip and 14,456 raw / 6,160 gzip bytes; seven clean Linux x64 builds record
-a 532.6 ms median without a timing comparison. This packet ships as the actual
-`0.16.9` patch, with `create-kudzu@0.1.134`, to retain the public
-`0.16.x` line.
-
-### `0.19.1`: Apache Answer Authentication Journey
-
-- **Purpose:** promote the existing build-only auth fixture to browser evidence.
-- **Acceptance:** anonymous, invalid login, valid login, token restore, shared
-  header/settings, 401 clear, replacement navigation, public zero-JS sibling.
-- **Stop condition:** client UI is treated as authorization.
-- **Done condition:** the complete journey passes against a deterministic server.
-
-Complete. The existing Apache Answer reduction now runs one required-Chrome
-journey against a deterministic native server. It proves anonymous startup,
-invalid and valid native form submission, token persistence and reload restore,
-layout-owned header/settings reads, a server 401 clearing storage and shared
-state, and replacement navigation back to the anonymous login document. The
-public sibling remains complete HTML with zero JavaScript.
-
-No authored fixture, compiler, IR, runtime, or deploy artifact changed. The
-packet adds one browser harness only: zero semantic primitives, ModuleIR kinds,
-passes, normalization entries, runtime concepts, public APIs, and browser bytes.
-The authentication fixture remains 17 deploy files and 36,441 raw / 15,598 gzip
-bytes; `/` owns 25,684 raw / 11,372 gzip bytes, `/settings` owns 26,818 raw /
-11,319 gzip bytes, and `/public` owns 0 JavaScript bytes. Worker and window
-graphs remain 907 raw / 477 gzip and 14,456 raw / 6,160 gzip bytes; seven clean
-Linux x64 builds record a 674.7 ms median without a timing comparison. This
-packet ships as the actual `0.16.10` patch, with `create-kudzu@0.1.135`, to
-retain the public `0.16.x` line.
-
-### `0.19.2`: Apache Answer Connected Authoring Journey
-
-- **Purpose:** connect query, route, detail, create/edit, validation, mutation,
-  refresh, admin, and logout.
-- **Acceptance:** behavior, accessibility, persistence, errors, navigation,
-  source retention, and output all pass.
-- **Stop condition:** fixture-only compiler semantics or broad package runtime is
-  introduced.
-- **Done condition:** this is a connected browser journey, not a whole-app claim.
-
-Complete. One Apache Answer-derived fixture now connects invalid and valid login,
-layout-shared authentication, page/order query loading, explicit refresh, native
-creation with server validation and retained draft, runtime detail routing,
-editing, reload persistence, keyed administration, delete-and-refetch, and
-logout against one deterministic server. Native controls and status/alert
-regions preserve the accessibility boundary, every retained source file is
-checked, and the public sibling remains complete HTML with zero JavaScript.
-
-The packet reuses existing Zustand normalization, search signals, dependency
-effects, runtime parameters, native handlers, bindings, keyed ownership, and
-enhanced navigation. It adds zero semantic primitives, ModuleIR kinds, compiler
-passes or core LOC, runtime concepts, package adapters, and public APIs. The new
-fixture emits 33 deploy files and 88,198 raw / 33,173 gzip bytes; `/` owns 27,196
-raw / 11,968 gzip bytes, `/questions` owns 51,577 raw / 19,801 gzip bytes,
-`/questions/add` owns 27,109 raw / 11,893 gzip bytes, runtime detail owns 32,783
-raw / 14,056 gzip bytes, admin owns 50,861 raw / 19,426 gzip bytes, and `/public`
-owns zero JavaScript. Worker and window graphs remain 907 raw / 477 gzip and
-14,456 raw / 6,160 gzip bytes; seven clean Linux x64 builds record a 640.5 ms
-median without a timing comparison. No production logic changed, so the public
-version remains `0.16.10` and no release is consumed.
-
-### `0.19.3`: Memos Feed And CRUD Journey
-
-- **Purpose:** establish the first durable Memos application slice.
-- **Acceptance:** login, loading/error, feed, pagination, create/edit/delete,
-  retained memo identity, refresh, and logout.
-- **Stop condition:** outline-only or static feed output is presented as Memos
-  migration.
-- **Done condition:** pinned upstream provenance and exact retained source are
-  recorded.
-
-Complete. A fixture pinned to Memos revision
-`e5ed6e7ec60e141d6e354f3ab59906c2c47dccac` now runs invalid and valid login,
-initial loading, a server pagination failure, explicit recovery, overlapping-key
-pagination, create, edit, delete, reload persistence, and logout against one
-deterministic server. The retained memo row keeps DOM identity through the
-failed request, successful pagination, and every CRUD refetch. Native labels,
-status and alert regions preserve the accessibility boundary, every authored
-source file is retained, and the public sibling emits zero JavaScript.
-
-The packet reuses existing Zustand normalization, dependency effects, keyed
-ownership, native forms and handlers, and enhanced navigation. It adds zero
-semantic primitives, ModuleIR kinds, compiler passes or core LOC, runtime
-concepts, package adapters, and public APIs. The fixture emits 19 deploy files
-and 56,167 raw / 21,634 gzip bytes; `/` owns 25,856 raw / 11,396 gzip bytes,
-`/feed` owns 45,690 raw / 17,836 gzip bytes, and `/public` owns zero JavaScript.
-Worker and window graphs remain 907 raw / 477 gzip and 14,456 raw / 6,160 gzip
-bytes; seven clean Linux x64 builds record a 761.8 ms median without a timing
-comparison. No production logic changed, so the public version remains
-`0.16.10` and no release is consumed.
-
-### `0.19.4`: Memos Realtime Journey
-
-- **Purpose:** add reaction and realtime refresh using the accepted long-lived
-  transport model.
-- **Acceptance:** mutation, server event, deduplication, reconnect, stale event
-  rejection, route cleanup, and memory gate.
-- **Stop condition:** a query or SSE package runtime is copied wholesale.
-- **Done condition:** browser and endurance acceptance pass.
-
-Complete. The connected Memos fixture now adds a native reaction mutation and
-an effect-owned WebSocket that refreshes the keyed feed from versioned server
-events. Required Chrome proves duplicate and stale version rejection, old-socket
-callback rejection after reconnect, retained memo DOM identity, exact route
-cleanup, and fresh ownership after re-entry and reload. Active sockets,
-listeners, and reconnect timers remain bounded and return to zero after each
-owner release. The public sibling remains complete HTML with zero JavaScript.
-
-The packet reuses existing native handlers, keyed ownership, effect-private
-lifecycle, invalidation, timers, and route release. It adds zero semantic
-primitives, ModuleIR kinds, compiler passes or core LOC, runtime concepts,
-normalization or adapter rules, and public APIs. The fixture emits 19 deploy
-files and 57,811 raw / 22,069 gzip bytes; `/` owns 25,856 raw / 11,396 gzip
-bytes, `/feed` owns 46,740 raw / 18,184 gzip bytes, and `/public` owns zero
-JavaScript. Worker and window graphs remain 907 raw / 477 gzip and 14,456 raw /
-6,160 gzip bytes; seven clean Linux x64 builds record a 620.7 ms median without
-a timing comparison. Browser-disabled tests pass 292/292, required-Chrome
-focused acceptance, package smoke, and artifact accounting pass. No production
-logic changed, so the public version remains `0.16.10` and no release is
-consumed.
-
-### `0.19.5`: Actual Budget Intake Decision
-
-- **Purpose:** determine whether a complete, legally and technically usable
-  workspace can become the third application proof.
-- **Acceptance:** pinned source, install/build path, dependency inventory, first
-  executable user journey, and honest source-retention denominator.
-- **Stop condition:** workspace dependencies or core packages remain absent.
-- **Done condition:** a later packet is authorized, or deferral is recorded with
-  no compatibility claim.
-
-Deferred by the stop condition. The pinned MIT revision
-`87e33e49eb4d666c33e0c0f172faf340d1b7aa0a` remains only a sparse desktop-client
-acquisition: 810 TS/TSX modules, 163,344 LOC, and 80 route declarations without
-the workspace tooling, component library, core backend, Yarn release, database,
-worker coordinator, or spreadsheet implementation needed to install and run an
-honest user journey. The prior static report probe is an exported report, not
-Actual Budget, and its 22/906 retained-line result is not a whole-workspace
-denominator. No later application packet, compatibility claim, compiler work,
-runtime, package adapter, public API, version, or release is authorized. Intake
-may resume only from a complete pinned workspace and executable dependency
-graph.
-
-## `0.20.x`: AI Tooling And Delivery Cost
-
-### `0.20.0`: Structured Diagnostics
-
-- **Purpose:** expose stable machine-readable errors from the existing compiler.
-- **Acceptance:** version, code, stage, severity, source range, message,
-  compatibility class, and safe suggestion; human diagnostics remain useful.
-- **Stop condition:** diagnostic codes expose pass filenames or unstable IR.
-- **Done condition:** agents no longer parse message substrings for maintained
-  fixtures.
-
-Complete. Authored
-source failures now propagate as one versioned diagnostic envelope with stable
-semantic code, stage, severity, project-relative source range, human message,
-nullable compatibility class, and nullable safe suggestion. `kudzu build
---json` emits only that deterministic envelope for structured failures, while
-ordinary builds and the development overlay retain readable source-located
-messages. Unresolved imports and re-exports, unsupported dynamic imports,
-TypeScript syntax errors, runtime-locale `react-i18next`, and keyed-row ref
-initializers prove the maintained boundary without parsing message text.
-
-The implementation centralizes 72 lines in `diagnostics.mjs`, reuses the
-existing compatibility classification function, and normalizes paths only at
-the project build boundary. It adds zero semantic primitives, ModuleIR kinds,
-compiler passes, normalization entries, runtime concepts, browser modules, or
-public application APIs. Production source is +111 net lines, including +8 in
-the measured core semantic file set; the increase is compiler/tooling error
-transport rather than semantic analysis. All representative deploy manifests
-and hashes remain unchanged. Worker and window graphs remain 907 raw / 477 gzip
-and 14,456 raw / 6,160 gzip bytes; seven clean Linux x64 release-candidate builds
-record a 1,034.4 ms median without a timing comparison. `npm run check`, 293/293
-browser-disabled and required-Chrome tests, focused CLI JSON tests, development error recovery, package smoke,
-artifact accounting, and `git diff --check` pass. The packet ships as
-`@kudzujs/core@0.16.11` with `create-kudzu@0.1.136`.
-
-### `0.20.1`: `kudzu inspect --json`
-
-- **Purpose:** expose reachable application and compatibility inventory.
-- **Acceptance:** filtered module/route/package/capability/owner/blocker facts,
-  deterministic order, schema version, and large-project context bound.
-- **Stop condition:** the command builds a second analyzer or dumps full raw IR.
-- **Done condition:** the first blocker can be selected without reading hundreds
-  of source or dependency files.
-
-Complete. `kudzu inspect --json` now runs the authoritative build and projects
-the existing reachable source, compatibility, ComponentAnalysis, ModuleIR,
-CapabilityIR, and route-artifact records into one versioned report. Modules,
-routes, packages, compatibility sites, capability families, semantic owners,
-and blockers sort before fixed section limits; totals and omitted counts keep
-large-project context explicit. Unreachable source, generated code, HTML, raw
-IR, captures, state values, and full artifact closures remain excluded.
-
-One real CLI fixture proves reachable filtering, static and interactive route
-facts, native package classification, state ownership, project-relative paths,
-deterministic output, and structured blocked inventory. A synthetic 101-module
-and 51-blocker check proves sort-before-truncation and first-blocker retention.
-The implementation is one 154-line projection over existing records plus the
-existing build/CLI seam; it adds no analyzer, semantic primitive, compiler pass,
-normalization entry, runtime concept, or browser module. Representative deploy
-manifests and hashes remain unchanged. Worker and window graphs remain 907 raw /
-477 gzip and 14,456 raw / 6,160 gzip bytes; seven clean builds record a 648.7 ms
-median without a timing comparison. Browser-disabled and required-Chrome suites
-pass 295/295 tests, package smoke and artifact accounting pass, and the packet
-ships as `@kudzujs/core@0.16.12` with `create-kudzu@0.1.137`.
-
-### `0.20.2`: `kudzu explain --route`
-
-- **Purpose:** trace one authored route to its selected browser artifacts.
-- **Acceptance:** source site, normalization provenance, semantic owner,
-  capability, entry/chunk/style/Worker, byte reason, and zero-JS explanation.
-- **Stop condition:** a second artifact graph is introduced.
-- **Done condition:** existing route records and artifact reports answer the
-  query with bounded output.
-
-Completed by `@kudzujs/core@0.16.13` and `create-kudzu@0.1.138`. The
-`worker-effects` fixture proves exact configured-base lookup for an effect-owned
-Worker route, authored source closure, `Dashboard` ownership, capability family,
-runtime/handler/Worker byte reasons, raw/gzip totals, hashes, deterministic
-output, structured missing-route diagnostics, and a static sibling with an empty
-JavaScript closure. The implementation is one 137-line bounded projection over
-existing source results, compatibility sites, route records, and the sole route
-artifact report plus the existing build/CLI seam. It adds no analyzer, semantic
-primitive, compiler pass, normalization entry, runtime concept, or browser
-module. Representative deploy manifests and hashes remain unchanged. Worker and
-window graphs remain 907 raw / 477 gzip and 14,456 raw / 6,160 gzip bytes; seven
-clean builds record a 605 ms median without a timing comparison. Browser-disabled
-and required-Chrome suites pass 296/296 tests, and package smoke and artifact
-accounting pass. Fixed 100-record section limits expose totals and omitted counts;
-the command requires one exact emitted route and intentionally excludes raw IR,
-HTML, generated code, captures, and state values.
-
-### `0.20.3`: Deterministic Normalize And Fix
-
-- **Purpose:** automate only edits proven safe and useful by recorded AI trials.
-- **Acceptance:** check/preview/write modes, exact ranges, idempotence, comments
-  retained, no unrelated formatting, and byte-equivalent output.
-- **Stop condition:** canonical compiler source replaces ordinary authored TSX.
-- **Done condition:** each fix has one diagnostic code and explicit preconditions.
-
-Deferred by the evidence gate without production changes or a release. A complete
-repository inventory found no recorded model prompt, model/tool version, raw
-attempt, correction cycle, token/cost trace, or repeated source edit proving any
-automatic fix useful. Existing diagnostics offer migration guidance but do not
-identify a uniquely safe edit; replacing a non-null ref initializer would change
-authored behavior. The existing lazy-state-literal normalization could support a
-narrow exact-range rewrite while preserving comments and output bytes, but no AI
-trial records it as a correction or cost reduction. Adding check/preview/write
-infrastructure now would therefore be speculative and violate the packet purpose.
-Resume this packet only after `0.20.4` records a repeated correction with one
-stable diagnostic code, a unique semantics-preserving replacement, and
-compiler-proven preconditions. No fixture, analyzer, fix command, formatter,
-compiler pass, runtime concept, browser module, package version, or release was
-added.
-
-### `0.20.4`: AI Delivery Protocol And Runner
-
-- **Purpose:** track fair Kudzu and React+Vite delivery attempts.
-- **Acceptance:** pinned model/tools/prompts/budgets, framework-neutral behavior
-  suite, raw failures, token/cost/tool/file/build metrics, source retention, and
-  browser artifacts.
-- **Stop condition:** Kudzu receives private compiler guidance unavailable to the
-  comparator.
-- **Done condition:** every attempt is reproducible or fully attributable.
-
-Completed by `@kudzujs/core@0.16.14` and `create-kudzu@0.1.139`. The
-repository-owned Node standard-library runner pins the model revision, pricing,
-tools, prompt, budgets, adapter, acceptance command, starter digests, optional
-public context, and paired schedule before creating isolated workspaces. It
-retains raw adapter/build/acceptance streams, final source, browser artifacts,
-hashes, attribution, token/cost/tool/file/build metrics, source retention, and
-both successful and failed attempts. The deterministic two-attempt-per-variant
-fixture proves equal ordering, one retained failure per variant, independent
-final builds, framework-neutral acceptance, and failure-inclusive cost per
-success. It is protocol evidence only and makes no Kudzu versus React+Vite
-delivery claim. No dependency, compiler pass, semantic primitive, runtime
-concept, application API, or browser byte was added. Browser-disabled and
-required-Chrome suites pass 297/297 tests, and package smoke passes.
-
-### `0.20.5`: Tooling Cost Validation
-
-- **Purpose:** prove that diagnostics/inspect/explain/fix lower total cost per
-  successful task.
-- **Acceptance:** before/after attempts on the same tasks; failures remain in the
-  numerator; success behavior remains identical.
-- **Stop condition:** token reduction comes from omitted behavior or hidden
-  context.
-- **Done condition:** useful tools are retained and non-improving tools are
-  removed or deferred.
-
-Measured on 2026-09-01 with five interleaved attempts per condition against one
-Apache Answer-derived static Bootstrap correction. Both conditions passed 5/5
-with identical accepted behavior, source retention, and browser artifacts, but
-plain-build baseline cost 67,860 tokens per success while structured
-build/inspect/explain cost 84,793. The structured bundle is therefore deferred
-as an AI default rather than credited with a cost reduction. Existing JSON,
-inspection, and explanation commands remain for their independently accepted
-machine-readable contracts; no speculative `fix` command, compiler pass,
-runtime concept, application API, dependency, or browser byte was added. Exact
-samples, medians, and ranges are recorded in `PERFORMANCE.md`, while the pinned
-protocol reproduces and retains raw attempt evidence under the ignored
-`test-results/ai-delivery/` path.
-
-## `0.21.x`: Production-Scale Proof
-
-### `0.21.0`: Functional Parity Matrix
-
-- **Purpose:** freeze complete greenfield, Answer, and Memos acceptance.
-- **Acceptance:** content, auth, forms, CRUD, shared data, large list, overlay,
-  editor, lazy load, realtime, errors, accessibility, and navigation pass.
-- **Stop condition:** any build-only result is counted as application success.
-- **Done condition:** every claimed capability has a browser journey and static
-  exclusion control.
-
-**Result:** closed through existing application semantics without a release.
-The required-Chrome 300-test matrix passes content, authentication, forms, CRUD,
-shared data, large lists, overlays, editor ownership, lazy loading, realtime,
-errors, accessibility, and navigation across the maintained greenfield, Apache
-Answer, Memos, and focused package fixtures. The one material matrix gap was the
-10,000-item pagination decision: its release/remount behavior existed only in an
-optional benchmark. One ordinary browser acceptance now proves Enter-key page
-advance, focus retention, the 101-200 and restored 1-100 ranges, bounded 100-row
-DOM, fresh off-range input state, and no browser exceptions. Greenfield `/help`,
-Answer and Memos `/public`, package `/static` routes, and the complete direct
-10,000-row route provide zero-JavaScript controls.
-
-The artifact inventory now includes the large-list fixture: 19 files,
-1,900,048 raw / 242,968 aggregate gzip bytes, deploy SHA-256
-`1bbaf3479ac90f923deb469c73eb93c0dd7364e7fe906919b167f2f854deda70`.
-Pagination owns 30,143 raw / 11,193 gzip bytes of existing JavaScript while the
-direct static route owns zero. Semantic primitives, IR kinds, compiler passes,
-production compiler/runtime LOC, normalization rules, adapters, runtime
-concepts, public APIs, and browser artifacts change by zero. `npm run check`,
-required-Chrome 300/300, package smoke, focused acceptance, and artifact
-inventory pass. No production feature changed, so no release is consumed.
-
-### `0.21.1`: Compiler And Route Scale
-
-- **Purpose:** measure clean/incremental compilation and artifact emission.
-- **Acceptance:** 100, 1,000, and separately provisioned 10,000 routes; module
-  graph time, parse/normalize/compile/render/write time, peak RSS, output digest,
-  and failure recovery.
-- **Stop condition:** projected or incomplete 10,000-route data is called a pass.
-- **Done condition:** repeated measurements and all material regressions are
-  explained or fixed.
-
-**Result:** opt-in internal telemetry records source reads, reachable
-graph discovery, TypeScript parse, normalize, compile/transform, render, and
-write boundaries without changing deploy output. Compile/transform excludes
-the measured TypeScript parse and Kudzu normalization but includes indivisible
-esbuild parse/transform for proven import-free modules. Render includes
-provisional page materialization, write includes finalization and staged-output
-promotion, and the end-to-end clean build is not their sum. The runner also proves that an
-incremental source edit matches a clean build of the changed source, compares
-paired incremental outputs when a baseline is supplied, names retained-session
-rather than incremental-only peak RSS, and fails the process when either clean
-or retained-session invalid-source recovery changes output. This adds no
-semantic primitive, compiler pass, runtime concept, public API, deploy artifact,
-or browser byte; production compiler/build instrumentation is 35 added and four
-removed lines.
-
-One warm-up and three fresh-process samples pass at 100 routes / 1,000 modules /
-101,100 lines and 1,000 routes / 10,000 modules / 1,011,000 lines. Clean-build
-medians are 7,444.3 and 77,515.9 ms with 672.3 and 3,182.3 MiB peak RSS;
-incremental medians are 1,838.0 and 18,132.1 ms, recompiling exactly 10 modules
-and rendering one page in every sample. Clean and changed digests differ,
-incremental output equals a clean changed-source build, and clean and retained
-failure probes preserve then recover the expected digest. Exact phase samples,
-bytes, and digests are recorded in `PERFORMANCE.md`.
-
-The full 10,000-route / 100,000-module / 10,110,000-line topology initially
-exhausted a 10,240 MiB V8 heap because the canonical module cache retained about
-90,000 complete TypeScript ASTs. A 1,024-record LRU bounds that existing cache
-without changing stable position-based symbols. The identical topology then
-completed under the same heap cap with 5,461.0 MiB clean-build RSS and 8,266.9
-MiB retained-session peak RSS. It emits 10,000 files / 2,237,780 B with clean
-digest `7a48bc40ff33c4faaef8a01ab90fcfe611a4bada5364a6fedca42f80ad5070d3`;
-incremental output matches a clean changed-source build, recompiles 10 modules,
-renders one page, and clean and retained recovery pass. One complete measured
-run is retained after two diagnostic runs completed measured stages but exposed
-cleanup and timeout defects; no projection or reduced topology is counted.
-The memory fix adds eight net production lines, no semantic primitive, pass,
-runtime concept, public API, deploy artifact, or browser byte.
-
-### `0.21.2`: Browser Performance And Memory
-
-- **Purpose:** protect interaction latency and long-running stability.
-- **Acceptance:** initial/session bytes, list/range latency, navigation, lazy
-  feature load, resource cleanup, repeated heap, detached DOM, cache, listeners,
-  and handles.
-- **Stop condition:** smaller JavaScript is preferred over behavior parity.
-- **Done condition:** complete features meet declared thresholds with bounded
-  long-session growth.
-
-**Result:** closed through existing route artifacts, keyed ownership, enhanced
-navigation, native ESM loading, and the released endurance harness. Seven fresh
-Chrome profiles keep 2,000-row append, filter, restore, and reverse medians at
-9.6, 17.6, 84.5, and 19.4 ms against a 100 ms median alarm. Project table save
-and list-to-detail navigation medians are 0.9 and 4.9 ms against the same alarm;
-the exact session graph owns 17 JavaScript files totaling 81,200 raw / 27,885
-aggregate gzip bytes. The static 10,000-row document loads in a 999.8 ms median
-against a 2,000 ms alarm. Pagination and windowing keep the DOM at 100 rows,
-load in 163.1 and 251.7 ms against a 1,000 ms alarm, and change range in 18.1
-and 23.7 ms against a 100 ms alarm.
-
-The lazy editor retains 12,982 raw / 6,153 gzip eager bytes and one 250,086 raw
-/ 80,890 gzip deferred CodeMirror chunk; two route owners share one native ESM
-module request and the static sibling remains zero JavaScript. Focused Chrome
-acceptance passes 3/3. The endurance gate passes five warm-ups, ten stale-
-prefetch races, 30 ownership cycles, and 101 navigations: editor mounts and
-disposals balance 71/71, browser state stays zero, documents and listeners stay
-at six and seven, DOM nodes move by one, and forced-GC heap grows 79,988 B under
-the existing 2 MiB or 15% alarm. No exception or failed request occurs. Three
-existing benchmark entry points now fail when their declared median budgets are
-exceeded; the artifact collector may name the current packet. Production source,
-semantic primitives, compiler passes, runtime concepts, APIs, deploy artifacts,
-and browser bytes change by zero. Exact samples and limitations are recorded in
-`PERFORMANCE.md`; no package release is consumed.
-
-### `0.21.3`: Package And Release Candidate
-
-- **Purpose:** verify the exact candidate as an installed product.
-- **Acceptance:** `npm run check`, `npm test`, required Chrome, focused
-  benchmarks, package smoke, tarball inspection, generated application, registry
-  dry run, docs, and release rollback procedure.
-- **Stop condition:** version metadata changes before the exact candidate passes.
-- **Done condition:** the reviewed commit is independently publishable.
-
-**Result:** released as `@kudzujs/core@0.16.17` and
-`create-kudzu@0.1.142`. The exact versioned tree passes `npm run check`, all 301
-remaining tests after the application contract, package smoke, generator
-creation, and both registry dry runs. Tarball inspection records the expected 85
-core files and four generator files with no bundled dependencies. The Node 22
-no-browser lane initially attempted the list acceptance despite
-`KUDZU_SKIP_BROWSER`; that test now follows the existing skip contract while the
-Node 24 required-Chrome lane retains the browser acceptance.
-
-Seven fresh exact-candidate Chrome profiles pass every maintained alarm. The
-project table-save/navigation medians are 0.7/3.6 ms; direct, pagination, and
-window load medians are 849.2, 166.6, and 184.5 ms; bounded range medians are
-13.0 and 19.7 ms. The keyed benchmark also passes all four 100 ms alarms. A
-fresh endurance run balances editor mounts/disposals at 71/71 across 101
-navigations, with zero document/listener growth, one additional DOM node, and
-144,172 B forced-GC heap growth under the 2 MiB alarm. Package contents preserve
-the existing production compiler/runtime and introduce no semantic primitive,
-pass, runtime concept, API, deploy artifact, or browser byte. The immutable
-rollback contract remains forward-only: a failed push, tag, publication, or
-registry verification keeps this packet active; a published npm version or tag
-is never moved and requires a forward-fix patch.
-
-Commit `e9b33c04a7a2a382b858069910e4473800e7b70a`, tag `v0.16.17`, the
-[GitHub release](https://github.com/kudzujs/kudzu/releases/tag/v0.16.17), and
-[publish workflow](https://github.com/kudzujs/kudzu/actions/runs/33704726784)
-are complete. npm exposes core `0.16.17` and generator `0.1.142`; a fresh
-registry-generated application resolves core `0.16.17` and passes its complete
-TypeScript and Kudzu build check.
-
-### `0.21.4`: AI Delivery Proof
-
-- **Purpose:** run the final equal-condition comparison on production-shaped
-  tasks.
-- **Acceptance:** highest or statistically tied success rate, lowest median cost
-  per successful task, complete raw attempts, maintainable source, and retained
-  browser/build advantage.
-- **Stop condition:** cherry-picked attempts, toy-only tasks, or unequal tooling.
-- **Done condition:** the evidence either authorizes `1.0.0` or names the next
-  measured blocker without changing the release criteria.
-
-The first independently reviewable evidence-integrity prerequisite is released
-as `@kudzujs/core@0.16.18`. A successful adapter result without its raw
-attribution trace previously remained eligible for the success-rate and median
-successful-cost denominators with synthetic zero metrics. The runner now
-requires the trace in the success predicate, retains that attempt as an
-incomplete failure, and withholds unsupported per-success claims. The focused
-three-test harness passes and the change adds one condition with no semantic
-primitive, compiler pass, runtime concept, application API, dependency, or
-application browser byte.
-
-This prerequisite makes no framework ranking. Existing complete fixtures cover
-the content, forms, CRUD/shared-state, and resource/realtime task classes. The
-pinned public `kudzu-based-bench` storefront at commit
-`f2d5be1a516c539e30f7125f6870d42b1dd02ecd` remains the smallest complete paired
-commerce source. The next exact action is to freeze equivalent pre-task starters
-and complete acceptance journeys for those five classes, then run at least five
-interleaved attempts per framework and task under one model, tool, context, and
-time contract.
-
-The corrected production comparison was recorded in GitHub release `v0.16.19`.
-Its npm workflow stopped before publication on an unrelated flaky project browser
-journey. The `v0.16.20` forward release passed every repository gate, but npm
-rejected its trusted-publishing signing certificate with HTTP 403. The immutable
-recovery release `@kudzujs/core@0.16.21` carries the unchanged evidence and
-`create-kudzu@0.1.146` targets it.
-The first batch is retained but excluded from ranking after hidden content/form
-selector assumptions and inconsistent React versions were found. Revision 2
-hashes the written contracts and executable acceptance, uses React 19.2.8 in all
-five task classes, and records Kudzu 11/25 attributable successes (44%) against
-React + Vite 24/25 (96%). One React realtime adapter exceeded 300 seconds without
-a trace and correctly remains incomplete; it was not selectively replaced.
-
-The gate fails: Kudzu has no successful task result for content or commerce, so
-its aggregate task-token median is unavailable. React's incomplete realtime
-attempt also leaves its aggregate failure-inclusive cost unavailable. Kudzu keeps lower accepted browser artifacts and zero-JavaScript static
-siblings, but does not meet the required success-rate or cost criteria. The raw
-invalid and corrected attempts are published as
-`kudzu-ai-delivery-0.21.4-gpt-5.6-sol.tar.gz`, SHA-256
-`4e37d15b4c4b0cf88720a3a7743996794b2e0172dbc43d40514f944fb01bd0f8`.
-The next evidence-selected work must address reactive collection normalization
-and count reuse, then reactive text inside selected conditional branches.
-`1.0.0` remains unauthorized. This packet adds no semantic primitive, compiler
-pass, runtime concept, public API, dependency, generated behavior, or browser
-byte.
-
-The first evidence-selected follow-up now normalizes one supported collection
-alias across a keyed map, a top-level `.length` count, reactive text, and a
-count-selected JSX branch. It reuses collection analysis plus existing binding,
-conditional, and keyed ownership: no semantic primitive, compiler pass, runtime
-concept, API, dependency, or shared-runtime byte was added. The implementation
-changes two existing compiler files by net 37 lines and extends one positive and
-one negative fixture. Required Chrome verifies count updates, empty-branch
-unmount, keyed remount, and retained source diagnostics; all 302 tests, package
-smoke, and `npm run check` pass. The exercised route ships the unchanged 36,219
-B shared runtime closure and 37,870 B total JavaScript including its existing
-route handlers. No benchmark contract or budget changes; the production AI
-comparison must be rerun only after the remaining selected-branch reactive-text
-blocker is addressed. This follow-up is released as `@kudzujs/core@0.16.22`
-with `create-kudzu@0.1.147`. `1.0.0` remains unauthorized.
-
-The second evidence-selected follow-up now lowers reactive scalar text inside a
-selected conditional branch through the existing binding descriptor and
-conditional ownership paths. It adds no semantic primitive, compiler pass,
-runtime concept, API, dependency, or shared-runtime source byte; one production
-compiler file grows by net 11 lines. A focused fixture proves same-branch text
-updates and branch replacement, and required Chrome, all 302 tests, package
-smoke, and `npm run check` pass. The five retained commerce sources now pass
-their complete acceptance with `empty`, `below`, and `reached` all true and a
-zero-JavaScript static sibling. Their route-specific checkout handlers add
-147-154 transferred bytes; no runtime module or benchmark contract changes.
-This does not revise the published equal-condition score without a complete
-rerun, so `1.0.0` remains unauthorized. This follow-up is released as
-`@kudzujs/core@0.16.23` with `create-kudzu@0.1.148`.
-
-The user-approved 2026-09-07 `0.16.24` continuation addresses the initial r3
-content patch's imported collection alias and direct filter-count failures.
-[`imported-article-search`](../../test/fixtures/imported-article-search/README.md)
-reproduced both failures before changes. The reduction extends existing import
-discovery, pure selector normalization, and count build-value lowering, with
-existing BindingIR/DerivedIR/KeyedBlockIR consumers and route/keyed/conditional
-ownership. It adds no primitive, IR kind, pass, runtime concept, API, or adapter;
-the two core files grow by net 39 lines. Three source variants produce identical
-JavaScript (32,190 raw / 12,324 gzip B) and six unsafe neighboring forms remain
-source-located errors. Required Chrome verifies search/count/empty transitions,
-retained keys, fresh remounts, and static zero-JavaScript exclusion. Check, the
-project ownership gate plus 308 tests, package smoke, and both package dry-runs pass.
-
-The full initial-source replay passes the unchanged content acceptance and emits
-the same 13 JavaScript files/bytes as the earlier retained inlined workaround
-(55,447 raw / 20,976 gzip B). This is a source replay, not a fresh model trial or
-replacement benchmark score; no timing or AI-cost improvement is claimed.
-[`PERFORMANCE.md`](../../PERFORMANCE.md#imported-search-source-replay-2026-09-07)
-records commands, provenance, accounting, and limits. Preserve the original r3
-evidence. The release also includes the reviewed r4 protocol/adapter lifecycle
-work: atomic partial traces, bounded incremental output, setup-inclusive deadlines,
-owned process-tree termination, and incomplete timeout attribution. Two Linux
-lifecycle regressions protect future-run integrity; Windows termination remains
-untested. The frozen r4 starters stay on published `0.16.23`; old raw records are
-not reinterpreted. This packet ships as `@kudzujs/core@0.16.24` with
-`create-kudzu@0.1.149`, without authorizing `1.0.0` or a new AI performance claim.
-
-The post-release 2026-09-07 r5 rerun now executes all 50 predeclared attempts with
-published `0.16.24` core pins, registry lock integrity, and canonical hashes in
-all five Kudzu starters. Revision 5 preserves the release-documented r4 pins as
-history rather than relabeling them; model `openai/gpt-5.6-sol`, OpenCode 1.18.27,
-budgets, prompts, public context, acceptance, React starters, schedules, and the
-r4 lifecycle adapter are unchanged. No compiler or runtime change is made.
-
-The frozen scorer records Kudzu 18/25 versus React + Vite 23/25, with per-task
-counts 3/5, 5/5, 3/5, 5/5, 2/5 versus 5/5, 5/5, 3/5, 5/5, 5/5. All tasks finish,
-but two Kudzu realtime deadline failures retain partial attribution, so suite
-status and Kudzu aggregate token cost remain incomplete. React's failure-inclusive
-task-token median is 81,111. Historical r2's published 11/25 versus 24/25 and all
-r3 raw evidence remain unchanged; no selective retries or replacement scores.
-
-This is not a clean ranking or `1.0.0` authorization. Four CRUD failures are
-exact accessible-name mismatches against a requirement absent from the supplied
-prompt; two content failures arise from the grader reading count and empty copy
-as one live-region string, one also exceeding the input budget. The next
-prerequisite is acceptance/prompt alignment review without retroactive rescoring.
-Real compiler intake remains in raw `query.trim` failures for direct count uses
-of normalized-query filters and realtime refs intended to survive pause/resume.
-One null-ref workaround builds but fails the first snapshot; two zero-ref
-attempts fail cleanup proof and time out. Accepted source retains declarative
-composition but some Kudzu attempts add derived/version state workarounds.
-Designated static siblings remain script-free; content's unchanged topic source
-emits list scripts, leaving the broader static-route contract unproven.
-
-Exact commands, nine-failure census, hashes, artifact medians, and scope limits:
-[`PERFORMANCE.md`](../../PERFORMANCE.md#0214-released-01624-rerun-2026-09-07).
-Local raw evidence is
-`test-results/ai-delivery-production/0.21.4-gpt-5.6-sol-r5-kudzu-0.16.24-20260907-01/`.
-Six integrity/lifecycle tests, `npm run check`, and the standalone ownership gate
-plus 308 tests pass. No commit, version bump, publication, or release is part of
-this continuation.
-
-### r5 Compiler Follow-Up
-
-The evidence-backed follow-up is implemented without changing the dirty r5/r6
-protocol, prompt, contract, or acceptance work. The two content reductions were
-confirmed red before compiler changes: direct JSX `.length` over a normalized
-query filter threw `query.trim is not a function`, and the unchanged static topic
-filter emitted list state/scripts. Existing count dependency registration now
-selects signal build values, while direct imported pipelines without selector
-states or collection aliases stay in ordinary build-time map execution.
-BindingIR, DerivedIR, KeyedBlockIR, and route/keyed/conditional ownership remain
-unchanged; reactive selectors retain the existing list path and negative checks.
-
-Realtime source tracing closes the proposed lifetime extension negatively.
-Attempts 0/3 correctly reject version refs without cleanup reset; moving them
-into invocation-private closures would lose the version on pause/resume. Attempt
-1 exposes a separate diagnostic gap: render-written null refs were accepted as
-DOM refs, which serialize an ID rather than the retained numeric value. Existing
-ref normalization now rejects those writes at source. No cross-invocation ref,
-resource primitive, runtime, or arbitrary callback support was added. Original
-state-based attempts 2/4 still pass the realtime browser journey.
-
-Accounting: semantic primitives, IR kinds, core passes, normalization entries,
-transform/adapter rules, runtime concepts/source, and APIs all +0; core semantic
-LOC +5, focused normalization LOC +2, one diagnostic guard, two positive and two
-negative reduced cases. The four search variants emit identical JavaScript at
-32,190 raw / 12,324 gzip B. All five final content sources plus the recovered
-first attempt-1 patch pass the prepared r6 acceptance, including every one of the
-ten static siblings and combined polite regions. Each deploy removes four topic
-JavaScript files totaling 20,539 raw / 7,299 gzip B, including 20,462 / 7,204 B of
-runtime files; every retained JavaScript file is byte-identical to r5.
-
-`npm run check`, the standalone ownership gate plus 315 required-Chrome tests,
-and package smoke pass with 1,200,000 ms command timeouts. Exact source locations,
-replay hashes, commands, output tables, diagnostic boundaries, and local evidence
-are in [`PERFORMANCE.md`](../../PERFORMANCE.md#r5-compiler-follow-up-2026-09-07).
-This is source replay, not AI rescoring: r5 remains 18/25 versus 23/25 with partial
-traces, and published r2 remains 11/25 versus 24/25. No model rerun, timing claim,
-commit, release, or `1.0.0` authorization follows. The next benchmark decision
-must use the aligned r6 protocol in a separately authorized complete run, not
-selective retries or broader compiler semantics without new evidence.
-
-### 0.16.25 Release Transaction
-
-The user-authorized patch packages the preceding r5 records, compiler fixes,
-render-ref diagnostic, and future-only r6 acceptance alignment as core 0.16.25
-and create-kudzu 0.1.150. The pre-release session records above remain historical.
-R6 accepts semantically named CRUD groups and combined polite count/empty regions,
-and checks all ten content static siblings. Its hashes and published 0.16.24
-starter pins remain frozen; testing 0.16.25 in a model run requires a new frozen
-revision. No fresh model run, score replacement, persistent-ref support, or
-`1.0.0` authorization is part of this release.
-
-## `1.0.0`: Stable Application Model
-
-`1.0.0` is authorized only after every required `0.21.x` gate passes. It means:
-
-- the small package-neutral application semantic model is stable;
-- complete production-shaped application journeys pass;
-- static routes retain zero JavaScript;
-- interactive routes retain only used capabilities;
-- React migration uses compatibility, normalization, adapters, and owned
-  external UI rather than a React runtime;
-- browser/build/performance and long-running memory gates pass;
-- the equal-condition AI benchmark meets the published success and cost gates.
-
-## Patch Session Procedure
-
-Before implementation:
-
-1. Read this plan, `MIGRATION_ROADMAP.md`, the current architecture map, and the
-   applicable capability packet.
-2. Confirm the package version and inspect uncommitted work without modifying
-   unrelated changes.
-3. Confirm every earlier patch is complete or explicitly replanned.
-4. Name the exact application journey, failing fixture, owner, producer, and
-   consumer boundary.
-5. Record the baseline output, browser bytes, runtime concepts, compiler/pass
-   LOC, and applicable benchmark.
-
-During implementation:
-
-1. Add the failing fixture first.
-2. Try native behavior and existing semantics before changing the framework.
-3. Keep the smallest correct patch and preserve ordinary React-shaped TSX.
-4. Add nearby negative diagnostics and static zero-cost exclusion.
-5. Do not broaden public APIs, IR, runtime, or package support beyond the packet.
-
-Before release:
-
-1. Run focused tests, `npm run check`, `npm test`, required Chrome, and package
-   smoke.
-2. Run affected output and performance gates with at least seven interleaved
-   measured samples where a comparison is claimed.
-3. Record fixtures, semantic primitives, core/pass LOC, runtime concepts,
-   browser bytes, output digest, benchmark deltas, AI-cost effect, risks, and
-   known limits.
-4. Update package version, lockfile, release notes, support documentation, and
-   registry-facing records only on the exact accepted release commit.
-5. Complete the release transaction below, then advance the active status to
-   the next patch.
-
-## Release Transaction
-
-Every completed patch and every new minor is a real release. A work session that
-does not complete its packet does not consume a version; the next session
-continues the same target version.
-
-1. Confirm the target version is the active packet and is not already present in
-   npm or Git tags.
-2. Update `package.json`, root lockfile metadata, `RELEASES.md`, README/support
-   records, and generator metadata when the release changes generated projects.
-3. Run the complete release gate on the exact versioned tree: focused tests,
-   `npm run check`, `npm test`, required Chrome, `npm run test:package`, affected
-   benchmarks, package inspection, and `git diff --check`.
-4. Inspect `git status`, the complete diff, and recent release history. Stage
-   only the intended packet and release files.
-5. Create one release commit using the established `Release Kudzu X.Y.Z`
-   convention. Do not amend a rejected release commit; fix forward and create a
-   new commit.
-6. Push the release commit and require its CI checks to pass.
-7. Create tag `vX.Y.Z` on that exact commit and push the tag. Never move or
-   repoint a published tag.
-8. Publish the matching GitHub release. The existing `Publish npm` workflow
-   checks out the tag, reruns check/test/package gates, verifies version/tag
-   equality, publishes `@kudzujs/core` and any changed `create-kudzu`, and checks
-   npm registry visibility.
-9. Require the publish workflow to pass, verify npm metadata and a fresh install
-   manually, and record the release URL and registry versions.
-10. Only after commit, push, tag, GitHub release, npm publication, registry
-    verification, and fresh-install verification succeed may the next patch
-    become active.
-
-If Git push, tag push, GitHub release, npm publication, or registry verification
-fails, the current packet remains active. Do not start or version the next
-packet. npm versions and published tags are immutable; recover with the same
-release transaction where possible or document and publish a forward-fix patch.
-
-## Active Status
-
-| Patch | Status | Next exact action | Blocker |
-|---|---|---|---|
-| `0.10.0` | Released | Preserve the published application baseline and release evidence. | None |
-| `0.10.1` | Released | Preserve the published state-scale evidence and explicit object-field limit. | None |
-| `0.10.2` | Released | Preserve the published shared-layout navigation and exact route-release evidence. | None |
-| `0.10.3` | Released | Preserve the published guarded persistence recipe and zero-runtime exclusion. | None |
-| `0.11.0` | Released | Preserve the published owned-fetch lifecycle and zero-query-runtime exclusion. | None |
-| `0.11.1` | Released | Preserve the published single-owner consistency and exact request-count evidence. | None |
-| `0.11.2` | Released | Preserve the published no-new-primitive architecture decision and zero-byte result. | None |
-| `0.11.3` | Released | Preserve the published optimistic rollback, duplicate prevention, navigation, and retry evidence. | None |
-| `0.11.4` | Released | Preserve the published bounded pagination, history, refresh, polling cleanup, and exact request-count evidence. | None |
-| `0.12.0` | Released | Preserve direct runtime entry, reload, rewrite, and native fallback evidence. | None |
-| `0.12.1` | Released | Preserve the published layout/history/focus/scroll evidence. | None |
-| `0.12.2` | Released | Preserve the published authentication and permission evidence. | None |
-| `0.12.3` | Released | Preserve the published route failure, focused retry, native fallback, and precommit retention evidence. | None |
-| `0.12.4` | Released | Preserve the published single-layout decision and zero-owner-chain evidence. | None |
-| `0.13.0` | Released | Preserve native constraints, server feedback, focused field errors, retained input, and retry evidence. | None |
-| `0.13.1` | Released | Preserve application-owned nested metadata, keyed row identity, release, and reset evidence. | None |
-| `0.13.2` | Released | Preserve multistep validation, versioned persistence, stale-save rejection, conflict retention, and reset evidence. | None |
-| `0.13.3` | Released | Preserve the published upload validation, cancellation, retry, route cleanup, and attachment mutation evidence. | None |
-| `0.14.0` | Released | Preserve native table CRUD, retained identity, keyboard access, and measured update evidence. | None |
-| `0.14.1` | Released | Preserve immutable object-state replacement, nested keyed identity, latest handlers, and exact row release. | None |
-| `0.14.2` | Released | Preserve bounded cursor loading, duplicate suppression, retry/end behavior, retained identity, and route cleanup. | None |
-| `0.14.3` | Released | Preserve the published 10,000-item pagination decision and measurement evidence. | None |
-| `0.14.4` | Closed by stop condition | Preserve pagination as the selected strategy until three independent fixtures authorize virtual ownership review. | Evidence favors pagination; zero qualifying fixtures |
-| `0.15.0` | Closed by native composition | Preserve the native dialog, focus, cancellation, keyed deletion, and route cleanup evidence. | No production change; no release consumed |
-| `0.15.1` | Released | Preserve key-scoped IDs, native Popover behavior, retained keyed identity, and route cleanup evidence. | None |
-| `0.15.2` | Closed by layout composition | Preserve owner-bounded queue, deduplication, dismissal, route persistence, live-region, and timer cleanup evidence. | No production change; no release consumed |
-| `0.15.3` | Closed by stop condition | Preserve native top-layer and in-place composition until three independent authored-range failures exist. | Zero qualifying failures; no release consumed |
-| `0.16.0` | Closed by existing effect ownership | Preserve real package mount, replacement, disposal, remount, and static exclusion evidence. | No production change; no release consumed |
-| `0.16.1` | Released | Preserve the retained CodeMirror instance boundary, bidirectional updates, error recovery, accessibility, and exact cleanup. | None |
-| `0.16.2` | Closed by existing effect ownership | Preserve retained Chart.js data/resize updates, listener and route cleanup, fresh remount, accessibility, and static exclusion. | No production change; no release consumed |
-| `0.16.3` | Released | Preserve state-owned order, keyed identity, keyboard parity, invalid-input recovery, package disposal, and static exclusion. | None |
-| `0.16.4` | Released | Preserve scoped GSAP presentation, reduced-motion fallback, dependency/owner cleanup, fresh remount, and static exclusion. | None |
-| `0.16.5` | Released | Preserve the reproducible endurance evidence, project notification ownership, and bounded stale-prefetch cache disposal. | None |
-| `0.17.0` | Closed by existing layout effect ownership | Preserve project notification connect, keyed update, reconnect, stale rejection, offline recovery, retained navigation, and exact cleanup evidence. | No production change; no release consumed |
-| `0.17.1` | Closed by stop condition | Preserve independent effect ownership until three applications prove the complete shared transport and bounded replay contract. | Zero qualifying applications; no release consumed |
-| `0.17.2` | Released as `0.16.5` | Preserve stale-prefetch rejection and bounded document, state, resource, listener, DOM, and external-instance ownership. | Patch release retained the `0.16.x` public version line |
-| `0.18.0` | Closed by existing artifact reporting | Preserve exact route/source ownership, bytes, hashes, preload equality, zero-JavaScript controls, and bounded prefetch policy. | No production change; no release consumed |
-| `0.18.1` | Released as `0.16.6` | Preserve one guarded literal package edge, native module-map deduplication, exact owner cleanup, deferred artifact ownership, and static exclusion. | Patch release retained the `0.16.x` public version line |
-| `0.18.2` | Released as `0.16.7` | Preserve on-demand retained editor mount, bidirectional updates, failure recovery, identity, exact cleanup, cached fresh remount, and static exclusion. | Patch release retained the `0.16.x` public version line |
-| `0.18.3` | Released as `0.16.8` | Preserve shared deferred ownership, interaction-only loading, native document module-map deduplication, exact owner cleanup, binding-aware import validation, and static exclusion. | Patch release retained the `0.16.x` public version line |
-| `0.19.0` | Released as `0.16.9` | Preserve the compatibility registry, reachable source inventory, exact ranges, and package-neutral output boundary. | Patch release retained the `0.16.x` public version line |
-| `0.19.1` | Released as `0.16.10` | Preserve anonymous, invalid/valid login, token restore, layout-shared session reads, 401 clear, replacement navigation, and public zero-JavaScript output. | Patch release retained the `0.16.x` public version line |
-| `0.19.2` | Closed by existing semantics | Preserve connected auth, query, route, create/edit validation, refresh, persistence, admin deletion, logout, source retention, accessibility, and public zero-JavaScript output. | No production change; no release consumed |
-| `0.19.3` | Closed by existing semantics | Preserve connected Memos login, loading/error, feed, overlapping-key pagination, refresh, CRUD, reload persistence, logout, source retention, accessibility, and public zero-JavaScript output. | No production change; no release consumed |
-| `0.19.4` | Closed by existing effect ownership | Preserve reaction mutation, version deduplication, stale-socket rejection, reconnect, keyed identity, exact route cleanup, fresh ownership, bounded handles/listeners/timers, and public zero-JavaScript output. | No production change; no release consumed |
-| `0.19.5` | Deferred by stop condition | Resume only from a complete pinned Actual Budget workspace with an executable dependency graph and honest retention denominator. | Sparse acquisition omits required workspace and core packages; no compatibility claim |
-| `0.20.0` | Released as `0.16.11` | Preserve stable diagnostic schema/codes, authored ranges, human errors, compatibility reuse, and zero browser/output delta. | Patch release retained the `0.16.x` public version line |
-| `0.20.1` | Released as `0.16.12` | Preserve bounded deterministic reachable inventory, first-blocker retention, existing-record reuse, and zero browser/output delta. | Patch release retained the `0.16.x` public version line |
-| `0.20.2` | Released as `0.16.13` | Preserve exact emitted-route lookup, authored provenance, semantic ownership, capability/artifact byte reasons and hashes, bounded deterministic output, structured missing-route diagnostics, and zero-JavaScript explanation. | Patch release retained the `0.16.x` public version line |
-| `0.20.3` | Deferred by evidence gate | Resume only after recorded AI trials prove one repeated correction safe and useful with a stable diagnostic code and compiler-proven preconditions. | No recorded AI trials; no production change or release consumed |
-| `0.20.4` | Released as `0.16.14` | Preserve pinned equal conditions, isolated paired attempts, raw failures, independent acceptance, attribution, failure-inclusive costs, source retention, and browser artifact evidence. | Deterministic fixture validates the protocol only; no framework ranking claimed |
-| `0.20.5` | Closed by measured stop condition | Preserve the negative tooling-cost result and do not present structured tools as an AI cost reduction. | Tool-assisted attempts cost 16,933 more tokens per success; no release consumed |
-| `0.21.0` | Closed by existing semantics | Preserve the complete browser parity matrix, large-list keyboard/release acceptance, and static exclusion controls. | No production change; no release consumed |
-| `0.21.1` | Released as `0.16.16` | Preserve 100/1,000/10,000-route phase, RSS, output, digest, incremental-equivalence, recovery, and bounded canonical-AST retention evidence. | Full 10,000-route report has one complete measured run after two measured-stage diagnostics; no projection used |
-| `0.21.2` | Closed by existing semantics | Preserve initial/session/lazy bytes, keyed/range/navigation median alarms, and bounded endurance ownership evidence. | Same-revision absolute gates; structural artifact bytes are not compressed network transfer |
-| `0.21.3` | Released as `0.16.17` | Preserve exact package, browser, benchmark, registry, and fresh-install evidence. | One initial navigation benchmark invocation stalled at the outer timeout; its clean seven-profile rerun passed all alarms |
-| `0.21.4` | r5 executed all 50 attempts; compiler follow-up verified; gate remains blocked | Preserve raw scores and partial traces. Direct-count/static-topic fixes and the render-ref diagnostic are verified; persistent version refs remain unsupported. Use aligned r6 only in a separately authorized complete run. | Local historical r5 scorer: Kudzu 18/25 versus React + Vite 23/25, with two partial Kudzu traces. Source replay passes all ten content static siblings but does not replace scores. Published r2 remains 11/25 versus 24/25. |