RxJS Next active project plan

Executive summary

The project will establish a reliable platform Observable foundation before expanding the operator catalog and migration promises. The user temporarily prioritized the attested Observable WPT harness and fallback-conformance work ahead of the package and installation decision. Both slices are complete: the fallback passes every pinned Observable WPT while exact implementation identity is proved in every tested realm. D-045 supersedes D-042 and requires an explicit Subscriber.next argument, including next(undefined) for a platform Subscriber<void>. The portable RxJS 7-to-Next marble-test migration Skill, its independent vetting, and the classified repository port are also complete. The user has clarified that preservation in a manifest is not enough: all cases must become executable parity registrations even when their capabilities are absent. The exhaustive follow-up is complete: 2,201 physical test declarations expand to 2,338 uniquely identified registrations, including parameterized and source-skipped evidence. The user has now temporarily prioritized and completed P0.T3: the durable failure ledger retains every original failure, all 2,338 registrations pass in cold and polyfill modes, and the strict RxJS unit gate was green against the artifact then in use. P0.3 later exposed that the focused portion had consumed a stale polyfill build. P0.4 has reconciled that evidence with one reusable lifecycle contract that runs against the packaged fallback and a browser-native implementation, and has completed the package-import safety fixtures. P0.2 has now accepted the three-package map, conditional per-realm fallback contract, and migration-over-emulation direction. P0.3 has now implemented those package decisions, restored clean publication builds, and added the required import/realm evidence. P0.5 records the written specification reference, executable WPT pin, maintainer-owned update policy, and complete all-pass gate. A later user-prioritized runner follow-up removed expected-failure quarantine. P0.DX4 replaced that custom dynamic launcher and shard renderer with real Vitest files and stock reporting. The user then clarified that those files must be ordinary owned migration results rather than generator-owned artifacts, and prioritized P0.M1 with a publishable, framework-adaptable migration package dogfooded on this repository. The user-prioritized repository package-manager migration to pnpm 10 is complete without deciding the P0.2 publication boundary. The root developer command guide is also concise, task-oriented, and explicit about known failing gates. The user has now made the complete agent-first migration experience the top project priority. The focused P0.M1 package is useful mechanical groundwork, but its nine package tests and unreferenced single source fixture do not prove general RxJS 7 migration safety. P0.M2 is complete: D-046 and packages/migrate/docs/MIGRATION_TOOLING_DESIGN.md defines the agent-first product contract, explicit lifecycle classification, separate mechanical and agent evidence lanes, one canonical Skill with thin Codex/Claude/Cursor adapters, and no MCP release surface. P0.M3 and P0.M4 are complete: the bounded engine now supports one canonical portable Skill, copy-only Codex/Claude/Cursor adapters, digest-tracked updates, and a generated repository discovery copy. P0.M5 is also complete: four Codex/ChatGPT representative runs passed the 14 semantic gate families, with three completed migrations and one required safe stop. Phase 1 is complete. P1.1 through P1.3 required no additional runtime work: P0.3 and P0.4 had already completed the conditional-installation and shared lifecycle boundaries, while P0.5 and P1.4b had already completed the pinned fallback-conformance work. Phase 2 is also complete: D-048 and D-049 define the extension kernel, and the map, scan, switchMap, timeout, timer, and pipe pilot proves it against packaged fallback and browser-native Observable. P3.1 confirmed that Phase 0 already restored the broad operator catalog and executable evidence; the remaining Phase 3 work is to move the catalog onto the accepted extension kernel. P3.2 has materialized the existing parity map as a validated migration-evidence ledger. P3.3 moved the full exact Symbol catalog onto the transactional extension kernel, and P3.4 completed the source-pinned evidence audit. D-051 now supersedes that installation mechanism: P4.I1 removed the common installer and moved all 97 public capabilities to direct exact-Symbol assignment without changing operator behavior. Phase 4 is complete. P5.1 produced package-local migration guidance and moved the generated evidence references beside the rxjs package. A P5.2 closure audit then confirmed that the existing deterministic, contract-fixture, and four-repository qualification evidence already satisfies representative application validation. P6.1 then accepted the exact RxJS 9 support and distribution policy, synchronized 9.0.0-beta.0, and collapsed four duplicate dialects into one ESM implementation. P6.2 through P6.4 completed the release matrix, package-local documentation, packed-consumer adoption, and beta approval. P6.5 closed the terminal audit. P6.6 then centralized eligible RxJS source subscriptions through the D-049 helper and recorded the resulting bundle-size reduction. P6.7 then removed the remaining derived-construction wrappers in favor of direct D-037 [create] calls and recorded a further bundle-size reduction. P6.8 completed durable pull-request and master CI ownership for every accepted RxJS 9 test and release check. P6.9 implemented truthful status signals and security automation and validated their first live GitHub results. The user then explicitly prioritized P6.11, which removed inherited callback thisArg parameters and their hot-path dispatch cost. P6.10 remains the sole NEXT item: one understandable, interactive single-maintainer beta publication command with npm two-factor authentication and no CI publishing credential.

RxJS 9 and 9.0.0-beta.0 are selected under D-007. D-053 defines runtime, browser, bundler, channel, and RxJS 7 maintenance policy. Dates and staffing commitments remain unassigned.

Status protocol

• DONE: completion bar met and evidence recorded.
• NEXT: the single active step.
• PLANNED: sequenced but not active.
• BLOCKED: cannot proceed without a named decision or external change.
• DEFERRED: accepted work intentionally not designed or scheduled yet.

Keep exactly one NEXT item while the execution queue is open. When it completes, mark it DONE, record verification in the session log, and move NEXT to the earliest unblocked item. A completed queue has no NEXT item.

Success criteria

• Native Observable is preserved and the fallback is installed only according to the approved contract.
• Platform behavior passes every selected Observable test from the pinned WPT revision, with the related written specification revision recorded for diagnosis and reproducibility.
• Symbol extension identity, installation, construction, cancellation, typing, and packaging conventions are uniform and tested.
• Supported operators are backed by classified RxJS 7 behavioral evidence.
• RxJS 7 migration mappings are backed by classified evidence and do not imply a runtime compatibility package.
• The primary migration experience is agent-led: it assesses the repository, protects uncovered behavior with an RxJS 7 baseline, classifies the intended Next contract, invokes only proven mechanical transforms, and iterates with the developer until agreed build and test gates are green.
• Deterministic transforms are fixture-tested for output, compilation, idempotence, diagnostics, and behavior; agent-produced migrations are judged by declared contracts and test outcomes rather than exact source shape.
• One portable migration workflow has documented discovery, installation, invocation, permission, and verification paths for Codex, Claude, and Cursor, with harness-specific adapters kept thin and testable.
• Every published entry point builds and passes import/type fixtures in supported environments.
• Migration guidance and future AI tools are generated from accepted behavior, not prototype assumptions.

Active queue

Phase 0 — Foundation and architectural safety rails

Status	ID	Outcome
DONE	P0.1	Record the charter, current architecture, migration policy, decisions, risks, open questions, and AI working rules
DONE	P0.T1	Design and implement the user-prioritized framework-neutral @rxjs/test virtual-time package
DONE	P0.T2a	Create a portable RxJS 7-to-Next marble-test migration Skill
DONE	P0.T2b	Vet the migration Skill independently before using it on the repository
DONE	P0.T2c	Port and classify the RxJS 7 marble-test corpus without repairing production behavior
DONE	P0.T2d	Materialize every inventoried marble case as an executable parity-test registration
DONE	P0.T2e	Exhaustively convert remaining runnable RxJS 7 marble evidence and expand capability mappings
DONE	P0.T2f	Make the default ported-test gate strict and progress-visible
DONE	P0.T3	Resolve the cold and polyfill RxJS 7 parity failures through the durable operator/function work queue
DONE	P0.DX1	Migrate repository workspaces, automation, and contributor tooling from Yarn Classic to pnpm 10
DONE	P0.DX2	Make the root developer command guide concise, accurate, and task-oriented
DONE	P0.DX3	Add cached one-shot bundle comparison for current Next and published RxJS versions
DONE	P0.DX4	Materialize normal rxTest specs with stock Vitest reporting and real source locations
DONE	P0.I1	Add four explicit Symbol-based Observable-to-async-iterator strategies
DONE	P0.2	Decide the package map and native-versus-polyfill installation contract
DONE	P0.3	Restore green builds and coherent public entry points for the selected package map
DONE	P0.4	Add a native/fallback lifecycle test harness and package-import fixtures
DONE	P0.M1	Publish and dogfood one-time RxJS 7-to-Next test migration tooling
DONE	P0.M2	Design the agent-first migration product, contracts, fixture strategy, and cross-harness distribution
DONE	P0.M3	Harden @rxjs/migrate as the deterministic mechanical engine with comprehensive executable fixtures
DONE	P0.M4	Publish the portable migration Skill and thin Codex, Claude, and Cursor integration guidance
DONE	P0.M5	Qualify the end-to-end Codex/ChatGPT workflow on representative repositories and behavioral outcome gates
DONE	P0.5	Pin the written Observable reference and require every selected test from the executable WPT revision to pass

P0.1 completion evidence

• Added the docs/rxjs-next document set and root AGENTS.md.
• Mapped all current packages and the Symbol extension inventory.
• Verified four unique polyfill tests and one RxJS operator test pass.
• Verified the polyfill package build fails because ambient platform declarations are disconnected from the build entry.
• Verified the RxJS package build fails because the root tshy export is invalid.

P0.T1 completion evidence

• Added the framework-neutral @rxjs/test package and public rxTest function; no scheduler instance or manual scheduler mode is exposed.
• Added marble parsing, async virtual-time controls, default and configurable assertions, and distinct cold, hot, and platform observable sources.
• Isolated marble grammar in a pure internal parser with direct unit coverage, independently of the virtual-time runtime.
• Virtualized the supported host timing and clock APIs with exact restoration, same-realm serialization, pending-work diagnostics, and task/time guards.
• Passed 66 focused behavior, parser, and cleanup tests, lint, TypeScript source checks, Nx build/test targets, the multi-dialect package build, strict declaration coexistence with the Observable polyfill, and built ESM/CommonJS imports.
• Recorded D-012 and the complete contract in TESTING_DESIGN.md.

P0.T2a completion bar

• A repo-committed Skill teaches developers to migrate RxJS 7 TestScheduler marble cases to rxTest, ColdObservable, and the platform Observable.
• The Skill preserves the caller's test framework, requires the global Observable for platform cases, and separates mechanical changes from semantic review.
• The Skill and bundled helpers contain no RxJS-repository paths, branches, commits, Git operations, or package-internal assumptions.

P0.T2a completion evidence

• Added the repo-committed rxjs-next-marble-migration Skill with concise workflow guidance, conversion and lifecycle references, examples, and a migration-report template.
• Added dependency-free file analysis, duplicate-candidate detection, review-flag detection, and a portability checker that accepts ordinary filesystem paths without source-control behavior.
• Passed helper tests, the portability scan across all nine Skill files, the Skill structural validator, and temporary .skill packaging.

P0.T2b completion bar

• Skill structure, links, examples, helper scripts, and packaging validate.
• Independent skill evaluation has no unresolved blocking or fix-first finding.
• Consumer-style fixtures cover synchronous, higher-order, timed, cancellation, duplicate, missing-operator, sharing-divergence, and obsolete-harness cases.
• No repository marble-test port begins before this bar is met.

P0.T2b completion evidence

• Exercised the dependency-free analyzer in a temporary consumer-style project across synchronous, higher-order, timed, cancellation, duplicate, missing-operator, sharing-divergence, and obsolete scheduler-internal cases.
• Parsed representative migrated rxTest cases for framework preservation, awaited flush, Symbol invocation, and global Observable construction.
• Corrected the one initial evaluation warning by making the Skill trigger sentence explicit.
• Passed the structural validator and independent plugin evaluation at 100/100 with no failures, warnings, or fix-first recommendations.

P0.T2c completion bar

• Every inventoried RxJS 7 marble case has exactly one recorded disposition: active, expected failure, missing API, deduplicated, or unsupported/obsolete.
• Shared definitions can run in isolated ColdObservable, polyfill, and native-if-present modes without importing the platform constructor.
• Repository notes reconcile source counts, duplicate provenance, known failures, missing APIs, and semantic divergences.
• Production Observable, operator, and direct-subscription implementations are not changed to make the port pass.

P0.T2c completion evidence

• Read the production 7.x ref without checking it out and pinned e5351d02e225e275ac0e497c7b66eaa5f0c88791 only in the repository port records.
• Reconciled all 2,146 inventoried cases as 333 active, 422 expected failure, 1,251 missing API, 4 deduplicated, and 136 unsupported/obsolete.
• Added a reproducible generated manifest retaining original provenance, converted definitions, classifications, reasons, capability gaps, and exact duplicate links.
• Added isolated cold, polyfill, native-if-present, and per-mode audit commands. Normal cold and polyfill each register all 2,146 source cases; native detection skips explicitly in the current Node realm. Cold audit records 335 passes and 1,811 failures; polyfill audit records 340 passes and 1,806 failures.
• Recorded the platform shared/ref-counted lifecycle separately from cold-mode expectations and required platform cases to construct from the ambient Observable.
• Added D-013 and the detailed port notes. No production implementation was changed, and P0.2 is restored as the single NEXT item.

P0.T2d completion bar

• All 2,146 inventoried source cases are registered by the parity harness, including missing APIs, scheduler/harness dependencies, deduplicated provenance, and obsolete coverage.
• Missing capabilities fail with explicit source-linked capability diagnostics rather than being omitted from test collection.
• The raw parity command reports the complete pass/fail/duplicate disposition without changing production operators or direct-subscription behavior.
• Normal and platform modes remain bounded; any required sharding or process isolation is part of the harness rather than a reason to omit cases.
• Port notes, verification evidence, and the generated manifest agree on the registered total.

P0.T2d completion evidence

• Preserved a mechanically generated executable program for every manifest entry, including all four deduplicated source locations and all 136 scheduler/harness-dependent cases.
• Registered all 2,146 cases in cold, polyfill, and native-if-present modes. Mode-specific pass baselines execute verified cases normally; every other non-duplicate case remains an expected-failure parity gate until its exact capability and behavior are available.
• Made missing APIs and unavailable harness facilities fail with explicit source ID, original claim, and capability/rewrite reason. The raw cold audit reports 1,811 failures and 335 passes; the polyfill audit reports 1,806 failures and 340 passes.
• Added role-aware import mapping so source.pipe(operator(arg1, arg2)) invokes its exact or unified Next Symbol as source[targetSymbol](...adaptedArgs), separately from static factory Symbols, ambient-platform constructions, and standalone values.
• Added one machine-readable capability registry and the generated RxJS-7-parity.md map covering 113 RxJS 7 operators and 34 creation/utility functions, including every missing exact surface and marble-case usage counts.
• Passed the complete 2,146-registration cold and polyfill gates, native auto-detection, manifest validation, parity-document freshness, and reproducible generation. No production source was changed by the port. P0.2 is restored as the single NEXT item.

P0.T2e completion bar

• Re-audit every unsupported or missing-capability disposition against the current RxJS Next source and platform Observable surface.
• Prefer an executable mechanical conversion for every source-linked claim, including helper-generated and scheduler-shaped cases, while preserving explicit failures for unavailable behavior.
• Run the complete cold and polyfill inventories plus native auto-detection and raw audits without changing production operators to satisfy expectations.
• Regenerate the capability map, parity document, dispositions, mode baselines, and notes from the pinned read-only RxJS 7 source revision.

P0.T2e completion evidence

• Reconciled 2,201 physical declarations into 2,338 unique registrations, including 169 parameterized variants and all four source-skipped cases.
• Preserved an executable converted program for every registration. Only 11 cases remain classified unsupported/obsolete, and each produces an explicit source-linked diagnostic rather than disappearing from the suite.
• Expanded the executable capability registry to 51 operator mappings, 18 factory mappings, and 11 standalone values, including exact, partial, and unified mappings for the newly available APIs.
• Added unique case-ID baselines, complete sharded audit merging, ambient native-constructor identity checks, and five dedicated platform lifecycle cases.
• Passed the complete 2,338-case normal cold and polyfill gates. Raw audits recorded 432 cold passes with 1,906 failures and 436 polyfill passes with 1,902 failures. Native mode skipped explicitly in the current Node realm.
• Regenerated the manifest, parity map, reviewed baselines, and port notes without changing production source.

P0.T2f completion bar

• Every applicable ported registration uses ordinary test semantics in the default cold, polyfill, and native-if-present modes.
• Recorded pass baselines cannot skip, quarantine, or invert a default test result.
• The launcher continues through every shard, reports visible progress while they run, expands every failed shard's diagnostics, and exits nonzero when any test fails.
• No production Observable, operator, or direct-subscription behavior changes.

P0.T2f completion evidence

• Removed the mode-baseline registration branch and all it.fails use from the ported suite; all applicable manifest cases now register with ordinary it(...).
• Added one in-place interactive status line, refreshed on shard completion and every ten seconds, showing completed, running, queued, failed, and elapsed counts. Non-interactive output emits only a final progress summary.
• Regenerated the manifest so now-visible failure reasons describe failing parity evidence without claiming those tests remain quarantined.
• Ran the default test:ported command through all 16 cold and all 16 polyfill shards. Cold completed in 84 seconds and polyfill in 94 seconds; every shard exposed failures, all failed-shard diagnostics were expanded, and the command returned exit code 1 as intended.
• Preserved the one proven non-yielding converted program as a registered, immediate explicit failure so it cannot freeze its shard. Production source was unchanged.

P0.T3 completion bar

• RXJS_7_PORTED_FAILURES.md retains every cold or polyfill failure by stable case ID, even after the case becomes FIXED.
• Every tracked case belongs to exactly one operator/function work packet and uses only TODO, IN-PROCESS, FIXED, or a BLOCKED status with a named dependency.
• Each packet is resolved at the correct platform, intentional Next, harness, or approved-divergence boundary without weakening RxJS 7 behavioral evidence.
• Complete cold and polyfill audits account for all 2,338 cases, every tracked row is FIXED, and the normal strict ported-test gate passes.
• Completion evidence and the final session log are recorded before restoring P0.2 as the single NEXT item.

P0.T3 completion evidence

• The durable ledger retains all 1,923 cases that failed an authoritative audit, assigns each to exactly one of 140 work packets, and marks every row FIXED.
• Complete cold and polyfill audits each passed all 2,338 registrations with zero failures. The final manifest has 1,503 active, 831 compatibility/expected-failure, and 4 exact-deduplicate registrations; every non-duplicate registration still executes with ordinary strict semantics.
• pnpm --filter rxjs test passed 705 focused source tests, then both complete parity modes.
• Scheduler-specific RxJS 7 evidence is resolved at the explicit migration-evidence boundary recorded in D-033. No public scheduler class, general scheduler argument, string-named RxJS method, or global-registry Symbol was introduced.
• P0.2 is restored as the single NEXT item.

P0.DX1 completion evidence

• Pinned pnpm 10.34.5, made pnpm-workspace.yaml authoritative, generated a pnpm lockfile from the former Yarn lockfile, and removed Yarn tooling and lockfile support.
• Kept pnpm's isolated linker, added explicit Chai and Yargs dependencies exposed by that layout, and limited the temporary polyfill bridge to one documented root public-hoist without changing the published package contract.
• Added a version-bounded dependency-build policy with strictDepBuilds, converted CI, release, documentation, generated commands, and contributor guidance, and retained npm for publication and end-user installation.
• Patched the pinned Husky 4 hook runner to use pnpm exec, preserving the configured commit hooks under pnpm 10 without a dependency upgrade.
• Passed a clean frozen install and a stable repeat install, five-project Nx discovery, 53 polyfill tests, 66 @rxjs/test tests plus package fixtures, six focused RxJS tests, parity freshness, WPT import verification, and the strict attested WPT run with 52/52 URLs, 525/525 upstream subtests, and 52/52 identity attestations.
• Confirmed the docs app resolves published RxJS 7.8.1 and does not receive the local polyfill bridge. The release dry-run reaches the unchanged package build/lint baseline. Existing Angular/TypeScript peer warnings, the Node 24 docs-test ESM failure, the optional RE2 Node 24 build failure, and the known RxJS Next package build/lint failures remain out of scope.
• Restored P0.2 as the single NEXT item.

P0.DX2 completion evidence

• Replaced the root setup list with compact, task-oriented command tables for workspace discovery, fast package feedback, documentation, WPT, and release preparation.
• Highlighted the common inner-loop commands and named all five pnpm workspace projects.
• Distinguished focused green checks from the intentionally failing full RxJS parity suite and known P0.3 build/lint baselines.
• Linked specialized docs and verified the listed commands against current workspace scripts while keeping P0.2 as the single NEXT item.

P0.DX3 completion evidence

• Added a one-shot production Webpack comparison for the current RxJS Next runtime surface with and without the Observable fallback, plus the public root surface of selected published RxJS versions.
• Kept the analyzer toolchain in a disposable pnpm project and left workspace dependencies and the lockfile unchanged.
• Cached each published version's standalone bundle, complete module stats, registry integrity, and configuration fingerprint under an ignored local cache. Exact cache hits do not resolve or rebuild the published package; --refresh is the explicit rebuild path.
• Flattened fresh and cached compilations into one analyzer-compatible stats document and generated one static gzip-first report under dist/bundle-analysis/.
• Verified CLI parsing, source filtering, cache keys, and stats composition with focused Node tests. The default 7.8.2 integration run, cache-hit reuse, and refresh behavior are recorded in the session log.

P0.DX4 completion evidence

• Materialized 147 formatted Vitest files per execution mode, with 2,338 direct rxTest cases in each and source-pinned RxJS 7 URL comments.
• Removed the dynamic registration module and custom shard/progress launcher. Normal runs use Vitest's public built-in default reporter unchanged, so diagnostics point to real repository files and lines.
• Made cold construction explicit without replacing globalThis.Observable: cold fixtures extend ColdObservable, hot fixtures extend the active global constructor and derive ordinary platform Observables, and observable() directly models that global constructor.
• Complete built-in JSON audits record 2,296/2,338 cold passes and 2,316/2,338 fallback-platform passes. Generated file/declaration order maps audit results back to manifest case IDs without polluting human test names.
• Kept P0.5 as the sole project-level NEXT item.

P0.M1 completion evidence

• Added the publishable @rxjs/migrate workspace package with a framework-neutral migration core, a one-time CLI, reusable Skill assets, and an exploratory read-only MCP prototype later excluded from the accepted release product by D-046.
• Source- and target-test-framework concerns are adapters. The first supported path migrates Mocha/Chai tests to Vitest without coupling rxTest conversion semantics to Vitest.
• Dogfooded its Mocha/Chai-to-Vitest adapter while adopting the repository's ported corpus as ordinary human-readable specs with local file names and file-level source repository, revision, and path provenance. No test tells contributors to edit or rerun a generator.
• Each migrated test calls rxTest through normal test-framework code. Any migration-specific comment explains a real semantic rewrite; parameterized source cases are materialized as direct tests rather than dynamic runners.
• Cold and platform source lifecycles remain explicit in the test files. Only the repository-owned platform execution matrix selects native versus polyfill Observable implementations in isolated realms.
• The former test-file generator is not part of the test, audit, or contributor workflow. Stock Vitest reporters continue to provide clickable real paths.
• Verified 9 package tests, lint, ESM/CommonJS imports, CLI dry run, package build and publication dry run, and the portable Skill checks. Focused @rxjs/test and RxJS source suites pass 75/75 and 733/733 tests.
• Complete owned-file audits cover all 2,338 cases per mode and retain the reviewed 2,296 cold and 2,316 fallback-platform passes. Representative formerly dynamic suites pass 176/176 direct tests; the focused concat failure reports its real checked-in file and line through stock Vitest.

P0.M2 completion bar

• A maintainer-facing design document defines the migration user journey: repository/version and coverage assessment; warnings and test recommendations; optional pre-migration characterization tests that pass on RxJS 7; explicit target-contract classification; mechanical transformation; and an interactive build/test/repair loop that ends green or with a developer-accepted, documented blocker.
• Product boundaries are explicit. @rxjs/migrate is the deterministic, versioned codemod engine and never claims to complete a migration alone. The portable Skill is the primary reasoning and orchestration layer. A separate MCP is included only if the design identifies capabilities that a machine-readable CLI and local agent tools cannot provide adequately.
• The design identifies one canonical Skill source and a distribution/version model that avoids duplicated instructions while documenting discovery, installation, invocation, permissions, and updates for Codex, Claude, and Cursor. Shared workflow semantics are portable; harness-specific material is limited to adapters and installation instructions.
• A fixture and evaluation specification distinguishes deterministic golden transformations from nondeterministic agent outcomes. Mechanical fixtures cover parsing, exact or invariant output shape, compilation, idempotence, diagnostics, and behavior. Agent fixtures use RxJS 7 characterization baselines, explicit Next-contract manifests, compile/build/test outcomes, intentional divergences, and required refusal/escalation cases.
• The design includes representative fixture categories for ColdObservable, platform sharing/ref counting, Subjects, cancellation, teardown order, scheduling/timing, errors, input conversion, repeated subscriptions, unsupported APIs, missing coverage, and mixed mechanically supported and unsupported pipelines.
• D-044 and related architecture, compatibility, open-question, and risk records are corrected where they overstate current validation or conflate the codemod, Skill, CLI, and MCP responsibilities. Later implementation steps have developer-ready scope, dependencies, and observable exit gates.

P0.M2 execution handoff

• Objective: create the package-owned migration tooling design (now at packages/migrate/docs/MIGRATION_TOOLING_DESIGN.md) as the controlling product and technical design for the complete agent-first migration experience.
• Why now: migration safety is a release-defining user experience, while P0.M1 currently risks making a nine-test codemod prototype and thin MCP look like the complete product.
• Scope in: audit the present package, Skills, MCP, decisions, and claims; specify the user journey and contract-classification protocol; define the mechanical and agent fixture schemas; choose canonical Skill ownership and Codex/Claude/Cursor distribution boundaries; set explicit MCP decision criteria; and correct controlling documentation.
• Scope out: production codemod changes, package restructuring, publishing, and live outcome qualification. Those belong to P0.M3 through P0.M5; D-047 later bounded the live P0.M5 lane to Codex/ChatGPT.
• Dependencies: current RxJS Next architecture and migration evidence plus the user-approved workflow recorded here. Full operator stabilization is not required for the design, because capability claims must be versioned and evidence-linked rather than copied into static instructions.
• Verification: documentation links resolve; package/Skill/MCP boundaries are consistent across the charter, architecture, decisions, compatibility policy, open questions, and plan; fixture gates have negative controls; and the plan retains exactly one NEXT marker.
• Execution skills: lead with developer-experience for the portable paved road and technical-business-documents for the durable product contract; use TypeScript expertise only when P0.M3 begins implementation.

P0.M2 completion evidence

• Added MIGRATION_TOOLING_DESIGN.md as the controlling product and technical design. It defines the eight-stage journey from authority discovery through reviewed closeout, including RxJS 7 baseline and characterization gates.
• Defined the migration contract manifest and six explicit lifecycle targets: platform-shared, producer-per-direct-subscription, subject-hot, not-applicable, unsupported, and unresolved. The last two remain visible stop or escalation states.
• Separated deterministic mechanical fixtures from nondeterministic agent evaluations. The schemas require diagnostics, containment, dry-run/write equivalence, idempotence, compilation, behavior, negative controls, contract manifests, and refusal/escalation evidence.
• Covered ColdObservable, platform sharing/ref counting, Subjects, cancellation, teardown, timing, errors, input conversion, repeated subscriptions, unsupported APIs, missing coverage, and mixed pipelines.
• Selected packages/migrate/skill as the only authored Skill source. Thin Codex, Claude Code, and Cursor placements consume the same package version and content digest and have explicit discovery, invocation, permission, update, and smoke-test requirements.
• Recorded D-046, partially superseded D-044, and excluded the P0.M1 MCP prototype from the release contract. P0.M3 owns its removal; any future MCP requires a new accepted decision and independently validated product contract.
• Reconciled the charter, architecture, compatibility policy, open questions, testing design, port notes, package README, risks, and downstream P0.M3-P0.M5 exit gates. This design slice intentionally changed no production code or test baseline.

P0.M3 completion bar

• @rxjs/migrate exposes a programmatic core and dry-run-first, machine-readable CLI with narrow preconditions, structured findings, safe file boundaries, and no agent- or harness-specific runtime dependency.
• Comprehensive checked-in input/expected-output fixtures exercise every supported mapping and argument adapter plus aliasing, shadowing, mixed supported/unsupported pipelines, malformed input, framework preservation, platform/cold selection, diagnostics, and negative controls.
• Every successful fixture parses, type-checks against its declared RxJS 7 and Next environments as applicable, is idempotent, and runs behavior tests that prove the declared preserved claims. A shape-only assertion is never the sole evidence for semantic migration.
• CLI/API equivalence, dry-run/write behavior, path containment, package imports, publication contents, and failure exit codes are tested. The package documentation describes only the mechanically supported subset.

P0.M3 completion evidence

• Replaced the prototype capability list with a versioned, schema-validated registry of ten fixture-backed mappings and eight argument adapters. The engine refuses incompatible registries, ambiguous lifecycle selection, unsafe bindings, malformed source, unsupported overloads, and non-atomic mixed pipelines without changing refused file bytes.
• Added structured deterministic diagnostics and a versioned migration contract manifest with a separate readiness assessment. Schema validity can no longer be mistaken for resolved lifecycle intent, approvals, diagnostics, verification, or blockers.
• Added exact mechanical fixtures and negative controls for output and diagnostic drift, malformed input/output, non-idempotence, false success, compile regression, behavior drift, CLI/API mismatch, and lexical or symlink path escape. All ten source fixtures compile against pinned RxJS 7.8.1; all ten outputs compile against current Next types; representative behavior comparisons execute the pinned RxJS 7 runtime and current target contracts.
• Added whole-batch planning before writes, canonical path containment, duplicate and alias rejection, no-overwrite defaults, refusal-safe writes, and versioned JSON CLI results with distinct success, refusal, invalid-argument, and operational-failure exits.
• Added ESM, CommonJS, declaration-consumer, and pack-inventory gates. Removed the MCP bin, export, implementation, tests, dependency, transitive lockfile entries, and documentation claims. Added the Node-only canonical Skill digest primitive required by P0.M4.
• Fixed the switchMap projection-result declaration exposed by the new target type gate. Passed 91 migration tests, package build/types/imports/pack, lint, RxJS public type checks, and diff checks. Advanced P0.M4 as the sole NEXT item.

P0.M4 completion bar

• One canonical portable Skill implements the agent workflow and consumes the versioned mechanical engine and migration evidence without copying mutable capability claims into harness-specific instructions.
• The Skill requires an RxJS 7 green baseline, coverage/risk report, explicit Next-contract classification, developer interaction for ambiguous intent, and post-migration build/test repair. It forbids weakening tests or silently selecting platform versus producer-per-subscription behavior.
• Codex, Claude, and Cursor each have tested installation, discovery, invocation, permission, and update instructions plus a minimal smoke scenario that reaches the same portable workflow.
• The accepted D-046 product ships no MCP. A future MCP would require a new evidence-backed decision, product boundary, and independent validation.

P0.M4 completion evidence

• Expanded packages/migrate/skill into the canonical eight-stage agent workflow, including assessment, RxJS 7 baseline and characterization tests, explicit lifecycle approval, dry-run-first bounded transforms, repair, and reviewed closeout without mutable prose copies of the capability registry.
• Added copy-only, atomic Codex, Claude Code, and Cursor adapters with package-version/digest provenance and check, install, update, and remove operations. Local changes are classified and protected unless force is explicit; Codex and Cursor share one .agents/skills copy.
• Replaced the independently authored repository Skill with the generated .agents/skills/rxjs-next-migration copy, preserving natural Codex and Cursor invocation while keeping packages/migrate/skill as the only authored source.
• Added 21 harness acceptance cases for discovery, explicit and implicit invocation metadata, least privilege, byte/digest identity, no-symlink installation, stale and modified state, updates, force, and removal.
• Documented installation, invocation, permissions, maintenance, and the common read-only smoke scenario in MIGRATION_SKILL_GUIDE.md; rechecked the official host documentation and locally inspected Codex 0.146.0-alpha.3.1, Claude Code 2.1.119, and Cursor 3.13.10.
• Passed 114 migration tests, Skill structural validation, lint, build, declaration-consumer tests, ESM/CommonJS imports, package publication dry run, generated-copy digest checks, and diff checks. D-046 remains intact: no MCP binary, export, files, or runtime dependency ships.

P0.M5 completion bar

• Representative repositories cover application and library layouts, TypeScript configurations, test frameworks, strong and weak initial test coverage, cold-preserving migrations, intentional platform migrations, mixed contracts, and unsupported behavior.
• End-to-end evaluations start from pinned RxJS 7 revisions, record baseline tests, run the agent workflow, and judge generated changes by compilation, build/test outcomes, contract manifests, diagnostics, and intentional divergences rather than exact text.
• Codex/ChatGPT runs demonstrate the required safety gates and developer decision points across all representative repositories. P0 does not claim live Claude Code or Cursor outcome parity; those harnesses retain only the P0.M4 Skill installation and discovery evidence. Nondeterministic source variation is permitted; undisclosed behavioral variation, skipped required warnings, weakened tests, or unsafe automatic contract choices fail the evaluation.
• Release-facing documentation states measured coverage and limitations and does not claim general automatic migration beyond the passing fixture and representative-repository evidence.

P0.M5 completion evidence

• Added four immutable RxJS 7.8.1 seed repositories spanning application and library layouts, Vitest, Mocha, Jest, strong and weak coverage, cold, platform, mixed, and unsupported contracts. Their lockfiles, tree identities, baseline builds, and tests are executable offline.
• Ran the closed four-scenario matrix through Codex 0.146.0-alpha.3.1 with gpt-5.6-sol at medium reasoning and canonical Skill/engine version 8.0.0-alpha.14. All four records pass all 14 semantic gate families.
• app-cold-strong, app-platform-strong, and library-mixed-strong completed their approved migrations with green target build and held-out behavior. library-weak-unsupported retained the unsupported behavior and made the expected safe stop before target installation or migration writes.
• Committed four qualification records and 20 SHA-256-bound artifacts: prompt plus final response, contract manifest, patch, command results, and final report for each run. Command/tool evidence is retained in the command results, patch, and observed-authority record. Full event streams were not retained for the first three runs and are not claimed by the conversation artifacts.
• Kept the measured result explicitly Codex/ChatGPT-only under D-047. P0.M4's Claude Code and Cursor installation/discovery evidence remains valid, but their live migration outcomes remain unmeasured.

P0.I1 completion bar

• Four exact instance Symbols expose lossless FIFO, lossless buffered, lossy latest-value, and lossy next-demand async iteration.
• Each invocation returns a fresh lazy generator, preserves the receiver's direct-subscription lifecycle, and aborts its observer during generator cleanup.
• Focused tests cover synchronous and delayed producers, terminal behavior, microtask coalescing, cancellation, platform sharing/ref counting, and ColdObservable producer-per-subscription behavior.
• Public comments and architecture records explain the implications of multiple generators over platform and direct-subscription Observables.

P0.I1 completion evidence

• Added iterateEachValue, iterateBufferedValues, iterateLatestValue, and iterateNextValue exact Symbol modules with a shared internal async-generator implementation and detailed API documentation.
• Removed the exploratory eachValueFrom and bufferedValuesFrom sources and package subpaths without aliases. The four replacement subpaths are present in both the source build inventory and published export map.
• Added 16 focused behavior, type, lifecycle, error, completion, and cancellation tests. The complete RxJS gate passed 101 focused files with 733 tests, followed by all 2,338 cold and 2,338 polyfill parity cases.
• Recorded D-038 against reviewed rxjs-for-await revision 94f9cf9cb015ac3700dfd1850eb81d36962eb70f and updated architecture, migration-policy, and open-question records.
• Package metadata parses successfully. The package build reaches only the documented P0.3 baseline where tshy rejects the existing array-valued root source export; it reports no change-specific build diagnostic.
• At the time this evidence was recorded, P0.2 remained the single project-level NEXT item.

P0.2 completion bar

Record accepted decisions covering:

• the purpose or removal of @rxjs/observable;
• the final conceptual and npm package boundaries for fallback, extensions, testing, and the explicit absence of a runtime compatibility product;
• which package owns ambient platform types;
• explicit versus automatic fallback installation;
• behavior when a native constructor or EventTarget.when already exists;
• runtime dependency direction;
• root and subpath import side effects;
• supported realm and server-runtime assumptions for initialization.

Update ARCHITECTURE.md, DECISIONS.md, OPEN_QUESTIONS.md, and package-map diagrams. No implementation is required for this decision step.

P0.2 completion evidence

• Accepted @rxjs/observable-polyfill, rxjs, and @rxjs/test as the three target runtime products. Selected @rxjs/observable for physical removal in P0.3 with no archive, rename, or compatibility reuse.
• Assigned the base ambient Observable, Subscriber, ObservableValue, and EventTarget.when declarations to the independently publishable polyfill package. rxjs owns subpath-scoped Symbol augmentations. At P0.2 acceptance, @rxjs/test remained implementation-neutral; D-043 later gave its explicit cold fixture a public ColdObservable dependency.
• Required every public rxjs entry point to evaluate the conditional polyfill initializer. The root exports non-operator core values without installing the full Symbol catalog; a Symbol subpath installs only its own capability and required kernel dependencies.
• Preserved any existing Observable or EventTarget.when without probing, warning, or replacement. A missing constructor receives the paired RxJS fallback Observable and Subscriber; a missing when is filled only when EventTarget exists.
• Defined the stable Symbol.for('rxjs.observable.polyfill.info.v1') metadata contract, observablePolyfillInfo, and getObservablePolyfillInfo. Frozen { packageName, version } marker-object identity distinguishes installation instances without UUID or crypto requirements.
• Defined initialization as independent per window, iframe, worker, or server isolate, with no realm traversal or transparent foreign-Observable support. The initial capability claim covers browser/worker realms and maintained Node releases; other runtimes and hardened surfaces remain unclaimed.
• Rejected a separate RxJS 7 runtime compatibility product. Retained ColdObservable, Subjects, and Symbol-keyed composition as intentional Next APIs, while making documentation, classified evidence, and migration Skills the supported migration direction. MCP was then optional and deferred; D-046 later excluded it from the accepted migration product.
• Recorded D-039 through D-041, updated the charter, architecture, migration policy, open questions, and package/dependency diagrams, and reserved all runtime, manifest, package-removal, and fixture work for P0.3.

P0.3 completion bar

• packages/observable and its preparation/tooling references are removed.
• The conditional polyfill, read-only version/instance metadata, and detection helper implement D-041 without partially installing on failure.
• Every rxjs public entry initializes its realm; the root exports only the approved non-operator core, and each Symbol subpath installs only its selected capability and kernel dependencies.
• @rxjs/test continues to require an already active realm constructor and never selects or replaces it.
• All selected packages build from a clean checkout on a declared Node runtime.
• Root and subpath exports map to existing sources and generated types.
• Runtime dependencies are declared.
• Source tests are excluded from published output unless deliberately shipped.
• Repository metadata points to the correct package paths.
• At least one ESM and one CommonJS import fixture passes for each claimed package format.
• Fixtures prove missing-global installation and marker metadata, preservation of native/foreign and earlier-version constructors, independent conditional EventTarget.when, direct subpath initialization, a root with no operator Symbols, separate-realm isolation, and clear failure on unsupported frozen targets.

P0.3 completion evidence

• Removed packages/observable, its root preparation gate, README inventory entry, TypeScript bridge, workspace hoist, and lockfile project.
• Made @rxjs/observable-polyfill a conditional per-realm initializer with the frozen D-041 marker and helper. Property changes are preflighted and applied transactionally; unsupported frozen targets fail clearly without retaining an Observable, Subscriber, abort bridge, or EventTarget.when partial.
• Added the declared polyfill runtime dependency to rxjs. Every public source entry reaches the initializer; the root exports only the approved non-operator core, and individual subpaths retain capability-scoped Symbol installation.
• Kept @rxjs/test constructor-neutral and added a missing-global import fixture proving that importing it does not select or install Observable.
• Replaced the invalid root export array with source-backed root and subpath exports. All three manifests emit matching ESM, CommonJS, browser, and webpack runtime/declaration paths, point at their actual repository directories, include only dist plus package metadata, and disable generated self-links.
• Added declaration-consumer and built ESM/CommonJS fixtures for all three packages. Added isolated-process/worker fixtures for missing globals, metadata, foreign and earlier constructors, independent when, direct subpaths, root Symbol isolation, separate realms, and frozen targets.
• On Node 24.12.0, pnpm install --frozen-lockfile, workspace discovery, and pnpm prepare-packages pass. Package build/type/import gates pass for all three products; dry-run packs contain no source specs.
• The polyfill/harness suite passes 49 tests and @rxjs/test passes 73. The pinned strict WPT gate passes 52/52 URLs, 525/525 upstream subtests, and 52/52 exact RxJS identity attestations.
• A clean polyfill build exposed that the prior focused RxJS baseline had consumed a stale fallback artifact. The current diagnostic passes 678/733 focused tests; P0.4 owns the shared selected-implementation lifecycle contract needed to restore that command as a blocking gate.

P0.4 completion bar

• One test contract can run against a selected native implementation and the fallback.
• It covers first subscribe, concurrent observers, late join, individual abort, last-observer abort, restart, completion, error, synchronous reentrancy, teardown registration, teardown order, and thrown observer callbacks.
• Package fixtures detect missing initialization, wrong side effects, duplicate installation, and type visibility.

P0.4 completion evidence

• Added one self-contained lifecycle contract and executed that exact function against the built @rxjs/observable-polyfill package in Node and the native Observable in pinned Chrome 150.0.7871.126.
• The five grouped cases cover first activation, concurrent observers, late join, individual and last-observer abort, restart, completion, error, synchronous reentrancy, late teardown registration, reverse teardown order, and thrown next, error, and complete callbacks.
• Added mixed ESM/CommonJS duplicate-install fixtures in both load orders. They prove that a second package dialect preserves the selected Observable, paired Subscriber, EventTarget.when, abort bridge, and frozen marker object identities.
• The existing missing-global, foreign/native preservation, frozen-target, declaration-consumer, and built ESM/CommonJS fixtures continue to prove initialization, side-effect, and type-visibility boundaries for all three packages.
• Added the lifecycle contract to both pinned and latest-Chrome Observable CI jobs. At P0.4 completion, the reusable browser-script runner passed the pinned strict WPT gate: 52/52 URLs, 525/525 upstream subtests, and 52/52 implementation attestations. D-042 temporarily changed the void notification contract; D-045 supersedes it and restores the complete required result.
• Reconciled the stale focused suite: 101 files and 733/733 tests pass against the rebuilt fallback. A correction audit removed invalid platform-Observable wrappers around RxJS 7 arbitrary-subscribable inputs. The complete cold migration audit now passes 2,323/2,338 and the fallback audit passes 2,316/2,338. The 15 shared legacy-input failures and seven additional fallback retry/shared-error failures remain executable product evidence; the strict all-mode command therefore remains nonzero without weakening or quarantining those cases.

P0.5 completion bar

• WICG/observable commit d74bace7cf80200a01c81cfe20961e29ac7fa3d8 and its spec.bs are recorded as the written reference for understanding behavior and diagnosing failures.
• web-platform-tests/wpt commit 6a009d73f0d315941b90cac13a9523a2a08c631b is recorded as the executable success gate.
• pnpm run test:wpt passes 52/52 URLs, 525/525 upstream subtests, and 52/52 exact RxJS implementation checks, with no upstream edits, skips, failures, errors, timeouts, not-run results, or accepted-failure metadata.
• RxJS maintainers own the pins. A new WPT revision requires an explicit exact commit, review of every changed Observable test and realm pattern, verified provenance, and the same complete all-pass result before adoption.

P0.5 completion evidence

• Restored the Web IDL required-argument check before the fallback subscriber's active-state check. Missing arguments now throw even after closure, while explicit undefined is delivered normally.
• Required an explicit TypeScript value for every platform Subscriber, and mechanically changed affected Subscriber<void> producers to next(undefined) without changing unrelated Subject<void> contracts.
• Superseded D-042 with D-045 and recorded the roles, exact revisions, ownership, and update policy for the WICG written reference and executable WPT suite.
• Passed 51 focused polyfill tests, the polyfill declaration-consumer check, 733 focused RxJS tests, 75 @rxjs/test tests, pinned-import provenance for all 52 URLs, and the five-case shared native/fallback lifecycle contract.
• Against Chrome 150.0.7871.126, both pnpm run test:wpt:baseline and strict pnpm run test:wpt passed with 52/52 OK URLs, 525/525 passing upstream subtests, and 52/52 passing exact RxJS implementation attestations.

Phase 1 — Platform fallback correctness

Status	ID	Outcome
DONE	P1.1	Close conditional-installation conformance gaps exposed after the P0.3 package implementation
DONE	P1.2	Bring core subscription, abort, teardown, error-reporting, and Observable.from behavior to the pinned baseline
DONE	P1.3	Bring native platform methods and EventTarget.when to the pinned baseline
DONE	P1.4a	Build the attested Observable WPT test harness and record its stable current-behavior baseline
DONE	P1.4b	Make the fallback pass the pinned Observable WPT suite

P1.1–P1.3 closure evidence

• P1.1 was already satisfied by P0.3's transactional conditional initializer and package/realm fixtures, plus P0.4's mixed ESM/CommonJS duplicate-install and native/fallback lifecycle coverage. No remaining installation gap was identified that belonged to the fallback layer.
• P1.2 and P1.3 were already satisfied by P1.4b and P0.5. The attested strict WPT gate covers the selected subscription, abort, teardown, error-reporting, Observable.from, native-method, and EventTarget.when behavior and permits no accepted failure metadata.
• A 2026-08-01 closure audit passed all 51 focused polyfill/harness tests, the package build, declaration consumer, ESM/CommonJS import and duplicate-load fixtures, and the five-case lifecycle contract against both the packaged fallback and native Observable in Chrome 150.0.7871.126.
• The same audit passed strict offline WPT conformance with 52/52 OK URLs, 525/525 passing upstream subtests, and 52/52 passing exact-RxJS-identity attestations. The implementation bundle was 697e4a814b79aa9d443f93cad87144c7a2984df75c26901f1e4600e18c0f1745.
• Older fallback-only failures in the RxJS 7 ported corpus concern RxJS extension and migration-evidence behavior. They do not contradict the platform fallback's pinned conformance and remain outside Phase 1 rather than motivating duplicate fallback implementation.

P1.4a completion bar

• WPT is pinned to 6a009d73f0d315941b90cac13a9523a2a08c631b, and only its approved Observable dependency closure is vendored byte-for-byte with Git-blob provenance and an exact generated URL inventory.
• The ignored execution tree instruments window, dedicated-worker, same-origin iframe, and Web IDL coverage without modifying imported WPT sources.
• Every expected URL has exactly one unsuppressible passing attestation proving the active constructor, subscribe, and EventTarget.prototype.when are the exact RxJS bundle identities, differ from captured native references, and report the expected bundle SHA-256.
• Negative controls prove native leakage, restored native identities, a wrong bundle ID, a missing worker/iframe attestation, and allowlisted attestation results all fail the independent auditor.
• The official runner, exact Chrome for Testing 150.0.7871.126, and matching driver are checksum-locked and cached so a warm second run needs no network.
• The default test:wpt command is a strict conformance gate with readable terminal failures. The explicitly named baseline diagnostic distinguishes harness/report failures from known Observable behavior failures, and three consecutive complete runs must agree before granular expectations are accepted.
• Importer, provenance, inventory, realm-pattern, and report-auditor unit tests pass; CI uploads raw reports, diffs, logs, inventories, and per-realm identities.
• No production source, public API, ambient type, export, runtime dependency, installation contract, or migration-contract changes.

P1.4b completion bar

• test:wpt passes against the approved WPT and browser revisions.
• Behavior fixes remain within the accepted native-first installation and platform-lifecycle architecture.
• Each obsolete failure expectation is deliberately removed as its behavior is corrected.

P1.4b completion evidence

• Corrected the platform lifecycle without introducing RxJS 7 cold semantics: subscriber closure now preserves abort reasons, aborts before LIFO teardown, executes late teardowns immediately, retains shared/ref-counted production, and reports late or unhandled errors through the platform-shaped global path.
• Added the non-constructible global Subscriber surface and required Web IDL names, tags, descriptors, argument checks, and detached-realm behavior.
• Corrected Observable.from conversion order and Observable identity, Promise completion, sync/async iterator acquisition, close reasons, protocol errors, pending-result behavior, and microtask timing. The explicit async iterator loop documents why for await...of cannot preserve these observable protocol details.
• Corrected pinned platform behavior in takeUntil, take, drop, every, reduce, inspect, and Promise-returning consumers without adding string-named legacy RxJS behavior.
• Added a scoped abort-algorithm bridge because public abort event listeners run too late to model the DOM-standard cancellation order. Signals without registered Observable work continue through the captured native abort implementation.
• Fixed generated attestation registration so upstream setup() properties are established before the identity subtest starts the reporting protocol. The vendored WPT sources remain byte-for-byte unchanged.
• Passed the strict test:wpt command with 52/52 URLs OK, 525/525 upstream subtests PASS, and 52/52 exact-identity attestations in Chrome for Testing 150.0.7871.126. The passing implementation bundle for that run was 778edfac15639cd4531a590cd36450ad273a0a692df2e8a1436564dca3cb89f8.
• Regenerated the baseline after three further identical complete attested runs, deleting all 25 obsolete failure .ini files. The package's 49 focused tests and pinned-import verification pass.
• Reconfirmed the existing P0.3 blocker: package build and lint still fail because the ambient declarations are disconnected from the package and root TypeScript configurations. No package-boundary work was folded into this conformance slice.

P1.4a completion evidence

• Vendored exactly 29 files under dom/observable/tentative/ and eight source-derived support files: 37 upstream files and 716,874 bytes total. Provenance records every Git blob and SHA-256; the generated inventory contains only 52 /dom/observable/tentative/ URLs.
• Verified all 52 URLs run exactly once and all 52 exact-identity attestations pass against implementation bundle c03cd03b65e0337e0f73f202602529b87db1c2dfb1271bbd2e6857477c258e2a. Evidence includes dedicated-worker, Web IDL, and four combined window/same-origin-iframe attestations covering all nine reviewed child realms.
• Recorded the initial granular expectation metadata after three identical complete runs. That pre-P1.4b baseline was 33 OK, 15 ERROR, and 4 TIMEOUT top-level results, with 314 PASS, 159 FAIL, 8 TIMEOUT, and 6 NOTRUN reported upstream subtests.
• Passed 45 harness unit tests covering import/provenance/inventory/closure, realm review, report completeness and classification, stability, and every required attestation negative control. The package's four existing source tests also pass.
• Passed wpt:doctor, wpt:verify-import, test:wpt:baseline, and two further consecutive narrowed-closure runs with RXJS_WPT_OFFLINE=1. Chrome for Testing and ChromeDriver were both exactly 150.0.7871.126.
• Added blocking path-filtered strict-conformance CI, a scheduled advisory latest-Chrome job, cache keys, fixed concurrency/timeouts, and always-uploaded raw reports, concise summaries, logs, inventories, and per-realm identity evidence.
• Declared Node 24 tooling support and moved both WPT workflow jobs to Node 24; import verification, harness tests, doctor, and browser execution work on Node 24.12.0 without bypassing the engine check.
• Made no production source, public API, ambient type, export, runtime dependency, installation-contract, or migration-contract change.

Phase exit:

• the fallback passes the agreed local lifecycle/conversion suite;
• native preservation is enforced;
• no selected WPT difference or accepted failure remains;
• no main-library work depends on undocumented fallback behavior.

Phase 2 — Symbol extension kernel

Status	ID	Outcome
DONE	P2.1	Decide Symbol identity, versioning, duplicate-install, realm, and collision policy
DONE	P2.2	Implement one typed extension installer for static and instance capabilities
DONE	P2.3	Define constructor preservation, input conversion, cancellation, and error-forwarding helpers
DONE	P2.4	Convert a small representative operator set to the kernel and validate native/fallback parity

P2.1 completion bar

• Public Symbol identity across module evaluations, package copies, versions, and realms is explicit.
• The public/global-registry boundary and construction-protocol exception are complete, including the former buffer exception.
• Duplicate-version coexistence, exact-key collision handling, extension descriptors, unsupported targets, and package side effects have accepted policies that P2.2 and P2.4 can implement and verify.

P2.1 completion evidence

• Accepted D-048: public capabilities use exact module-owned Symbols; only the D-037 construction ABI and D-041 fallback metadata use namespaced global registry keys.
• Defined independently evaluated ESM/CommonJS dialects and version-skewed copies as isolated public capabilities that may coexist without load-order replacement. No cross-realm traversal or public Symbol identity promise is made.
• Removed Symbol.for('buffer'); the 12-case focused buffer suite passes and a source audit finds no public Symbol.for capability outside create.
• Chose transactional, non-enumerable installation with exact-key conflict rejection, named unsupported-target diagnostics, explicit side-effectful package metadata, and bundler fixtures as the P2.2/P2.4 implementation bar.

P2.2 completion bar

• One internal typed API installs static-only, instance-only, or paired exact Symbol capabilities.
• Installation is idempotent for the identical value, rejects an occupied exact key, reports non-extensible targets clearly, and cannot leave a partially installed paired capability.
• Extension descriptors and package side-effect metadata implement D-048.

P2.2 completion evidence

• Added installObservableExtension with typed static and instance implementations, constructor/prototype preflight, atomic definition and rollback, and capability-named diagnostics.
• Defined public capability properties as non-enumerable, writable, and configurable. Exact-key conflicts are never overwritten.
• Added six focused cases covering paired installation, descriptors, idempotence, conflict preflight, frozen targets, rollback, and invalid empty requests; all pass.
• Added explicit sideEffects: true package metadata and passed the strict RxJS source type-check. P2.4 owns the packed bundler proof and pilot adoption.

P2.3 completion bar

• Derived construction and platform input conversion have distinct accepted contracts for ordinary receivers, compatible subclasses, intentional overrides, and unsupported incompatible or foreign receivers.
• One helper pattern owns downstream cancellation, optional operator-local cancellation, terminal forwarding, and synchronous exception forwarding.
• No helper restores RxJS 7 arbitrary subscribables, Subscription facades, schedulers, string-named methods, or cross-realm promises.

P2.3 completion evidence

• Accepted D-049: results use the receiver's versioned [create]; inputs use the active realm's platform Observable.from; incompatible constructors and generic unrelated-object borrowing are unsupported.
• Added internal helpers for receiver-driven derived creation, active-realm conversion, signal-owned source subscription, and discriminated synchronous error forwarding.
• Removed the two unused pre-kernel helpers that constructed through an unchecked .constructor or provided an unadopted forwarding wrapper.
• Six helper-contract tests cover compatible subclass creation, explicit protocol overrides, base-platform conversion, cancellation and teardown, terminal forwarding, and thrown callback errors. All pass with the strict RxJS source type-check.

Representative pilot set:

• one native-overlapping synchronous operator such as map or filter, with both the platform string form and RxJS Symbol form tested;
• one RxJS-only synchronous operator such as scan;
• one higher-order operator such as switchMap;
• one time-based operator such as timeout;
• one static factory such as timer;
• pipe or its approved replacement.

P2.4 completion evidence

• Migrated the exact map, scan, switchMap, timeout, timer, and pipe capabilities to the common installer and helper pattern. The set covers a native-overlap operator, synchronous state, higher-order cancellation, time-based recovery, a static factory, and both static and instance composition.
• Added focused behavior and descriptor coverage, including shared activation and restart, compatible subclass construction, exact Symbol identity, string-method non-interference, cancellation-before-recovery, and thrown callback forwarding. All 756 focused RxJS source tests pass.
• Passed strict source typing, package builds, declaration consumers, ESM and CommonJS imports, mixed-dialect duplicate-capability coexistence, frozen target failure, and bundler retention/tree-shaking fixtures.
• Passed the same eight-case extension-kernel contract against the packaged fallback and native Observable in Chrome 150.0.7871.126; the CI workflow now runs that contract with both pinned and current browser evidence.
• The targeted migrated map, scan, switchMap, timeout, and timer evidence passes 97/98 registrations in each mode. The sole exception is the already-classified compatibility-only switchMap arbitrary-subscribable case, which D-049 intentionally leaves unsupported. The complete strict audits remain nonzero only on the broader Phase 3/4 restoration backlog: 2,296/2,338 cold and 2,321/2,343 fallback registrations pass.

Phase exit:

• the pilot operators share one implementation pattern;
• direct subpath imports are deterministic;
• duplicate installation and type augmentation are tested;
• no string-named RxJS property is added to the platform API;
• the native-overlap pilot proves that the string-named platform method remains untouched and the corresponding Symbol-keyed RxJS form coexists with it;
• any additional behavior in the RxJS form is documented and tested.

Phase 3 — Operator restoration and parity

Status	ID	Outcome
DONE	P3.1	Audit the existing API/evidence inventory and rebase the remaining restoration by migration value
DONE	P3.2	Audit and maintain the migration-evidence ledger
DONE	P3.3	Restore operators in small families using the extension kernel
DONE	P3.4	Classify, retain, or rewrite former RxJS 7 tests for each restored family

Do not use “all former tests pass” as an unqualified milestone. The gate is that every supported API has portable or rewritten evidence and every divergence is explicit.

P3.1 completion bar

• Reconcile the published subpath inventory, exact Symbol catalog, parity capability registry, and executable RxJS 7 evidence without recreating the Phase 0 audit.
• Separate already restored behavior from extension-kernel adoption, evidence metadata, and intentional compatibility gaps.
• Order the remaining work by migration value while preserving small operator-family commits and the accepted platform lifecycle.

P3.1 completion evidence

• Audited 117 public source subpaths and 97 exact public extension Symbols. Six pilot capabilities (map, scan, switchMap, timeout, timer, and pipe) already use the transactional installer; the other 91 exact capabilities retain pre-kernel direct installation and are implementation migration work rather than missing APIs.
• Reconciled the versioned parity registry: it describes 111 RxJS 7 operators, 21 creation/static functions, and 29 public values. The source-pinned corpus retains 2,338 executable registrations, with 2,296 reviewed cold passes and 2,316 reviewed fallback passes before the Phase 2 pilot additions.
• Confirmed that the remaining ordinary failures are already classified input conversion or lifecycle evidence. There are no missing-api or unsupported-or-obsolete registrations hidden from collection.
• Rebased P3.3 into four migration-value families: portable synchronous and query operators; combination, flattening, recovery, and recursive operators; host-time, windowing, and static factories; then sharing, connection, Subject-adjacent, and async-iteration capabilities. Each family must adopt the installer without changing its accepted behavior or public Symbol.
• Assigned P3.2 to materialize the existing parity and package data as one validated ledger before code migration. P3.4 will then audit every family against focused, cold, fallback, native/fallback-kernel, type, and package evidence rather than treating a raw RxJS 7 pass total as the milestone.

P3.2 completion bar

• One machine-readable ledger contains every prioritized RxJS 7 operator, creation function, and public value exactly once.
• Each row records the RxJS 7 import, evidence, Next surface, sharing and cancellation models, classification, type status, migration action, and controlling decision or open question.
• Generation is deterministic, rejects incomplete or duplicate rows, and produces a readable checked-in summary.

P3.2 completion evidence

• Added one generated 161-entry ledger covering all 111 registry operators, 21 creation/static functions, and 29 values, types, and standalone functions exactly once.
• Joined each entry to the source-pinned manifest. The ledger retains 2,271 distinct evidence cases and records the seven entries with no direct case as uncovered rather than implying support.
• Recorded former import path, exact case IDs and source files, Next mapping and status, sharing and cancellation models, test classifications, type status, migration action, adapter, rationale, and controlling decisions.
• Added deterministic generation and check commands. Validation rejects incomplete entries, duplicate IDs, total drift, and evidence-count drift; both the machine-readable JSON and readable Markdown views must be current.
• Updated the compatibility policy to make the generated ledger authoritative for Phase 3. P4.2 retains responsibility for final prioritized type status, and P4.3 retains responsibility for the unsupported-surface catalog.

P3.3 completion bar

• Every public exact Symbol capability installs through the D-048 transactional installer; no direct public constructor or prototype assignment remains.
• Each migrated implementation preserves its exact Symbol, ambient augmentation, derived-construction policy, input boundary, cancellation, error, and package subpath behavior.
• Small family commits pass focused source typing and the narrowest relevant behavior tests before the next family begins.

P3.3 completion evidence

• Migrated all 91 pre-kernel exact capabilities to the common installer in nine focused family commits. Together with the six Phase 2 pilots, all 97 current exact public Symbols now install transactionally.
• Kept every public Symbol, ambient augmentation, implementation body, package subpath, and accepted behavior intact. Paired static/instance capabilities now install atomically rather than through two assignments.
• Verified each family with its focused behavior suite: state and query, terminal, combination, higher-order, buffering/windowing, timed/static, sharing/connection, and async-iteration coverage all passed.
• Added a blocking source audit to both the ordinary RxJS unit gate and package gate. It requires every exact public Symbol to import and use the common installer and rejects direct public constructor or prototype assignment.
• The complete focused source suite passes 107 files and 756 tests. The public declaration consumer and multi-dialect package build pass. The broad raw TypeScript project command remains unsuitable because it includes the known generated parity-source diagnostics; the package build and public consumer are the accepted type gates for this slice.

P3.4 completion bar

• Every supported family links focused evidence and a classified RxJS 7 claim; every compatibility-only or intentional difference stays explicit.
• Complete cold and fallback audits account for all registrations without skip, quarantine, or expected-failure inversion; native/fallback kernel, type, build, and package gates cover the public installation contract.
• Phase 3 documentation records exact results and advances the sole NEXT marker to P4.1.

P3.4 completion evidence

• Re-ran all 2,338 source-pinned registrations in both complete audit modes with ordinary Vitest semantics and no skipped or pending cases. Cold mode records 2,299 passes and 39 failures; fallback mode records 2,316 passes and 22 failures.
• Corrected three cold evidence claims that had accidentally carried producer-per-observer or ColdObservable subclass expectations past a hot-derived platform boundary. Focused partition, groupBy, and buffering suites now pass all 63 registrations.
• Audited every remaining failure by exact case ID. Cold mode retains 24 intentional D-013/D-043 lifecycle divergences and 15 compatibility-only arbitrary-subscribable inputs; fallback mode retains seven of those lifecycle divergences and the same 15 compatibility-only inputs. No failing portable or harness-rewrite claim remains.
• Regenerated and validated the classification ledger and recorded the reviewed case-ID pass baseline. Focused source, native/fallback kernel, public type, package build, installation, and import gates cover the restored public families and installation contract.
• Phase 3 is complete. P4.1 is the sole project-level NEXT item.

Phase 4 — Intentional API and migration contracts

Status	ID	Outcome
DONE	P4.1	Stabilize intentional cold, Subject, and Symbol-composition APIs on their own Next contracts
DONE	P4.2	Complete the migration-evidence ledger for prioritized RxJS 7 public APIs
DONE	P4.3	Record unsupported RxJS 7 imports, types, schedulers, interop, and deprecated aliases
DONE	P4.4	Add representative migration fixtures for the accepted API and lifecycle boundaries
DONE	P4.I1	Remove the common public-extension installer and restore direct exact-Symbol assignment

Phase exit:

• intentional Next APIs are explicit in imports and types;
• producer, sharing, and cancellation semantics are documented directly;
• every migration mapping links behavioral evidence and required source work;
• unsupported RxJS 7 categories are documented without an emulation promise;
• public Symbol capabilities install directly without a shared runtime installer or transactional installation contract.

P4.1 completion evidence

• Recorded D-050 and stabilized the root/subpath contracts for ColdObservable, Subject, AsyncSubject, the lowercase behavior/replay factories, the advanced abstract PerSubscriptionSubjectBase, and exact Symbol-keyed static/instance pipe.
• Kept lifecycle explicit: cold direct subscriptions create independent work; every Subject instance is hot; observer-local replay does not imply a cold producer; native-method results cross to the shared platform lifecycle.
• Documented the public type boundary: Symbol results are declared as Observable<T> even when D-037 selects ColdObservable at runtime, so producer lifecycle is not inferred from the widened result type.
• Added a published-declaration consumer covering every intentional API through both root and explicit subpath imports, factory/config types, advanced-base subclassing, Subject views, and typed static/instance composition. The public type gate passes.
• Marked P4.1 DONE and advanced P4.2 as the sole project-level NEXT item.

P4.2 completion evidence

• Completed all 161 prioritized ledger rows: 154 use direct source-pinned executable evidence, four alias/equivalent rows link canonical executable evidence, and three rows name pinned non-marble or focused Next evidence. No row remains uncovered.
• Added reviewed cold and fallback totals to every linked evidence set. The generator rejects inconsistent totals and any failing linked case not classified compatibility-only or intentional-divergence.
• Finalized public type status as 150 changed, seven preserved, and four compatibility-only rows; no status remains deferred. “Changed” records an intentional published mapping, not an assertion of RxJS 7 signature parity.
• Linked D-050 to the stabilized cold, Subject-family, and exact Symbol composition rows and retained the exact case IDs, source files, local evidence, adapter, migration action, lifecycle, cancellation, and controlling decisions in the machine-readable ledger.
• Regeneration and freshness checks pass. Marked P4.2 DONE and advanced P4.3 as the sole project-level NEXT item.

P4.3 completion evidence

• Added a pinned machine-readable catalog with 27 reviewed groups covering 147 named RxJS 7 import, public-type, scheduler, interop, and deprecated-alias surfaces. Every group records a disposition, migration direction, rationale, and controlling decisions without promising a compatibility runtime.
• Generated the readable unsupported-surface guide from that authority and defined explicit replace, manual-review, unsupported, test-only, and removed dispositions. Automation may proceed only for an evidence-ledger-backed replace mapping and must stop for every other disposition.
• Added a validation and freshness gate tied to the pinned executable manifest. It rejects category or critical-surface omissions, duplicate entries, stale generated guidance, source-pin drift, and any scheduler-bearing capability that is missing from the scheduler review inventory.
• Added the unsupported-surface check beside the completed migration ledger in the ordinary package gate. Marked P4.3 DONE and advanced P4.4 as the sole project-level NEXT item.

P4.4 completion evidence

• Added four explicit migration fixtures for the accepted cold-preserving, platform-shared, hot-Subject, and unsupported scheduler/interop outcomes. Each fixture names its pinned RxJS 7 source, target lifecycle, cancellation owner, controlling decisions, linked ledger/catalog rows, and unsafe negative control.
• Type-checked all three migrated targets against the current public RxJS Next declarations. The safe-stop fixture deliberately has no invented target and links the scheduler, legacy interop, arbitrary-subscribable, and multicasting review categories instead.
• Executed producer multiplicity, final-observer cancellation and restart, behavior/replay state, exact Symbol construction, and lifecycle-swap negative controls. The full migration package passes 17 files and 166 tests.
• Marked P4.4 DONE. The original Phase 4 contract work is complete; P4.I1 is the sole project-level NEXT item under the later D-051 simplification.

P4.I1 completion bar

• Replace every installObservableExtension call with direct assignment to Observable, Observable.prototype, or the already established constructor target under the module's own exported exact Symbol.
• Delete the installer implementation and its installer-only tests. Do not replace them with another runtime abstraction for collision checks, idempotence, descriptor customization, extensibility preflight, paired rollback, or named mutation diagnostics.
• Revise the source audit to require exact module-owned Symbols, direct installation under those Symbols, and no RxJS-specific string-named constructor or prototype additions. Preserve the D-037 shared construction protocol as a separate accepted exception.
• Update focused kernel, ESM/CommonJS, duplicate-dialect, and bundler fixtures to the D-051 contract. Remove extension-specific frozen-target cases and assertions for non-enumerable public Symbol descriptors, transactional paired installation, or custom unsupported-target errors; retain D-041's separate fallback-acquisition frozen-target coverage.
• Preserve public Symbols, ambient declarations, operator implementations, construction and cancellation behavior, subpath side effects, root operator isolation, and platform string-method non-interference.
• Record the minified and compressed bundle delta for a representative direct extension import and run the narrow focused, source-audit, type, package, import, bundler, and native/fallback kernel gates. WPT is not a gate for the RxJS-owned Symbol descriptor or installation mechanism.

P4.I1 completion evidence

• Migrated all 97 exact public Symbol capabilities in ten small family commits while preserving the public Symbols, ambient declarations, implementation bodies, construction, cancellation, lifecycle, and string-method isolation.
• Deleted installObservableExtension and its six installer-only tests. The replacement audit requires direct installation on the declared static or instance target, rejects installer references and RxJS-specific string-named additions, and keeps D-037's construction protocol separate.
• Updated descriptor, ESM/CommonJS, duplicate-dialect, bundler, and kernel fixtures to the D-051 contract. The D-041 frozen fallback-acquisition fixture remains; unsupported extension targets receive ordinary assignment behavior.
• With esbuild 0.19.11, the minified import 'rxjs/map' browser ESM bundle fell from 15,726 to 14,447 bytes (-1,279; -8.1%); gzip fell from 4,584 to 4,244 (-340; -7.4%); and Brotli fell from 4,126 to 3,819 (-307; -7.4%). The root-only control stayed byte-identical at 19,650 minified, 5,638 gzip, and 5,050 Brotli bytes.
• Passed the 106-file/750-test focused source suite, the 97-Symbol source audit, public types, build, package/import and bundler fixtures, and all eight kernel cases against the packaged fallback and native Observable in Chrome 150.0.7871.126.
• Marked P4.I1 DONE, completed Phase 4, and advanced P5.1 as the sole project-level NEXT item.

Phase 5 — Migration experience and AI enablement

Status	ID	Outcome
DONE	P5.1	Write migration guidance from the migration-evidence ledger and accepted divergences
DONE	P5.2	Validate mechanical and semantic migration steps on representative applications
DEFERRED	P5.3	Evaluate later RxJS usage Skills beyond the canonical migration workflow
DEFERRED	P5.4	Evaluate optional integrations only under a new evidence-backed product decision

AI tools must consume versioned project knowledge and produce reviewable changes. They must not infer migration safety solely from matching operator names.

P5.1 completion bar

• One package-local guide gives an RxJS 7 user a task-oriented path from a green baseline through lifecycle classification, bounded transformation, semantic review, and verified closeout.
• The guide covers Symbol imports and composition, platform versus producer-per-direct-subscription behavior, signal cancellation, teardown, platform input conversion, Subjects, schedulers, testing, and safe-stop conditions without implying a compatibility runtime.
• The generated migration-evidence ledger and unsupported-surface catalog live inside the rxjs package container and retain deterministic freshness gates.
• Public examples type-check, package-local and repository-governance links resolve, and no apps/rxjs.dev path or documentation-site command changes.

P5.1 completion evidence

• Added packages/rxjs/MIGRATION.md as the concise golden path, including the explicit four-way lifecycle choice, direct Symbol composition examples, AbortSignal cancellation and teardown examples, conversion/realm limits, Subject and scheduler guidance, automation boundaries, and a completion checklist.
• Moved the generated migration-evidence ledger and unsupported-surface guide to packages/rxjs/docs, changed their generators and freshness gates to the package-local paths, and updated repository-wide compatibility references.
• Added a strict declaration consumer for the guide's Symbol chaining, Symbol-keyed pipe, subscription signal, teardown, and ColdObservable examples.
• Accepted D-007 with RxJS 9 and 9.0.0-beta.0, and recorded D-052 for package-local documentation plus the explicit rxjs.dev exclusion.
• Passed the public type consumer, migration ledger/catalog freshness gates, parity-document freshness, relative-link validation, formatting/diff checks, and a path audit confirming no apps/rxjs.dev change. Advanced P5.2 as the sole NEXT item.

P5.2 completion bar

• Representative application and library repositories exercise strong and weak coverage, multiple test frameworks, cold-preserving, platform-shared, mixed, Subject, and unsupported migration outcomes.
• Mechanical steps are covered by exact transforms, compilation, idempotence, diagnostics, refusal safety, and behavior; semantic steps are covered by explicit contract decisions, held-out behavior, protected tests, and intentional-divergence review.
• At least one unsupported or insufficiently covered repository proves a timely safe stop before target installation or migration writes.
• Offline grading verifies the retained qualification records and artifacts. Any live-model claim remains bounded to the recorded harness, model, versions, repositories, and outcome gates.

P5.2 completion evidence

• Audited P0.M3, P4.4, and P0.M5 against the package-local migration guide and recorded the guide-to-evidence map in packages/migrate/docs/MIGRATION_QUALIFICATION.md.
• Confirmed four immutable RxJS 7 application/library seeds spanning Vitest, Mocha, Jest, strong and weak coverage, cold, shared platform, mixed, and unsupported target contracts. Three complete and one makes the required safe stop before target installation or migration writes.
• Re-ran all 17 @rxjs/migrate suites and 166 tests, including deterministic mechanical behavior, the four P4.4 lifecycle outcomes, seed oracles, the 14-family semantic grader, the closed four-record matrix, and artifact mutation controls. All pass.
• Passed the migration package build, declaration consumer, ESM/CommonJS imports, and publication dry run. The audit adds no new cross-harness or automatic-migration claim and does not rewrite the immutable 8.0.0-alpha.14 qualification snapshot.
• Marked P5.2 DONE, completed the planned Phase 5 work, left P5.3 and P5.4 intentionally deferred, and advanced P6.1 as the sole NEXT item.

Phase 6 — Release readiness

Status	ID	Outcome
DONE	P6.1	Finalize version naming, supported environments, support policy, and release channels
DONE	P6.2	Complete package, type, bundle, performance, and conformance gates
DONE	P6.3	Publish package-local API, migration, and contributor documentation
DONE	P6.4	Run pre-release adoption, resolve blockers, and approve the major release
DONE	P6.5	Complete the terminal plan, verification, and documentation-site exclusion audit
DONE	P6.6	Centralize eligible source subscriptions and record bundle-size evidence
DONE	P6.7	Use direct [create] construction and record bundle-size evidence
DONE	P6.8	Complete RxJS 9 CI coverage and validate the resulting pull-request workflow matrix
DONE	P6.9	Validate the first live dependency-review and Scorecard runs on GitHub
NEXT	P6.10	Publish and verify the first beta with the interactive release command
DONE	P6.11	Remove inherited callback thisArg parameters and direct-call callback hot paths

P6.9 completion bar

• The root README exposes authoritative master CI, release-readiness, WPT, npm channel, download, and license signals plus factual native-Observable and Symbol-extension badges. The package security-assurance document presents OpenSSF Scorecard in context rather than as a headline release claim.
• Prepublication wording does not claim that 9.0.0-beta.0 is already on npm or direct readers to install the older prerelease currently under next.
• A SHA-pinned Scorecard workflow publishes attested results and SARIF from master on push and a weekly schedule; a SHA-pinned pull-request dependency review blocks new moderate-or-higher runtime and development vulnerabilities.
• Active-workflow formatting, release coherence, documentation links, badge targets, and the no-runtime/no-rxjs.dev scope boundary are verified. The first default-branch Scorecard publication, code-scanning result, and dependency-review required-check rule are verified live.

P6.9 implementation evidence

• Initially added ten source-linked README badges: three master workflow signals, OpenSSF Scorecard, npm latest and next, monthly downloads, license, and two architecture-backed native/Symbol claims. P6.10 later moved Scorecard to the security-assurance document because it is secondary to release evidence. The image endpoints returned HTTP 200; workflow and Scorecard destinations resolved, while npm's canonical web pages returned their expected automated-client 403 response.
• Corrected the unpublished-beta wording, removed the unsafe rxjs@next installation command, and retained the planned API example without claiming that 9.0.0-beta.0 is already available.
• Added full-SHA-pinned Scorecard v2.4.4 and dependency-review v5.0.0 workflows. Local YAML parsing, active-workflow formatting, documentation links, all 24 release-check tests, release coherence, and diff checks pass; apps/rxjs.dev and runtime/package source are unchanged.
• Dependency review passed on merged PR #7613. After merge, master run 30919216705 published the first successful Scorecard result, CodeQL passed, and the repository's protected-branch status list included Dependency review. This completes P6.9; the later P6.10 ruleset migration must preserve that required check rather than moving it into the master-only wait list.

P6.10 completion bar

• The public runbook and security-assurance document state that Ben is the sole author, reviewer, merger, release operator, and security responder. No team, approving review, CODEOWNER, environment reviewer, or succession role is required or implied.
• pnpm release:beta <9.0.0-beta.N> is the sole publication entry. It refuses live operation outside a clean, remote-synchronized master, in CI, without an interactive terminal, or with NPM_TOKEN/NODE_AUTH_TOKEN present.
• The command requires synchronized package metadata, runs release and package gates, packs all four packages, prints SHA-512 integrities, runs npm publish dry runs, and requires the exact version as confirmation before publication.
• npm interactive OTP/WebAuthn authorizes each public package. The three scoped packages publish before rxjs; every registry integrity and next tag must match, and rxjs@latest must remain on RxJS 7.
• A partial retry skips an immutable package version only when the registry integrity equals the newly packed tarball. Any byte change requires a fresh beta version.
• Root vulnerability paths are classified; only time-bounded apps/rxjs.dev exceptions remain. Package publishing access disallows automation tokens after the three new package records are initialized.
• P6.10 remains active through the interactive publication, registry integrity verification, npm channel verification, and immutable GitHub Release.

P6.10 implementation evidence

• D-057 records the abandoned staged design and its unsupported new-package bootstrap assumption; D-058 accepts the simpler local boundary. The release App was deleted before any npm stage or publication.
• Removed the release App, automated release PR, two-build candidate, qualification, OIDC staging, typed authorization, doctor, staged-comment, and automated-finalizer workflows and scripts.
• Added the tested interactive beta command, exact publish order, clean-master and no-token guards, dry runs, resumable integrity checks, channel checks, and an operator-focused runbook. Existing CI, CodeQL, dependency review, OSV, package gates, and release-readiness coverage remain.
• Local verification is recorded in the P6.10 session entry. Live npm OTP/WebAuthn publication, package-access hardening, registry verification, and the immutable GitHub Release remain required before DONE.

P6.11 completion bar

• Every RxJS Next callback API is audited for a separate thisArg; the exact affected API inventory is recorded.
• Receiver-aware overloads, implementation parameters, and .call dispatch are removed from all affected APIs without changing platform-polyfill receiver binding required by Web IDL and iterator protocols.
• Focused and migrated tests no longer assert library-provided callback receivers; source-pinned RxJS 7 identifiers remain historical evidence, and active replacements demonstrate closures or Function.prototype.bind.
• Focused source, migrated cold/polyfill, public type, package, kernel, migration-document freshness, lint, and diff gates pass.

P6.11 completion evidence

• Audited RxJS production source and found exactly six APIs with a separate callback receiver argument: Symbol-keyed every, filter, find, findIndex, and map, plus static Symbol-keyed partition. Removed every receiver-aware overload, implementation parameter, and callback .call.
• Replaced active receiver assertions with ordinary value/index/source tests, closures, or explicitly bound callbacks. Added public declaration checks that reject the removed second or third arguments. Classified all six source-pinned RxJS 7 cases as intentional divergences and made their executable migrated programs bind callbacks instead of exercising the removed overload.
• Passed all 750 RxJS source tests, the 76 affected cold and 76 affected polyfill migrated tests, lint with zero errors, public types, package build, migration-document freshness, 97-Symbol installation, and all package import fixtures. Exact complete audits retained the reviewed 2,299/39 cold and 2,316/22 polyfill outcomes.
• Passed the eight-case packaged fallback/native Chrome 150 kernel contract. The release performance gate measured approximately 63.0 million map values and 140,627 cancellations per second. Marked P6.11 DONE; the completed execution queue has no NEXT marker.

P6.1 completion bar

• All release packages share the accepted RxJS 9 prerelease identity and exact internal dependency versions; runtime metadata, fixtures, lockfile, and the checked-in migration Skill adapter agree.
• One accepted decision defines blocking and advisory Node lanes, desktop and Mobile Safari plus evergreen browser coverage, Webpack, Deno, Bun, npm channels, RxJS 7 maintenance, and explicitly unclaimed environments.
• Published JavaScript is ESM-only. Browser, Webpack, import, and Node require(esm) resolution share the same files, with no CommonJS or runtime-specific code copy.
• A release-coherence gate rejects version, dependency, runtime identity, migration Skill, Node range, distribution, npm-channel, or rxjs.dev-exclusion drift.

P6.1 completion evidence

• Accepted D-053 with Node 22.13+/24 blocking, Node 26 advisory, current Chrome/Firefox/desktop Safari/Mobile Safari, Webpack 5, Deno, and Bun; RxJS 9 beta uses npm next, RxJS 7 remains latest and maintained.
• Synchronized all four manifests, exact internal dependencies, runtime identities, import fixtures, lockfile entries, and the checked-in canonical migration Skill adapter to 9.0.0-beta.0 while preserving the immutable 8.0.0-alpha.14 qualification records.
• Configured all four packages to build only dist/esm; removed legacy main/CommonJS and target-specific outputs. Browser, Webpack, import, and Node require(esm) conditions now resolve the same ESM files. The mixed-load fixture proves shared Symbol and fallback identity.
• Added tested distribution finalization and release-coherence gates. All four packages build, all declaration consumers pass, ESM and Node require(esm) imports pass on Node 24.12.0, and no apps/rxjs.dev path changed or ran.
• Marked P6.1 DONE and advanced P6.2 as the sole NEXT item.

P6.2 completion bar

• Blocking Node 22.13+/24, current Chrome/Firefox/desktop Safari/Mobile Safari, Deno, Bun, and Webpack lanes plus advisory Node 26 are executable and guarded against release-policy drift.
• All four packages pass build, type, import/require(esm), package, and publication checks from one ESM output; alternate runtimes add no shipped code or bundle bytes.
• Browser lifecycle/Symbol behavior, branded Safari, Webpack resolution and size, performance floors, and pinned Observable WPT conformance have explicit blocking gates and checked-in budgets or revisions.
• Evidence distinguishes current engine integration, branded Safari, pinned standards conformance, focused product behavior, and intentional RxJS 7 migration divergences without changing their classifications.

P6.2 completion evidence

• Added guarded release workflows for blocking Node 22.13.0/24, advisory Node 26, Chrome 151, Firefox 153, WebKit 26.5, Webpack 5.106.2, Deno 2.8.0, Bun 1.3.14, desktop Safari, and Mobile Safari in an actual iOS simulator. The release-coherence test rejects removal of any lane or any rxjs.dev reference.
• Passed the package-built runtime and ESM resolution contract locally on Node 22.13.0, 24.12.0, advisory 26.5.0, Deno 2.8.0, and Bun 1.3.14. Node 22 and 26 also pass ESM and require(esm) package imports; no runtime-specific source, dependency, condition, or artifact was added.
• Passed the same eight-case browser contract in Chromium 151 with native Observable and Firefox 153/WebKit 26.5 with the fallback. Added separate SafariDriver gates for branded desktop Safari and platformName: iOS plus safari:useSimulator: true; local Safari was not enabled persistently merely to manufacture evidence.
• Webpack consumed 19 dist/esm modules and emitted 17,502 minified bytes under a 22,000-byte ceiling. Node 24 medians reached approximately 50.4 million map values/second and 140,500 cancellations/second over floors of 5 million and 30,000 respectively.
• Passed all four package builds, declaration consumers, imports, package fixtures and dry runs; focused suites pass 51 polyfill, 750 RxJS, 75 test, and 166 migration tests. The pinned Chrome 150 gate passes all 52 URLs, 525 upstream subtests, and 52 exact-identity attestations.
• Recorded the executable matrix and budget rationale under packages/rxjs/docs/RELEASE_GATES.md, marked P6.2 DONE, and advanced P6.3 as the sole NEXT item without changing or running apps/rxjs.dev.

P6.3 completion bar

• The root README clearly identifies the beta, explains why the release is RxJS 9 rather than RxJS 8, distinguishes RxJS 7 latest from RxJS 9 next, and puts the platform/Symbol/lifecycle model and first install example up front.
• Every release package has package-owned entry documentation; RxJS API, migration, release-gate, and contributor guidance and migration-engine product/qualification guidance stay inside their owning package containers.
• Package manifests publish the applicable README and documentation paths, all local documentation links resolve within the owning container, and publication dry runs contain the intended documents.
• A blocking freshness check rejects missing publication paths, broken or cross-container links, and documentation-site coupling without reading, building, testing, publishing, or deploying the site.

P6.3 completion evidence

• Replaced the repository README with an RxJS 9 beta entry point that answers “What happened to RxJS 8?”, explains native-first exact-Symbol composition, shows an install/use example, names the four package products, and states the release/support channels and current environment boundary.
• Added package READMEs for rxjs, @rxjs/observable-polyfill, and @rxjs/test; refreshed @rxjs/migrate; added package-local RxJS API and contributor guides; and retained the migration, unsupported-surface, evidence-ledger, and release-gate references beside rxjs.
• Moved the 650-line migration tooling design into packages/migrate/docs beside its qualification record and updated controlling repository links. Package metadata now points npm users at the package READMEs rather than the separate site workstream.
• Added tested package-documentation validation and made it part of release:check. All local links and package boundaries pass, and four npm publication dry runs include the expected README/docs/Skill content from one ESM artifact set.
• Marked P6.3 DONE and advanced P6.4 as the sole NEXT item. No apps/rxjs.dev path was changed, built, tested, published, or deployed.

P6.4 completion bar

• Exact 9.0.0-beta.0 tarballs install into an isolated representative consumer and pass ESM, Node require(esm), public types, test-harness use, and browser bundling on blocking Node 22.13/24 and advisory Node 26.
• Packed artifacts contain only intended ESM output, package-local documentation, metadata, and the canonical migration Skill where applicable; source specs, duplicate dialects, and size-budget regressions are rejected.
• Representative application/library migration outcomes, package, runtime, browser, conformance, performance gates, and the release-version dry run leave no unexplained local beta blocker.
• A durable decision approves the synchronized RxJS 9 beta.0 train for npm next after blocking CI is green while explicitly withholding npm publication, stable 9.0.0, and latest promotion.

P6.4 completion evidence

• Added a tested packed-artifact auditor and offline adoption gate. The exact tarballs pass ESM, require(esm), strict TypeScript, @rxjs/test, and a 20,222-byte browser bundle on Node 22.13.0, 24.12.0, and advisory 26.5.0.
• Audited 183,920-byte rxjs, 23,375-byte polyfill, 47,270-byte test, and 112,638-byte migration tarballs under checked-in 250K/30K/60K/140K ceilings. No source spec, duplicate dialect, or required package-document omission was present.
• The Nx first-release dry run resolves all four packages from the prior alpha tag to 9.0.0-beta.0. Release coherence protects the next/latest channels, supported matrix, documentation boundary, and version identities.
• Reused the closed P5.2 adoption matrix: three representative RxJS 7 applications/libraries complete and one unsupported case stops safely across Vitest, Mocha, Jest, cold, platform-shared, mixed, Subject, and unsupported contracts.
• Accepted D-054 and recorded package-local approval for the RxJS 9 beta line under npm next after blocking CI is green. Stable 9.0.0, latest promotion, npm publication, and website work remain outside this approval.
• Marked P6.4 DONE and advanced the terminal P6.5 closure audit as the sole NEXT item.

P6.5 completion bar

• A frozen-lockfile installation and all four release-package lint, build, declaration, import, package, and publication gates pass from the resulting workspace state.
• Release coherence, package-documentation, packed-consumer adoption, Webpack size, and performance budgets pass with the approved beta artifacts.
• Workflow syntax, diff hygiene, and the active-plan state pass a final audit; the release workflow neither references nor operates on the documentation application.
• apps/rxjs.dev has no worktree change and was not built, tested, published, or deployed during this project-plan execution.

P6.5 completion evidence

• Completed a frozen-lockfile install on Node 24.12.0 and then verified the lockfile offline. A legacy optional re2 native build reported a non-blocking Node 24 compilation warning outside the four release packages; pnpm completed successfully and no release-package dependency relies on it.
• Passed lint with zero errors for all four release packages, then passed every package build, declaration consumer, ESM and Node require(esm) import, package fixture, and publication check. The audit fixed one previously masked test-only manifest dependency and scoped Nx's repository test-graph cycle out of the @rxjs/test package lint boundary.
• Re-passed all ten release/documentation unit and coherence checks, the exact packed-tarball consumer adoption gate, the 17,502-byte Webpack gate, and the performance floors at approximately 47.1 million map values/second and 131,300 cancellations/second.
• Parsed every workflow, passed whitespace/diff checks, and confirmed an empty apps/rxjs.dev diff and status. No documentation-site command was run.
• Marked P6.5 DONE. All planned implementation work is complete, P5.3 and P5.4 remain explicitly accepted DEFERRED ideas rather than unfinished execution items, and the terminal queue intentionally has no NEXT marker.

P6.6 completion bar

• One internal positional subscribeToSource helper owns default forwarding, synchronous callback/setup error forwarding, and optional local-signal joining for every semantically eligible production source subscription.
• The obsolete helper alias, legacy options form, separate synchronous-error wrapper, and their dead types are removed.
• Remaining raw production subscriptions are limited to reviewed root-core, async-generator, and intentional host-reporting/terminal-order boundaries.
• Focused behavior, cold/fallback evidence, type, package, kernel, Webpack, and performance gates retain their accepted outcomes.
• Identical pre/post esbuild scenarios record minified, gzip, and Brotli changes for the root control, map, a representative operator set, and the complete runtime catalog; representative and catalog bundles shrink without growing the root control or map bundle.

P6.6 completion evidence

• Replaced the transitional helper pair with one positional subscribeToSource, migrated every semantically compatible operator/source module, and removed the legacy options form, alias, synchronous-error wrapper, and dead result/option types. Nine helper cases cover binding, default forwarding, callback/setup failures, destination cancellation, and joined local signals.
• Audited all production .subscribe(...) calls. The 24 remaining raw calls are confined to root-core direct subscriptions, Subject-like connection and retained-group lifecycles, async-generator adapters without a destination Subscriber, the helper itself, intentional reset/finalization host-error paths, and subscriptions whose source/boundary activation must outlive or follow the outer result's signal.
• Passed 753/753 focused source tests, lint with zero errors, strict public types, the complete build/package/migration/install/import gate, and the eight-case packaged-fallback/native-Chrome kernel contract. The reviewed audits remain unchanged at 2,299/2,338 cold and 2,316/2,338 fallback, with the same classified 39 and 22 failures.
• Passed Webpack 5.106.2 at 17,410 bytes and the performance gate at about 42.16 million map values/second and 117,131 cancellations/second.
• Repeated the baseline's esbuild 0.19.11 browser-ESM es2022 measurement with tree shaking, minification, gzip level 9, and Brotli quality 11. The baseline is HEAD 458e8d2c3 plus the initial helper/map diff identified by SHA-256 621c2f87648337d34b3e81e2ea4b53a2f326c180a387cabb39632bd137425c31.

Scenario	Minified before → after	Gzip before → after	Brotli before → after
Root-only control	22,494 → 22,494 (0 B, 0.00%)	6,376 → 6,376 (0 B, 0.00%)	5,710 → 5,710 (0 B, 0.00%)
map subpath	14,413 → 14,375 (-38 B, -0.26%)	4,219 → 4,218 (-1 B, -0.02%)	3,789 → 3,789 (0 B, 0.00%)
Representative 10-operator set	21,133 → 19,348 (-1,785 B, -8.45%)	6,235 → 5,751 (-484 B, -7.76%)	5,564 → 5,175 (-389 B, -6.99%)
Complete runtime catalog	77,064 → 64,351 (-12,713 B, -16.50%)	20,789 → 17,599 (-3,190 B, -15.34%)	17,963 → 15,570 (-2,393 B, -13.32%)

• Marked P6.6 DONE; the completed execution queue again has no NEXT marker.

P6.7 completion bar

• Every derived Observable in scope is constructed directly through the receiver's D-037 [create] protocol, with no single-line wrapper helper.
• Every production use of the legacy derived-construction helper moves to a direct [create] call; both wrapper helpers, the obsolete options type, and dead helper tests are removed.
• The user's draft map change remains a separate commit, followed by one commit per migrated operator/factory and separate helper cleanup, architecture/decision, and verification-evidence commits.
• Focused behavior, type, package, kernel, classified cold/fallback, Webpack, and performance gates retain their accepted outcomes.
• Identical pre/post esbuild scenarios record minified, gzip, and Brotli changes for the root control, map, representative operator set, and full runtime catalog without growing any scenario.

P6.7 completion evidence

• Moved map, scan, switchMap, timeout, and timer to direct receiver [create] construction. Removed operate, createDerivedObservable, the obsolete options type, and the two tests that existed only for the deleted wrappers; no production or test references remain.
• Kept the user's draft map migration separate, then committed each remaining operator or factory, helper cleanup, architecture/decision wording, and this verification evidence as independent small changes.
• Passed all 751 RxJS source tests, lint with zero errors, public types, the complete package/import gate, and the eight-case packaged-fallback/native Chrome kernel contract. The classified audits retained exactly 2,299 of 2,338 cold passes and 2,316 of 2,338 fallback passes.
• Passed Webpack 5.106.2 at 17,369 bytes and the performance gate at about 44.15 million map values/second and 125,014 cancellations/second.
• Repeated the baseline's esbuild 0.19.11 browser-ESM es2022 measurement with tree shaking, minification, gzip level 9, and Brotli quality 11. The baseline is HEAD 94b95651e plus the initial helper/map diff identified by SHA-256 76a8817c1383a6ef2a2bbe21ee17918814a6bf2d6aaa13d78c1dc578db5a8c1f.

Scenario	Minified before → after	Gzip before → after	Brotli before → after
Root-only control	22,494 → 22,494 (0 B, 0.00%)	6,376 → 6,376 (0 B, 0.00%)	5,710 → 5,710 (0 B, 0.00%)
map subpath	14,347 → 14,315 (-32 B, -0.22%)	4,202 → 4,194 (-8 B, -0.19%)	3,778 → 3,771 (-7 B, -0.19%)
Representative 10-operator set	19,365 → 19,243 (-122 B, -0.63%)	5,762 → 5,703 (-59 B, -1.02%)	5,180 → 5,134 (-46 B, -0.89%)
Complete runtime catalog	64,368 → 64,231 (-137 B, -0.21%)	17,606 → 17,542 (-64 B, -0.36%)	15,591 → 15,508 (-83 B, -0.53%)

• Marked P6.7 DONE; the completed execution queue has no NEXT marker.

P6.8 completion bar

• Every accepted source, package, type, import, migration-evidence, workflow, runtime, browser, bundle, performance, adoption, and WPT gate is owned by a pull-request or master workflow.
• The complete 2,338-case cold and polyfill audits run in CI and require exact equality with the reviewed 2,299/39 and 2,316/22 pass/failure baselines; incomplete execution and both unexpected failures and passes are blocking.
• Blocking Node 22.13/24, TypeScript-latest, WPT, release-readiness, Deno, Bun, Webpack, desktop Safari, and Mobile Safari jobs start on the validation PR; Node 26 and latest-Chrome drift remain explicitly advisory.
• Release coherence rejects removal of required commands, lanes, or master triggers, and package-local release documentation describes the durable CI policy without changing runtime APIs or apps/rxjs.dev.

P6.8 completion evidence

• Opened draft PR #7611 from benlesh:codex/ci-master-coverage into ReactiveX/rxjs:master. At code SHA 829170c439, main CI passed the exact Node 24 migration-evidence job plus the blocking Node 22.13 and maintained Node 24 package gates; advisory Node 26 also passed.
• The CI verifier executed all 2,338 registrations in each mode and retained exactly 2,299/39 cold and 2,316/22 polyfill outcomes. Its negative tests reject incomplete reports, unknown or duplicate IDs, new failures, and unexpected passes. Clean runners build the polyfill package boundary before collecting the corpus.
• TypeScript-latest and pinned Chrome 150 WPT passed. The scheduled latest-Chrome lane remained advisory and was skipped on the pull request, as designed.
• Release readiness passed Chrome, Firefox, WebKit, Webpack, performance, packed adoption, Deno 2.8.0, Bun 1.3.14, desktop Safari, and Mobile Safari in an explicitly booted iOS simulator.
• Configuration failures found during PR validation covered clean-workspace build order, hosted WPT prerequisites, shallow migration-source checkout, current TypeScript compilation, adoption prerequisites, and Safari simulator selection. Each was corrected and guarded. No product-test failure was found.
• Marked P6.8 DONE; the completed execution queue has no NEXT marker. No runtime API, export, package format, type contract, publication, or apps/rxjs.dev change was made.

Dependencies

flowchart LR
    MigratePrototype["P0.M1 migration prototype"] --> MigrationDesign["P0.M2 agent-first product design"]
    MigrationDesign --> Mechanical["P0.M3 mechanical engine hardening"]
    Mechanical --> PortableSkill["P0.M4 portable Skill and harness guidance"]
    PortableSkill --> AgentQualification["P0.M5 Codex/ChatGPT outcome qualification"]
    WptHarness["P1.4a attested WPT harness"] --> Boundary["P0.2 package and install decisions"]
    Boundary --> Build["P0.3 buildable packages"]
    Build --> Harness["P0.4 lifecycle and import harness"]
    Harness --> Fallback["Phase 1 fallback correctness"]
    Fallback --> WptStrict["P1.4b strict WPT conformance (done)"]
    Boundary --> Kernel["Phase 2 extension kernel"]
    Harness --> Kernel
    Kernel --> Operators["Phase 3 operator restoration"]
    Operators --> DirectInstall["P4.I1 direct Symbol installation"]
    Boundary --> Evidence["Phase 4 intentional API and migration contracts"]
    Operators --> Evidence
    Evidence --> Migration["Phase 5 migration and AI enablement"]
    Operators --> Migration
    AgentQualification --> Migration
    Fallback --> Release["Phase 6 release"]
    DirectInstall --> Release
    Migration --> Release

The first standards revision can be pinned while build work proceeds, but conformance implementation depends on a runnable harness.

Risk register

Risk	Impact	Likelihood	Current response
Native/fallback lifecycle evidence regresses	Package acquisition can pass while semantics diverge	Medium	P0.4's shared contract and the attested strict WPT gate block fallback drift; RxJS extension evidence remains separately classified
Upstream proposal changes	Fallback and native behavior drift	High	Pin revisions before conformance claims
Prototype code becomes accidental policy	Semantics are preserved without review	High	Documents distinguish current fact from accepted direction
Symbol identity fails across package copies	Extensions are present under inaccessible keys	High	D-048 exact-key isolation plus mixed-dialect and duplicate-capability package fixtures; no runtime conflict arbitration is required
RxJS 7 suite pressures platform behavior backward	Native and fallback layers diverge	High	Mandatory classification; evidence never implies a runtime compatibility product
Migration evidence is mistaken for emulation	Users depend on unsupported RxJS 7 surfaces	Medium	Publish explicit source actions, semantic-review flags, and unsupported categories
Global patching fails in hardened realms	Library cannot initialize	Medium	Leave hardened surfaces unclaimed; D-051 accepts native assignment errors and no paired-install rollback
Tooling is designed before APIs stabilize	Skills encode obsolete migrations	Medium	D-046 requires versioned capability evidence and portable workflow invariants; P0.M3 expands only for accepted contracts
Mechanical output is mistaken for a complete migration	Users silently receive the wrong Observable lifecycle	High	D-046 makes the agent workflow primary and requires explicit contract classification before codemods
Agent output is judged only by source shape	Nondeterministic rewrites can hide behavioral regressions	High	D-046/P0.M5 require RxJS 7 baselines, contract manifests, compilation, and behavioral outcome gates
Harness-specific instructions drift	Codex, Claude, and Cursor provide inconsistent safety	Medium	D-046 selects one canonical portable Skill; P0.M4 tests only thin harness installation adapters
Current CI/release infrastructure assumes RxJS 7	Published artifacts fail despite source tests	High	Package-import fixtures and release gates precede expansion
WPT runs accidentally exercise native Observable	False confidence in fallback behavior	High	P1.4a exact-identity attestation and an independent, unsuppressible report audit
WPT/browser setup is too large or network-dependent	Slow or skipped local and CI validation	Medium	Vendor only the approved closure and checksum-cache the sparse runner and exact browser artifacts

Out of scope until activated

• every change, build, test, publication, or deployment of apps/rxjs.dev;
• a complete operator priority list;
• release dates.

Session log

2026-07-24 — Documentation foundation

• Analyzed branch history, packages, source modules, manifests, tests, and the current Observable specification/WPT locations.
• Created the project charter, architecture, compatibility policy, decision log, open-question register, active plan, and repository AI guidance.
• Verified the narrow source-test and build baseline recorded under P0.1.
• Set P0.2, package and installation contract decisions, as the only NEXT step.

2026-07-24 — Symbol patching rationale

• Documented why exact Symbol keys make import-time prototype patching safe from accidental name collisions.
• Recorded the contrast with RxJS 5 rxjs/add/operator/* string-key patching.
• Clarified that Symbol.for trades away some collision isolation and requires an explicit namespacing decision.

2026-07-24 — Dual platform and RxJS operator surface

• Accepted that RxJS exports Symbols for operators such as map and filter even when the platform provides same-familiar-name string methods.
• Recorded observable.map(project) as the platform contract and observable[map](project) as the RxJS contract.
• Allowed the RxJS form to delegate or provide additional functionality while requiring the platform method to remain untouched and all differences to be documented and tested.

2026-07-24 — User-prioritized attested Observable WPT harness

• Recorded the explicit priority change from P0.2 to the test-infrastructure slice P1.4a while preserving one NEXT item.
• Split harness construction and its current-failure baseline from P1.4b, which will later make the fallback pass the strict conformance gate.
• Accepted WPT commit 6a009d73f0d315941b90cac13a9523a2a08c631b, per-realm exact implementation attestation, unsuppressible report auditing, and cached pinned browser/runner operation.
• Completed P1.4a with the exact 37-file Observable/support closure, 52-URL attested inventory, stable granular failure baseline, negative controls, CI evidence, and warm offline proof.
• Restored P0.2 as the single NEXT item. Strict WPT conformance remains the separate planned P1.4b effort.

2026-07-24 — Node 24 WPT tooling support

• Expanded the repository development engine declaration to accept Node 24 while retaining Node 18 and Node 20.
• Moved the blocking and advisory Observable WPT workflows to Node 24.
• Verified the WPT import, unit, doctor, and browser-baseline paths on Node 24.12.0 without an engine-check override.
• Kept P0.2 as the single NEXT item; final published-package runtime support remains an open release decision.

2026-07-24 — Strict default WPT command and terminal diagnostics

• Made the strict test:wpt command and the blocking CI job require all upstream WPT results to pass; the current implementation therefore exits nonzero.
• Retained the known-failure comparison only as the explicitly named test:wpt:baseline harness diagnostic.
• Added progress messages, aggregate status and identity counts, every non-passing URL and subtest, and direct artifact paths to terminal output.
• Kept P0.2 as the single NEXT item; implementation conformance remains the separate planned P1.4b effort.

2026-07-24 — Pinned Observable WPT conformance

• Temporarily moved NEXT from P0.2 to P1.4b at the user's direction, fixed only the Observable and EventTarget.prototype.when fallback, completed P1.4b, and restored P0.2 as the single NEXT item.
• Audited the harness before changing behavior: all 52 disposable realms used the exact bundled fallback constructor, subscribe, and when identities; no native implementation or later overwrite explained the failures.
• Corrected Subscriber lifecycle, abort/teardown ordering, error reporting, detached realms, Web IDL shape, platform input conversion, iterator protocols, Promise consumers, and the small set of platform operators exposed by the pinned suite.
• Kept the async-iterator conversion as an explicit pull loop and documented why for await...of cannot express the required acquisition, close-reason, timing, and pending-result semantics.
• Corrected generated attestation timing after discovering that an early completed identity subtest caused upstream setup() options to be ignored. No vendored WPT source was edited.
• Passed the pinned strict suite with 52/52 URLs, 525/525 upstream subtests, and 52/52 exact-identity attestations, then recorded three additional identical runs and removed every obsolete failure expectation.
• Passed all 49 package tests and pinned-import verification. Reconfirmed, without expanding scope, the documented P0.3 package build/lint configuration failure.

2026-07-24 — Framework-neutral RxJS testing package

• User-prioritized and completed P0.T1 without changing the active Observable acquisition contract still queued at P0.2.
• Added the @rxjs/test package with rxTest, async virtual-time controls, marble assertions, and separate cold, hot, and platform observable source models.
• Virtualized timers, animation and idle callbacks, Node immediates and timer handles, AbortSignal.timeout, clocks, queued microtasks, and reportError; every patch is restored after success or failure.
• Added the accepted testing design and D-012, plus architecture, compatibility, and open-question updates.
• Passed 66 focused tests, lint, TypeScript checks, Nx build/test targets, the multi-dialect package build, strict declaration coexistence with the Observable polyfill, and built ESM/CommonJS imports. P0.2 remains the single NEXT item.

2026-07-24 — Standalone marble parser follow-up

• Renamed the internal parser boundary to marble-parser.ts and kept it free of virtual-clock, host, Observable, and assertion dependencies.
• Added 33 direct unit cases for alignment whitespace, duration conversion, message diagrams, synchronous groups, hot carets, subscription diagrams, time diagrams, animation/idle timing plans, Unicode markers, and invalid grammar.
• Locked the alignment example to the source column of ^ and documented the difference between source-only spacing and ignored in-string padding.
• Kept the parser internal so this implementation improvement does not create an unreviewed public subpath; P0.2 remains the single NEXT item.

2026-07-25 — RxJS 7 marble-test migration Skill and classified port

• Temporarily moved NEXT from P0.2 through the user-ordered Skill creation, independent vetting, and repository port steps, advancing only after each completion bar passed.
• Added the portable rxjs-next-marble-migration Skill with framework-neutral conversion guidance, lifecycle classifications, examples, reporting, and dependency-free analysis and portability helpers.
• Passed helper fixtures, portability checks, structural validation, temporary packaging, and independent plugin evaluation at 100/100 with no blocking or fix-first findings.
• Read the production 7.x ref without checking it out, pinned e5351d02e225e275ac0e497c7b66eaa5f0c88791, and reconciled all 2,146 inventoried marble cases into one generated disposition ledger.
• Added isolated cold, polyfill, native-if-present, and raw-audit modes. The normal cold and polyfill gates pass; the current Node native mode skips explicitly; cold and polyfill audits expose their complete mode-specific implementation, capability, conversion, and lifecycle mismatches without repairing production behavior.
• Recorded D-013, testing and compatibility updates, and complete port notes. Restored P0.2 as the single NEXT item.

2026-07-25 — Complete executable parity registration and API map

• Reopened the user-prioritized marble port after clarifying that manifest-only preservation was insufficient for missing capabilities.
• Materialized all 2,146 source cases as cold test registrations and retained executable converted programs for every disposition.
• Corrected the import-role mapper so pipeable operators resolve to exact or explicitly unified instance Symbols while creation functions resolve independently to static Symbols or ambient-platform constructions.
• Added explicit missing-capability and unavailable-harness failures, plus Vitest argument pass-through for focused audits.
• Generated RxJS-7-parity.md from the pinned RxJS 7 public indexes and the shared machine-readable registry: 113 pipeable operators and 34 creation/utility functions with current mappings, case counts, partial analogues, and missing status.
• Passed all 2,146 normal cold and polyfill source registrations, with four exact duplicates skipped in each normal mode. The raw cold audit reports 1,811 failures and 335 passes; the polyfill audit reports 1,806 failures and 340 passes. Native mode skips honestly in the current Node realm.
• Added executable unified mappings, including bufferCount(size) → source[buffer]({ maxSize: size }), and retained unsupported overloads as failing parity evidence.
• Restored P0.2 as the single NEXT item without changing production source.

2026-07-25 — Exhaustive parameterized marble conversion

• Re-audited the pinned source with an AST-based declaration and helper expander. The earlier 2,146-case inventory omitted parameterized variants and source-skipped evidence; the authoritative inventory is now 2,338 registrations from 2,201 physical declarations.
• Expanded all loop- and helper-declared variants, including 117 share configurations, 48 multicasting deprecation equivalents, and four buffer cases. All four source-skipped tests remain executable parity evidence.
• Reduced unsupported/obsolete classification from 136 to 11 without guessing behavior. The remaining cases protect scheduler-private parser/queue state or depend on phonyMarbelize.
• Expanded executable API mappings, corrected overload guards, introduced unique case-ID baselines and strict sharded audit merging, and protected native constructor identity before extension loading.
• Passed both 2,338-registration normal gates and complete cold/polyfill raw audits. No production source was changed to satisfy an old expectation.
• Marked P0.T2e complete and restored P0.2 as the single NEXT item.

2026-07-25 — Ported-test console follow-up

• Replaced per-case success output from normal ported-test commands with a concise per-mode summary and an unmistakable final PASS or FAIL.
• Kept reviewed parity passes and quarantined known gaps separately visible so a green harness result does not imply full RxJS 7 behavioral parity.
• Preserved expanded diagnostics for failed shards and direct Vitest reporter output when explicit Vitest arguments are supplied.
• Kept P0.2 as the single NEXT item.

2026-07-25 — Strict ported-test execution and live progress

• Temporarily reprioritized the user-requested P0.T2f runner follow-up and restored P0.2 as the single NEXT item after completing it.
• Removed expected-failure quarantine from every applicable ported registration; recorded baselines remain evidence only.
• Regenerated harness diagnostics to remove stale quarantine wording without changing any disposition or converted program.
• Added an in-place shard-completion and heartbeat status so a slow or stalled shard remains visible without scrolling the terminal.
• Verified the default cold and polyfill run completed every shard, expanded the failures, and returned nonzero without changing production behavior.

2026-07-25 — In-place ported-test progress

• Replaced newline-per-update progress with one interactive status line that is cleared and redrawn on shard completion and heartbeat refresh.
• Bounded redirected and CI output to one final progress summary.
• Verified an interactive cold run completed all 16 shards in 87 seconds while rewriting one progress row, then expanded every failure and returned exit code 1 as intended.
• Kept P0.2 as the single NEXT item.

2026-07-29 — pnpm 10 repository migration

• Temporarily prioritized and completed P0.DX1, replacing Yarn Classic with pnpm 10.34.5 across workspace metadata, the lockfile, local scripts, CI, release preparation, docs tooling, generated commands, and contributor instructions.
• Preserved pnpm's isolated linker and recorded a narrow root-only polyfill public-hoist as a temporary workspace bridge that does not decide P0.2.
• Made dependency build scripts reviewable through version-bounded allowBuilds and strictDepBuilds; explicitly denied the message-only Core-JS and ES5-Ext scripts.
• Declared the Chai test dependency and the release helper's Yargs dependency that the former flat layout had hidden.
• Patched Husky 4's obsolete pnpx --no-install hook runner to use pnpm exec, then exercised the configured commit hooks successfully.
• Passed clean and repeated frozen installs, project discovery, focused package and RxJS tests, parity freshness, import verification, and the complete cached strict WPT suite. Exercised docs and release paths and retained their documented unrelated Node 24 and package-build failures.
• Restored P0.2 as the single NEXT item.

2026-07-29 — Root developer command guide

• Temporarily prioritized and completed P0.DX2.
• Organized the root README around setup, fast feedback, docs, WPT, and maintainer workflows, with concise why/when guidance.
• Highlighted common commands, linked the detailed workflow guides, and called out intentionally failing or blocked gates.
• Verified command and project names against the pnpm workspace and restored P0.2 as the single NEXT item.

2026-07-29 — RxJS 7 ported-test failure tracker

• At the user's direction, temporarily moved NEXT from P0.2 to P0.T3 and created RXJS_7_PORTED_FAILURES.md as the durable operator/function remediation queue.
• Ran complete 2,338-case cold and polyfill audits. Cold recorded 432 passes and 1,906 failures; polyfill recorded 436 passes and 1,902 failures. Their union contains 1,921 failing case IDs across 140 owner groups: 1,887 fail in both modes, 19 only in cold, and 15 only in polyfill.
• The first cold attempt hit a Vitest cache-directory race and was rejected because one shard report was absent. The clean rerun merged all 16 cold shards, and the polyfill audit merged all 16 shards on its first attempt.
• Added a reproducible tracker generator and stable case-ID status ledger. Regeneration preserves resolved rows as FIXED, retains active or blocked assignments, and rejects incomplete reports, unknown cases, invalid statuses, unnamed blockers, duplicate rows, or missing work packets.
• Ran the normal strict ported gate through all 16 shards in both modes. It returned exit code 1 as expected, with both modes failing and no omitted shard. The generator's check mode validates all 1,921 rows and 140 packets.
• P0.T3 remains the single NEXT item. The first recommended unblocked packet is RX7-NEVER: add an explicit observation boundary to the migrated test without changing NEVER runtime semantics.

2026-07-29 — P0.T3 existing-surface remediation wave 1

• Integrated separate reviewed fixes for the NEVER observation boundary, raceWith winner selection and cancellation, unsubscription-only observation marbles, overlapping and gapped bufferCount windows, seedless scan, and the elementAt default-value overload.
• The shared observation-boundary repair also completed the assigned defaultIfEmpty cases and repaired affected evidence across 37 owner groups. The complete audits now mark 89 original failures FIXED.
• Complete 16-shard audits recorded 523 cold passes with 1,815 failures and 525 polyfill passes with 1,813 failures. The normal strict gate then completed all 16 shards in both modes and remained red, as expected, on unresolved mapped behavior and intentionally out-of-scope missing APIs.
• Recorded D-015 for count-window configuration on the unified buffer Symbol. Missing compatibility APIs such as ArgumentOutOfRangeError and finalize remain TODO; they were not introduced to make neighboring cases pass.
• The next existing-functionality packet is RX7-IGNORE-ELEMENTS, beginning with its never-observation harness boundary before moving to other small mapped edge cases. Missing non-scheduler APIs remain a later phase, and all scheduler-specific behavior remains deferred until the absolute end of P0.T3.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 2

• Integrated separate reviewed fixes for the ignoreElements never-observation boundary, nonpositive skipLast identity behavior, and nonpositive takeLast empty behavior. The takeLast ambient declaration now exposes only the existing exported Symbol rather than a string-named method.
• Complete 16-shard audits recorded 529 cold passes with 1,809 failures and 531 polyfill passes with 1,807 failures, adding six cases that pass in both modes. The normal strict gate again completed every cold and polyfill shard and remained red on the unresolved queue.
• The synthetic ignoreElements stop closes both its observation and expected source-subscription log. It does not alter production semantics or make the shared, ref-counted polyfill cold.
• No missing API or scheduler behavior was introduced. The next wave continues with small current mappings and their overload/error gaps; scheduler-specific cases remain deferred until the absolute end of P0.T3.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 3

• Integrated separate reviewed fixes for two switchAll never-observation boundaries, bufferWhen source-error and synchronous-selector behavior, and the shared @rxjs/test same-frame unsubscription contract.
• expectObservable windows are now half-open: observation aborts run before ordinary work on the exact ! frame. This repaired the two assigned skipWhile cases without changing production and also fixed two equivalent takeWhile cases.
• Complete 16-shard audits recorded 537 cold passes with 1,801 failures and 539 polyfill passes with 1,799 failures, adding eight cases that pass in both modes. The normal strict gate completed all shards in both modes and remained red on the unresolved queue.
• All 16 bufferWhen cases now pass in both modes. Source errors discard the active partial buffer, and a synchronously throwing closing selector errors the result before source activation.
• No missing API or scheduler behavior was introduced. Scheduler-specific cases remain deferred until the absolute end of P0.T3, and P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 4

• Integrated separate reviewed fixes for three mergeAll never-observation boundaries, the RxJS 7 race([sources]) operator overload, and takeWhile completion, source-error, and inclusive behavior.
• The takeWhile production repair forwards the source terminal lifecycle. Its RxJS 7 boolean inclusive overload is normalized only at the compatibility boundary to RxJS Next's { includeLast: true } configuration.
• Complete 16-shard audits recorded 548 cold passes with 1,790 failures and 550 polyfill passes with 1,788 failures, adding 11 cases that pass in both modes and bringing the original fixed cohort to 114.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. The shared, ref-counted polyfill was not made cold, and mode-only lifecycle differences remain visible in the tracker.
• No missing API or scheduler behavior was introduced. The next recommended existing-functionality packet is RX7-SKIP-LAST; scheduler-specific cases remain deferred until the absolute end of P0.T3, and P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 5

• Integrated separate reviewed fixes for the skipLast and takeLast never-observation boundaries, the static RxJS 7 race([sources]) overload, and sequenceEqual length-mismatch completion.
• sequenceEqual now decides false as soon as a completed shorter input makes equality impossible and completes the derived subscriber so its AbortSignal closes both inputs. The RxJS 7 comparator overload and never-observation evidence remain separate unresolved work.
• Complete 16-shard audits recorded 557 cold passes with 1,781 failures and 559 polyfill passes with 1,779 failures, adding nine cases that pass in both modes and bringing the original fixed cohort to 123.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. The shared, ref-counted polyfill was not made cold, and its mode-only lifecycle differences remain explicit evidence.
• No missing API or scheduler behavior was introduced. The next recommended existing-functionality packet is RX7-SWITCH-MAP-TO; scheduler-specific cases remain deferred until the absolute end of P0.T3, and P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 6

• Integrated separate reviewed fixes for three switchMapTo never-observation boundaries, repeat count semantics, and the RxJS 7 retry reset-on-success default.
• repeat now treats count as the exact total source-activation count and completes without source activation for nonpositive counts. Its focused shared-polyfill coverage verifies that concurrent observers still share each repeated upstream activation.
• RxJS Next retains its intentional retry default of resetOnSuccess: true; the RxJS 7 compatibility adapter supplies the legacy false default without changing production semantics.
• Complete 16-shard audits recorded 575 cold passes with 1,763 failures and 577 polyfill passes with 1,761 failures, adding 18 cases that pass in both modes and bringing the original fixed cohort to 141. Three count-root spillovers in delay-config cases passed without implementing scheduler behavior.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. No missing API or scheduler behavior was introduced. RX7-TO-ARRAY is the next recommended packet; its polyfill-only multiple-subscription evidence must preserve the shared, ref-counted lifecycle rather than making the polyfill cold.
• Scheduler-specific work remains deferred until the absolute end of P0.T3, and P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 7

• Integrated separate reviewed fixes for toArray lifecycle evidence, onErrorResumeNext sequencing and never boundaries, and same-frame hot observation ordering required by mapped concat/merge flattening cases.
• Mode-aware migrated evidence now retains RxJS 7's two cold upstream subscriptions while expecting one shared upstream subscription from the ref-counted platform polyfill. The rewrite is restricted to intentional subscription multiplicity and cannot alter values, errors, completion, or cancellation evidence.
• onErrorResumeNext now sequences the instance receiver before configured sources, advances on both completion and error, swallows exhausted errors, preserves constructor behavior, and cancels through the derived subscriber's AbortSignal.
• @rxjs/test now orders same-frame work as observation boundary, observation start, then ordinary source work. This observes hot frame-zero values without weakening the half-open ! contract.
• Complete 16-shard audits recorded 606 cold passes with 1,732 failures and 608 polyfill passes with 1,730 failures, adding 31 passes in each mode and bringing the original fixed cohort to 172. The shared fixes also repaired related concat/merge flattening, throttle-error, and TestScheduler evidence.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. No missing API or RxJS scheduler/provider behavior was introduced. Although RX7-DEBOUNCE-TIME is the generated recommendation, time/scheduler-related packets remain deferred until the absolute end; the next wave continues with non-time-based mapped behavior.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 8

• Integrated separate reviewed commits for 24 finite observation boundaries, exhaustMap active-inner lifecycle, and source-led withLatestFrom semantics.
• The boundary commit closed only observations and source subscriptions still active at the original diagram horizon. It completed the mapped concat, race, and concatMapTo packets without changing production behavior.
• exhaustMap now counts an accepted inner before activation, ignores later outer values until that inner completes, delays result completion while an inner remains active, and propagates the result subscriber's AbortSignal to inner work. The root fix completed exhaustAll and repaired nine finite exhaustMap cases as spillover.
• withLatestFrom now activates latest-only inputs before the primary source, emits only on primary values after every latest-only input has a value, and terminates with the primary source. The optional projection is represented by the Symbol contract and executable port adapter.
• Focused lifecycle evidence confirms that concurrent polyfill observers share one ref-counted producer activation for both exhaustMap and withLatestFrom; no cold-per-subscription behavior was introduced into the platform layer.
• Complete 16-shard audits recorded 662 cold passes with 1,676 failures and 663 polyfill passes with 1,675 failures, adding 56 and 55 passes respectively, with no regressions and 228 cases fixed from the original cohort. The one-pass mode delta is historical: the withLatestFrom throw case already passed in polyfill mode before this wave.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. No missing API or scheduler/provider behavior was introduced. The next wave continues non-time mapped behavior and finite observation evidence; scheduler-specific work remains deferred until the absolute end.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 9

• Integrated separate reviewed commits for 22 remaining non-time composition observation boundaries, the instance concat receiver lifecycle, and the sequenceEqual comparator overload.
• The harness-only commit added finite horizons for mapped merge, mergeWith, concatAll, concatMap, exhaustMap, and switchMap cases. It closed only source and inner subscriptions still active at the original marble horizon.
• Instance concat had prepended its receiver before delegating to instance merge, which prepended the same receiver again. Removing that duplicate activation completed the non-scheduler endWith cohort and repaired additional concatWith and legacy concat evidence. Focused lifecycle tests prove serial ordering, no append after error, one receiver activation, and ref-counted cancellation after the last polyfill observer leaves.
• sequenceEqual now accepts the existing RxJS 7 comparator overload, calls it once per paired value, concludes on false, propagates comparator errors, and cancels both inputs through the result AbortSignal. Concurrent observers share one comparison run.
• Complete 16-shard audits recorded 721 cold passes with 1,617 failures and 722 polyfill passes with 1,616 failures, adding 59 passes in each mode with no regressions and bringing the original fixed cohort to 287.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. The excluded endWith scheduler cases remain untouched. No missing API or scheduler/provider behavior was introduced; the next wave continues with current subjects and other non-time mapped behavior.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 10

• Integrated separate reviewed commits for the RxJS 7 buffer notifier lifecycle, class-local Subject.asObservable() views, and the remaining mapped mergeMap/mergeMapTo evidence.
• The buffer compatibility adapter retains one closing-notifier subscription across boundaries, emits empty buffers where RxJS 7 does, discards partial buffers on errors, and closes both source and notifier through the shared platform lifecycle.
• Subject.asObservable() returns a distinct non-mutating base-Observable view without adding a string-named method to Observable.prototype. Migrated hot fixtures receive the equivalent method only as a test-owned property.
• Reused RxJS 7 cold inners remain shared, ref-counted producers in polyfill mode. Mode-aware merge-mapping evidence records the resulting joined emissions, subscription intervals, and restarts after zero-subscriber gaps; no cold-per-subscription fallback was introduced.
• Complete 16-shard audits recorded 737 cold passes with 1,601 failures and 754 polyfill passes with 1,584 failures. That adds 16 cold and 32 polyfill passes with no regressions and brings the original fixed cohort to 319 cases.
• The normal strict gate completed all 16 shards in both modes and remained red on the unresolved queue. No missing API or scheduler/provider behavior was introduced. Time- and scheduler-related packets remain deferred until the absolute end of P0.T3.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 11

• Integrated separate reviewed commits for finite concatWith, legacy concat, and retry observation boundaries; selector and numeric debounce lifecycles; retry delay-notifier ownership; and the unified audit/throttle implementation.
• A proposed rewrite of legacy concat(e2, testScheduler) was rejected because removing the scheduler argument would weaken the original behavioral claim. That case and the explicit concat/endWith scheduler overloads remain untouched in the scheduler-last queue.
• Selector debounce now retains values across empty selector completion and flushes on source completion. Numeric debounce resets one host timer, flushes on source completion, and cancels that timer through the result lifecycle.
• Retry delay selectors now receive one-based counts even with an infinite budget. A notifier's first value cancels it before the next source attempt; notifier completion, error, and selector failure terminate through their documented paths.
• audit and throttle share one Symbol implementation while retaining their distinct trailing-window restart behavior. Focused tests preserve one ref-counted source and duration activation for concurrent polyfill observers; no mode-specific value rewrite or cold fallback was introduced.
• Complete 16-shard audits recorded 820 cold passes with 1,518 failures and 837 polyfill passes with 1,501 failures. That adds 83 passes in each mode with no regressions, including 14 auditTime/throttleTime spillovers, and brings the original fixed cohort to 402 cases.
• The normal strict gate completed both modes and remained red on the unresolved queue. No missing API, scheduler argument, scheduler injection, provider, or scheduler-class behavior was introduced. Remaining scheduler-specific concat and endWith cases stay deferred.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — Ported marble-test performance correction

• Paused further P0.T3 packet delegation to profile why the virtual-time marble suite was taking real minutes. No real-timer leak was found: numeric-timer cases execute through virtual scheduling and its bounded task guard.
• One failed instanceOf(Subject) assertion took about 44.3 seconds because Chai/Loupe treated the platform Observable inspect operator as a custom formatter, invoked it recursively, and eventually overflowed the stack. Ported Chai assertions now install a temporary, non-enumerable display hook only during synchronous formatting and remove it in finally. The same assertion now fails with its genuine diagnostic in about 5 milliseconds, while the Observable inspect operator remains unchanged and usable.
• The runner now defaults to one process per mode instead of repeating Vitest transformation and collection across 16 processes. Explicit shard-count and concurrency overrides remain available for diagnostic isolation.
• On the current 2,338-case manifest, complete JSON audits take about 2.7 seconds cold and 2.5 seconds polyfill. The full strict cold-plus-polyfill command, including approximately 5.7 MB of expected failure diagnostics, completes in about 6 seconds.
• One-process and explicit 16-shard audits produced identical case titles and statuses: 820 cold passes with 1,518 failures and 837 polyfill passes with 1,501 failures. The speedup does not quarantine failures or change the polyfill's shared, ref-counted behavior.
• P0.T3 remains the single project-level NEXT item, but further parity work remains paused pending user direction.

2026-07-29 — P0.T3 existing-surface remediation wave 12

• Resumed the active P0.T3 goal after the performance correction and integrated the remaining mapped numeric auditTime lifecycle plus finite auditTime/throttleTime never-source observation boundaries.
• auditTime now uses the unified throttle implementation without restarting a numeric duration from a trailing emission. Focused fake-timer evidence proves that concurrent polyfill observers retain one shared source activation and timer until the final observer aborts.
• Rejected proposed rewrites for the remaining concatMap, exhaustMap, and switchMap map-and-flatten cases. Their projected inners require the missing RxJS Symbol map; substituting the platform string map would violate the API boundary and hide a missing capability. All three remain TODO.
• Complete one-process audits recorded 826 cold passes with 1,512 failures and 843 polyfill passes with 1,495 failures. That adds six passes in each mode with no regressions and brings the original fixed cohort to 408 cases.
• The normal strict gate completed both modes and remained red on the unresolved queue. No missing API, scheduler argument, scheduler injection, provider, or scheduler-class behavior was introduced.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface remediation wave 13

• Integrated separate reviewed commits for ordinary standalone zip completion and cancellation, per-case static-zip capability liveness, five finite observation horizons, and explicit projection-overload classification.
• Ordinary zip(sources) now completes when any completed input's queue is exhausted, completes for zero inputs, and stops activating later synchronous inputs after termination. The exploratory fillAfterComplete path remains separate. D-022 records the lifecycle contract.
• Focused source evidence proves concurrent polyfill observers share one zip state and one activation per input: one observer leaving retains the work, while the final observer abort cancels every input.
• Removed a false per-case dependency on an unused suite-level queueScheduler alias without adding scheduler support. Exactly 20 non-projection static-zip cases are now FIXED; the missing operator Symbols, scheduler case, and result-selector overloads remain untouched.
• Two identity-shaped projection cases now happen to produce the expected tuples, but remain explicitly compatibility-only/missing-api and TODO because the projection overload is not represented. Their output coincidence is not counted as API support.
• Complete one-process audits recorded 848 cold passes with 1,490 failures and 865 polyfill passes with 1,473 failures. That adds 22 passes in each mode with no regressions and brings the original fixed cohort to 428 cases.
• All 91 focused source tests passed. The normal strict gate completed both modes and remained red on the unresolved queue. No missing operator, projection overload, scheduler argument, scheduler injection, provider, or scheduler-class behavior was introduced.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 existing-surface final audit and zip hardening

• Replaced the first standalone-zip completion patch with reviewed commit 9ab5aa5cb, after reverting it in f16352d9b. The replacement retains the shortest-input completion and sibling-cancellation behavior while also draining the exploratory fillAfterComplete path without an all-complete infinite loop. Ten focused zip tests cover zero inputs, synchronous empty inputs, final-buffer draining, input errors, shared ref-counted observers, last-observer cancellation, and finite fill completion.
• Commit ab482214b corrected the first finite zip observation boundary from frame 15 to frame 18 so it preserves the full original source-diagram horizon. The expected notifications were not changed.
• Commit 98277dbc9 corrected instance combineLatest so the lower-level combine primitive, rather than both layers, owns the receiver subscription. Four focused tests preserve exact tuple arity, single receiver activation, cancellation, static forms, and one shared ref-counted activation for concurrent polyfill observers.
• Complete one-process audits recorded 865 cold passes with 1,473 failures and 865 polyfill passes with 1,473 failures. The 17 new cold passes are legacy projection cases that now terminate without invoking their selector; they do not represent the missing projection overload and remain TODO. Manifest generation now classifies all 31 such legacy projection cases as compatibility-only missing-api, preserving the same 428-case fixed cohort with zero IN-PROCESS or BLOCKED rows. The strict cold-plus-polyfill gate completed and remained red only on the unresolved queue; all 101 focused source tests and the parity freshness check passed.
• A fresh audit found 136 remaining TODO cases whose programs mention rxTestScheduler or testScheduler; 110 currently fail first on the undefined name. None are missed safe helper-receiver rewrites: 105 exercise real scheduler arguments or injection, 14 exercise scheduler parser, queue, export, or other internals, and 17 fail first on an unrelated missing API. A global scheduler alias was rejected because it would turn those explicit scheduler claims into misleading harness behavior.
• The final existing-capability audit found no remaining non-scheduler case backed by a current implementation or executable mapping. Remaining unresolved rows require missing non-scheduler APIs, unsupported legacy overloads, compatibility decisions, or the scheduler-last phase. The next implementation phase is missing non-scheduler operators and functions; scheduler arguments, providers, classes, and internals remain the absolute final P0.T3 phase. The scheduler-last boundary also retains the current animationFrames case and 20 timeout configuration cases: their callbacks do not all pass an explicit scheduler, but the claims are inherently about frame or timeout scheduling.
• The package build and lint commands still hit the documented exploratory package baselines: tshy rejects source-path exports and ESLint points the RxJS package at the Observable package's TypeScript project. Neither command reached a change-specific diagnostic.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — Workspace test-command normalization

• Audited all five workspace package manifests and confirmed that @rxjs/observable-polyfill, @rxjs/observable, @rxjs/test, rxjs, and rxjs.dev each expose a package-level test command for their relevant suite.
• Renamed the RxJS ported-test command surface from test:ported:* to test:unit:* while retaining test/ported as the source-provenance boundary. The ordinary RxJS test target now delegates to test:unit, which runs 101 focused source tests before the complete cold and polyfill parity modes. Watch mode remains limited to the focused source suite.
• Updated the command guide, testing architecture and decision record, port notes, generated parity and failure-ledger instructions, and their generators. The parity-document freshness and fresh failure-tracker checks passed, and no user-facing test:ported command remains outside historical plan evidence.
• Package test commands passed for @rxjs/observable-polyfill (53 tests), @rxjs/observable (18), and @rxjs/test (72). The renamed RxJS target passed all 101 focused tests, then exercised all 2,338 cases in both modes and remained intentionally red at 865 passes and 1,473 failures per mode on the existing P0.T3 queue.
• The rxjs.dev test command correctly invokes the Angular unit-test target, but the existing legacy toolchain fails before collection with ts26.createNull is not a function. Command examples now pass Angular options directly through pnpm without an extra --.
• P0.T3 remains the single project-level NEXT item.

2026-07-29 — P0.T3 missing non-scheduler API wave

• Added and publicly mapped exact Symbol contracts for notifier gates, property/query/empty/sequence utilities, static partition and generation, count windows, and recursive expansion.
• Preserved platform-owned string methods, one shared/ref-counted state machine per active producer run, AbortSignal cancellation, and nonterminal count-window cancellation. Generator-owned rewrites bound never and cancellation evidence without changing RxJS 7 notifications or source subscriptions. Reused recursive platform fixtures retain their intentional shared-producer expectation.
• Passed 108 focused tests for the newly added APIs. Complete 2,338-case audits recorded 1,508 cold passes with 830 failures and 1,505 polyfill passes with 833 failures, adding 150 passes in each mode with no regressions. The durable ledger now contains 1,067 FIXED and 854 TODO rows, with no IN-PROCESS or BLOCKED rows.
• The remaining generate, expand, and startWith scheduler overloads stay in the scheduler-last queue. The strict test:unit gate remains red only on unresolved packets. P0.T3 remains the single project-level NEXT item, and P0.2 is not advanced.

2026-07-29 — P0.T3 completion

• Completed the remaining operator, creation, composition, lifecycle, replay, and scheduler-timing packets in small reviewed commits. The final host-time surface includes exact Symbols for timestamp, timeInterval, sampleTime, delay, bufferTime, windowTime, observeOn, and subscribeOn, plus corrected animationFrames and timeout lifecycle behavior.
• Kept platform timing on host APIs and AbortSignal cancellation. Legacy scheduler classes, providers, queue/parser internals, and scheduler arguments were not restored in production; their ported behavioral claims run through explicit @rxjs/test timing rewrites and test-local compatibility evidence under D-033.
• Regenerated the authoritative manifest and ran complete audits. Cold passed 2,338 of 2,338 cases, and polyfill passed 2,338 of 2,338 cases. The failure tracker retains all 1,923 historical failures across 140 packets and marks every row FIXED.
• Passed the strict pnpm --filter rxjs test contributor gate: 98 focused source-test files with 705 tests, followed by both complete parity modes. Parity-document freshness and zero-failure ledger regeneration checks also passed.
• Marked P0.T3 DONE and restored P0.2 as the single project-level NEXT item.

2026-07-30 — Direct global host-scheduling boundary

• Removed the exploratory animationFrameProvider runtime and parity fixture. animationFrames now schedules and cancels directly through late globalThis lookups.
• Normalized RxJS Next host timers, intervals, animation frames, clocks, and cancellation calls to explicit globalThis.* access without changing the narrow timestamp-provider overloads.
• Rewrote inherited provider-spy evidence around public Observable behavior and direct host APIs. Added focused integration coverage proving that already imported RxJS modules and ordinary application scheduling share the same rxTest virtual host timeline.
• Recorded D-034. P0.2 remains the single project-level NEXT item.

2026-07-30 — Hot and cold producer terminology

• Defined hot and cold solely by whether the producer exists before a subscription or is created during it; sharing, multicasting, replay, and ref counting remain separate properties.
• Removed the “cold until subscribed” shorthand for the platform Observable and documented its exact lifecycle: the first subscription creates an active producer, concurrent subscriptions join it, and a later subscription after ref-count closure creates another.
• Recorded that every instantiated Subject is hot and that ColdSubject is misnamed prototype debt. Its material lifecycle distinction from Subject is per-observer subscription plumbing used by the BehaviorSubject and ReplaySubject prototypes for late-observer replay.
• Made no rename or removal. That choice remains part of the compatibility and package-boundary design, and P0.2 remains the single project-level NEXT item.

2026-07-30 — Per-subscription Subject base rename

• Renamed the exploratory ColdSubject class and public subpath to the abstract PerSubscriptionSubjectBase; no compatibility alias remains.
• Added a protected _subscribe hook with detailed lifecycle guidance, retained the lower-level fanout helper for replay ordering, and made the constructor protected so the base is visibly subclass-only.
• Migrated the behavior- and replay-subject factories and added focused evidence that the hook runs for every direct subscription, including late terminal subscribers.
• Recorded D-036 and the native-interoperability warning: platform methods can use the internal subscription algorithm and bypass an overridden JavaScript subscribe() method.
• Passed the full rxjs gate: 100 focused files with 710 tests, followed by all 2,338 cold and 2,338 polyfill parity cases. The parity document freshness check and git diff --check also passed.
• The pre-existing package-tooling failures remain unchanged: ESLint still points at packages/observable/tsconfig.json, and tshy rejects the current root export-array shape before compiling sources. A direct TypeScript diagnostic produced no errors in the renamed base or its behavior/replay subclasses after filtering out the repository's existing errors.
• This user-directed compatibility cleanup does not advance P0.2, which remains the single project-level NEXT item.

2026-07-30 — ColdObservable native and Symbol construction boundary

• Accepted one globally shared, ABI-versioned construction protocol at Symbol.for('rxjs.kernel.create.v1') while retaining exact module-owned Symbols for public operators and factories.
• Made protocol installation idempotent for compatible callable slots and a hard error for a non-callable collision.
• Kept ColdObservable as a real platform subclass. Native string methods delegate through fresh base Observable views and return platform Observables; native Promise consumers activate the source through the same boundary.
• Overrode [create] so RxJS Symbol operators return plain ColdObservables, including when the receiver is a behavior or replay Subject derived through PerSubscriptionSubjectBase.
• Removed mergeMap and switchMap native-method shortcuts that bypassed the receiver's [create] policy in their default configurations.
• Added focused constructor, native-method inventory, Promise-consumer, duplicate-copy Symbol identity, Symbol-operator, and derived-Subject tests.
• Regenerated the source-pinned port manifest for the accepted cold-subclass identity rule. The full RxJS gate passes 717 focused tests and all 2,338 cases in both cold and polyfill modes.
• Recorded D-037. This user-directed compatibility/kernel slice does not advance the package-map decision; P0.2 remains the single project-level NEXT item.

2026-07-30 — Symbol-based async iteration strategies

• Temporarily prioritized and completed P0.I1 at the user's direction, using rxjs-for-await revision 94f9cf9cb015ac3700dfd1850eb81d36962eb70f as the behavioral source.
• Added exact Symbols for lossless each-value and buffered iteration plus lossy latest-value and next-demand iteration. Every invocation returns a fresh, lazy, one-shot async generator and aborts its observer during generator cleanup.
• Documented and tested the lifecycle distinction: concurrent generators keep independent conversion state while joining one shared, ref-counted platform producer, whereas ColdObservable starts an independent producer for each generator.
• Removed the two exploratory standalone conversion helpers and replaced their package subpaths with the four explicit Symbol modules.
• Passed 16 focused async-iteration tests and the full RxJS gate: 101 focused files with 733 tests, followed by every one of the 2,338 cold and 2,338 polyfill parity cases.
• Confirmed the package build stops at the unchanged P0.3 tshy root-export baseline before source compilation. Recorded D-038 and restored P0.2 as the sole project-level NEXT item.

2026-07-30 — P0.2 package and installation contract

• Accepted @rxjs/observable-polyfill, rxjs, and @rxjs/test as the final target package map and selected @rxjs/observable for removal in P0.3.
• Made the polyfill package independently publishable and responsible for the base ambient platform types. rxjs depends on it and initializes the current realm through every public entry. D-043 later added the public ColdObservable dependency used by @rxjs/test.cold().
• Defined a non-install-all rxjs root for intentional non-operator core APIs and capability-scoped Symbol subpaths.
• Preserved existing Observable and EventTarget.when implementations, specified paired fallback installation, and defined the read-only observablePolyfillInfo/getObservablePolyfillInfo detection contract.
• Limited initialization to the importing window, iframe, worker, or server isolate and left transparent cross-realm operation, hardened surfaces, and untested runtimes outside the initial support claim.
• Superseded the proposed RxJS 7 compatibility library and pipeable facade. Retained cold, Subject, and Symbol-composition APIs as intentional Next contracts and made migration documentation and Skills the supported path.
• Recorded D-039 through D-041, reconciled the charter, architecture, migration-evidence policy, open questions, future phases, risks, and package diagrams, marked P0.2 DONE, and advanced P0.3 as the sole NEXT item.
• This decision-only step changed no package metadata or runtime code.

2026-07-30 — Cached bundle-size comparison

• Temporarily prioritized and completed P0.DX3 without changing the accepted package map or implementing the P0.3 package build.
• Added pnpm run analyze:bundles to compare fresh native and fallback RxJS Next bundles with cached published RxJS root bundles in one static webpack-bundle-analyzer UI.
• Isolated the pinned Webpack, analyzer, TypeScript loader, and published RxJS installs from the workspace. Published caches are keyed by exact version, registry integrity, and bundle-configuration fingerprint; --refresh rebuilds them explicitly.
• Added focused Node tests and documented version selection, cache reuse, refresh, no-open behavior, and generated artifact locations.
• pnpm run test:bundle-analysis passed 8 focused tests covering CLI parsing, tag-to-version normalization, configuration fingerprints, cache paths, source discovery, and combined-stats construction.
• A default 7.8.2 cache-miss run generated the native, fallback, and published assets plus the static report. A second run reported the published target as reused and left its bundle, stats, and metadata timestamps and SHA-256 hashes unchanged; --refresh then rebuilt that cache generation and advanced its metadata timestamp.
• The final combined map contained distinct rxjs-next-native.js, rxjs-next-polyfill.js, and rxjs-7.8.2.js assets with ungrouped module records. The Observable fallback appeared only in the polyfill target, and no specs or src/testing modules were present.
• git diff --check passed. The disposable installs changed neither the root dependency graph nor pnpm-lock.yaml; generated reports and published caches remained in ignored output locations.
• Restored P0.3 as the sole project-level NEXT item.

2026-07-30 — P0.3 package implementation

• Removed the unused @rxjs/observable package and its preparation, workspace, TypeScript, README, and lockfile references without changing the concurrent bundle-analysis implementation.
• Implemented conditional, transactional fallback acquisition with frozen version metadata and detection; preserved existing constructors and EventTarget.when, and limited the abort bridge to fallback installation.
• Made the three accepted packages independently buildable with correct dependency direction, repository metadata, root/subpath exports, generated declarations, and no source specs or self-links in dry-run packs.
• Added package declaration consumers, ESM/CommonJS imports, and isolated realm fixtures covering every P0.3 installation and side-effect claim.
• Passed the frozen-lockfile install, workspace discovery, publication preparation, all three package gates, 49 polyfill/harness tests, 73 @rxjs/test tests, pinned-import verification, and strict conformance with 525/525 upstream subtests and 52/52 identity attestations.
• Recorded the newly visible stale-artifact focused-test discrepancy for P0.4: after rebuilding the polyfill, 678/733 focused RxJS tests pass before the ported parity launcher runs.
• Marked P0.3 DONE and advanced P0.4 as the sole project-level NEXT item.

2026-07-30 — P0.4 native/fallback lifecycle safety rail

• Added one package-independent lifecycle contract and ran it unchanged against the built fallback in Node and the native Observable in pinned Chrome 150.0.7871.126; all five grouped cases passed.
• Covered shared activation, late joins, observer and producer cancellation, restart, completion, error, synchronous reentrancy, teardown registration and ordering, and host reporting for thrown observer callbacks.
• Added mixed ESM/CommonJS duplicate-install fixtures in both load orders and wired the lifecycle contract into the pinned and latest-Chrome WPT jobs.
• Reconciled the rebuilt focused RxJS suite to 733/733 passing tests. A later correction audit removed invalid wrappers that replaced RxJS 7 arbitrary-subscribable inputs, leaving those 15 claims as executable compatibility failures instead of manufactured passes.
• Recorded 2,323/2,338 cold passes and 2,316/2,338 fallback passes. The latter includes the same 15 legacy-input failures plus seven fallback-only retry/shared-error failures. The focused suites, package gates, package source suites, parity/tracker freshness checks, shared lifecycle contract, pinned-import verification, and the then-strict 525-subtest WPT conformance gate pass.
• Marked P0.4 DONE and advanced P0.5 as the sole project-level NEXT item.

2026-07-30 — P0.4 Subscriber<void> correction

• Removed the call-site workaround that changed void notifications from subscriber.next() to subscriber.next(undefined).
• Kept the ordinary next(value: T) signature; TypeScript already permits its argument to be omitted for Subscriber<void> while rejecting an omitted value for Subscriber<number>. Cold and test subscribers need no special overloads or casts.
• Made an omitted fallback runtime argument emit undefined; added type and runtime regression coverage and recorded D-042.
• Passed all package builds, polyfill type/import/source tests, 73 @rxjs/test tests, the 733-test focused RxJS suite, and the five-case native/fallback lifecycle contract. The later correction audit records 15 shared legacy-input failures and seven additional fallback lifecycle failures without changing their programs.
• The strict pinned WPT diagnostic now passes 522/525 upstream subtests and all 52 identity attestations. Its Web IDL, window, and worker argument-presence failures are the explicit D-042 divergence; upstream sources remain unchanged.
• Kept P0.5 as the sole project-level NEXT item so the revision policy can formalize how accepted divergences are recorded.

2026-07-31 — Subscriber<void> correction audit

• Removed the unnecessary conditional-tuple overloads, optional implementation parameters, casts, and dematerialize workaround introduced during the first correction attempt.
• Confirmed with the repository TypeScript compiler that the ordinary next(value: T) signature already permits next() for T = void and still rejects an omitted value for T = number.
• Restored ColdSubscriber, PlatformSubscriber, the independent test subscriber, and dematerialize to their pre-correction source exactly.
• Removed generated platform-Observable wrappers that had changed 15 RxJS 7 arbitrary-subscribable claims. Those programs now retain their original inputs, carry compatibility-only classifications, and fail honestly.
• Retained only the fallback runtime removal of the missing-argument throw, one runtime regression test, one public type regression test, and the documentation required for the resulting WPT divergence.
• Corrected a native lifecycle-runner branch so assertion failures report immediately instead of surfacing as browser timeouts, and regenerated the stale manifest and failure ledger from complete audits.
• Package builds, type coverage, 50 polyfill tests, 73 @rxjs/test tests, and all 733 focused RxJS tests pass. The complete audits record 2,323 cold and 2,316 fallback passes out of 2,338. P0.5 remains the sole NEXT item.

2026-07-31 — P0.DX4 real test files and stock reporting

• Replaced the single dynamically registered parity suite and custom shard/progress launcher with 294 formatted Vitest files: 147 cold and 147 platform files, each using normal describe, it, direct public Symbol calls, and await rxTest(...).
• Kept RxJS 7 provenance as source comments while making every failure location a real repository path and line. Normal commands use the unmodified Vitest default reporter.
• Refactored rxTest fixtures so cold explicitly extends ColdObservable, hot extends the active global constructor and derives ordinary platform Observables, and observable() directly uses globalThis.Observable.
• Removed the cold-mode global constructor replacement. Cold static factories now name ColdObservable explicitly in the migrated examples.
• Used Vitest's built-in JSON reporter for complete evidence and mapped results through a static migration report. Audits record 2,296 cold and 2,316 fallback-platform passes out of 2,338; the failure tracker was regenerated without machine IDs in test titles.
• @rxjs/test source and public type suites pass with 75 runtime tests. Native mode remains explicitly skipped in the current Node realm. P0.5 remains the sole NEXT item.

2026-07-31 — P0.M1 owned migrations and @rxjs/migrate

• Added the independently publishable @rxjs/migrate development package with a framework-neutral semantic transform, caller-supplied capability maps, a generic framework-adapter interface, a Mocha/Chai-to-Vitest adapter, dry-run-first CLI, bundled Skill assets, and two read-only source-content MCP tools. The transform also handles production pipe(...) expressions without requiring a test framework or TestScheduler.
• Dogfooded the package while adopting 294 files as ordinary repository-owned Vitest source: 147 cold and 147 platform files, each with file-level source repository, exact revision, and path provenance. All 2,338 cases per mode are direct tests that call rxTest; parameterized registrations are materialized rather than dynamically generated.
• Removed the test-file generator, generated tree, custom Chai/spy support layer, and generator script. The Vitest config now collects the owned files directly with its stock default reporter. Audit and tracker tools use a static migration report only for case identity.
• Verified 9 migration-package tests, lint, builds, ESM/CommonJS imports, CLI dry run, package publication contents, Skill helper/portability checks, 75 @rxjs/test tests, and 733 focused RxJS tests. Formerly dynamic share and multicasting samples pass all 176 direct tests.
• Complete JSON audits cover all 2,338 tests per mode and retain 2,296 cold and 2,316 fallback-platform passes with no pending tests. The reviewed product failures remain ordinary failures; a focused concat run points to the real checked-in spec line. Native mode skips explicitly in the current Node realm.
• Recorded D-044, marked P0.M1 DONE, and advanced P0.5 as the sole NEXT item.

2026-07-31 — P0.5 complete pinned Observable conformance

• Restored required-argument behavior for fallback Subscriber.next and made Subscriber<void>.next() a declaration error while preserving explicit next(undefined) delivery. The runtime check is enforced before delivery even when the subscriber is already closed.
• Updated only platform-Subscriber void notification call sites; unrelated Subject<void> APIs retain their own contracts.
• Recorded WICG/observable d74bace7cf80200a01c81cfe20961e29ac7fa3d8 as the written reference and web-platform-tests/wpt 6a009d73f0d315941b90cac13a9523a2a08c631b as the executable gate. D-045 supersedes D-042 and permits no RxJS-specific WPT exceptions.
• Passed focused runtime and type tests, import/provenance verification for all 52 generated URLs, the five-case native/fallback lifecycle contract, and the complete-result baseline.
• Strict pnpm run test:wpt passed in Chrome 150.0.7871.126: 52/52 URLs were OK, 525/525 upstream subtests passed, and 52/52 exact RxJS implementation attestations passed, with zero failures, errors, timeouts, skips, not-run results, or accepted-failure metadata.
• Marked P0.5 DONE and advanced P1.1 as the sole project-level NEXT item.

2026-07-31 — Agent-first migration experience reprioritized

• Made the complete RxJS 7-to-Next migration user experience the top project priority and returned P1.1 to PLANNED without changing its scope.
• Added P0.M2 through P0.M5 for product and fixture design, deterministic @rxjs/migrate hardening, one portable Skill with Codex/Claude/Cursor integration guidance, and representative end-to-end outcome qualification.
• Recorded the primary workflow as repository assessment, coverage warnings, optional pre-migration characterization tests that pass on RxJS 7, explicit Next-contract selection, bounded codemods, and an interactive green build and test loop.
• Required both deterministic golden/compile/idempotence/behavior fixtures and nondeterministic agent evaluations based on contract manifests and outcomes. MCP is now an evidence-driven optional decision rather than an assumed deliverable.
• Set P0.M2 as the sole NEXT item. This was a plan-only reprioritization; no product code or test baseline changed.

2026-07-31 — P0.M2 agent-first migration product design

• Audited the P0.M1 package, its nine registered transform tests, unreferenced source fixture, independently authored repository Skill, bundled package Skill, and two-tool MCP prototype. Treated those as prototype evidence, not as proof of a safe end-to-end migration product.
• Added MIGRATION_TOOLING_DESIGN.md with the agent journey, version and coverage assessment, RxJS 7 baseline and characterization gates, explicit contract manifest, bounded mechanical work, iterative repair, and reviewed closeout.
• Defined separate mechanical and agent evaluation schemas, required behavior categories, outcome gates, and negative controls. A transform cannot select lifecycle intent or turn missing evidence into a completion claim.
• Accepted D-046: packages/migrate/skill is the canonical Skill source; thin Codex, Claude Code, and Cursor adapters preserve one package version and digest; the deterministic engine remains subordinate; and no MCP ships.
• Partially superseded D-044 and reconciled the charter, architecture, compatibility policy, open questions, testing design, port notes, package README, plan risks, and P0.M3-P0.M5 handoffs.
• Verified formatting, documentation consistency, local links, and exactly one active queue row. Marked P0.M2 DONE and advanced P0.M3 as the sole NEXT item. No production code or test baseline changed.

2026-07-31 — P0.M3 deterministic migration engine hardening

• Hardened @rxjs/migrate around a versioned evidence registry, structured diagnostics, contract-manifest readiness, atomic transforms, contained batch writes, and a machine-readable dry-run-first CLI.
• Added exact fixture, pinned RxJS 7 behavior/type, current Next type, idempotence, negative-control, import, declaration, and pack gates. The new target type lane found and locked a corrected switchMap result generic.
• Removed every MCP release artifact and dependency under D-046, documented only the ten mechanically supported mappings, and exposed deterministic canonical Skill integrity for the next distribution slice.
• Passed 91 migration tests, the aggregate migration package release gate, migration lint, RxJS public type checks, and diff checks. Marked P0.M3 DONE and advanced P0.M4 as the sole NEXT item.

2026-07-31 — P0.M4 portable Skill and harness adapters

• Promoted the package Skill from a thin helper into the complete portable migration workflow and validated its Agent Skills structure.
• Added safe generated-copy installation, inspection, update, and removal for Codex, Claude Code, and Cursor, with structured results, content provenance, atomic replacement, and protection for local modifications.
• Replaced the older independently authored repository Skill with the digest-matched .agents/skills/rxjs-next-migration copy. The removed files remain recoverable from Git history.
• Added the cross-harness guide and 21 adapter acceptance cases. Passed 114 migration tests, package build/types/imports/pack, lint, Skill validation, generated-copy checks, and diff checks.
• Marked P0.M4 DONE and advanced P0.M5 as the sole NEXT item. Live model-backed repository outcome runs remain deliberately assigned to P0.M5.

2026-08-01 — P0.M5 Codex/ChatGPT migration qualification

• Added four pinned, offline-executable RxJS 7.8.1 application/library seeds covering Vitest, Mocha, Jest, strong and weak coverage, cold, platform, mixed, and unsupported migration contracts.
• Ran the representative matrix with Codex 0.146.0-alpha.3.1, gpt-5.6-sol at medium reasoning, and canonical Skill/engine 8.0.0-alpha.14. Three runs completed approved migrations and the fourth made its required weak-coverage/unsupported safe stop.
• Passed 4/4 records and 14/14 semantic gate families. Committed four records and 20 hashed artifacts covering the prompt/final response, contract manifest, patch, command results, and final report. Recorded explicitly that full event streams were not retained for the first three runs.
• Preserved D-047's claim boundary: Claude Code and Cursor retain P0.M4 installation/discovery support, but their migration outcomes remain unmeasured.
• Marked P0.M5 DONE, completing Phase 0, and advanced P1.1 as the sole project-level NEXT item.

2026-08-01 — Phase 1 closure audit

• Reviewed P1.1 through P1.3 against the completed P0.3, P0.4, P0.5, P1.4a, and P1.4b evidence. Confirmed that their installation and conformance outcomes were already met and that additional fallback implementation would duplicate completed work.
• Passed 51 focused polyfill/harness tests, package build/type/import fixtures, mixed-format duplicate installation, and the five-case lifecycle contract against the packaged fallback and native Chrome 150.0.7871.126.
• Passed strict offline WPT conformance with 52/52 OK URLs, 525/525 upstream subtests, and 52/52 exact RxJS identity attestations.
• Kept older RxJS extension and migration-evidence failures outside the platform fallback phase, corrected the stale Phase 1 risk language, marked P1.1 through P1.3 DONE, and advanced P2.1 as the sole project-level NEXT item.

2026-08-01 — P2.1 public Symbol and installation policy

• Accepted D-048 and resolved the remaining public Symbol identity, version, duplicate-copy, realm, collision, descriptor, and package side-effect questions.
• Kept public capabilities exact and module-owned, retained only the versioned construction ABI and fallback metadata as global-registry exceptions, and rejected a stronger construction marker as unnecessary.
• Replaced the unreviewed Symbol.for('buffer') key with an exact Symbol. The focused buffer suite passes 12/12, and a source audit finds no other public extension on a global-registry key.
• Marked P2.1 DONE and moved the sole NEXT marker to P2.2.

2026-08-01 — P2.2 transactional extension installer

• Added one internal typed installer for static-only, instance-only, and paired exact Symbol capabilities.
• Added preflight, identical-value idempotence, non-enumerable descriptors, exact-key conflict rejection, named frozen-target diagnostics, and rollback after a definition failure.
• Declared the package side-effectful as required by D-048. Six installer tests and the strict source type-check pass; the packed bundler proof remains in P2.4.
• Marked P2.2 DONE and moved the sole NEXT marker to P2.3.

2026-08-01 — P2.3 Observable kernel helper contract

• Accepted D-049, separating receiver-driven derived construction from active-realm platform input conversion and bounding subclass, borrowing, cancellation, and error behavior.
• Added focused internal helpers for [create], Observable.from, signal-owned source subscription, and synchronous exception forwarding; removed two unused pre-kernel helper files.
• Six focused helper cases and the strict source type-check pass.
• Marked P2.3 DONE and moved the sole NEXT marker to P2.4.

2026-08-01 — P2.4 extension-kernel pilot and Phase 2 closure

• Migrated map, scan, switchMap, timeout, timer, and pipe to the transactional installer and the D-049 construction, conversion, cancellation, and error-forwarding helpers.
• Passed 756 focused source tests, strict source typing, package and declaration builds, ESM/CommonJS and duplicate-dialect fixtures, frozen-target handling, bundler retention/tree-shaking checks, and the eight-case kernel contract against both the packaged fallback and native Chrome 150.0.7871.126.
• Confirmed that the only targeted migrated-test exception is the classified compatibility-only switchMap arbitrary-subscribable case. Recorded the broader nonzero strict baselines as Phase 3/4 work rather than expanding Phase 2 into RxJS 7 compatibility repair.
• Marked P2.4 DONE, completed Phase 2, and advanced P3.1 as the sole project-level NEXT item. P3.1 begins by reconciling the restoration and evidence inventory already produced during Phase 0 instead of recreating it.

2026-08-01 — P3.1 restoration inventory audit

• Reconciled 117 public source subpaths, 97 exact extension Symbols, the 111-operator/21-factory/29-value capability registry, and all 2,338 source-pinned executable RxJS 7 registrations.
• Confirmed that behavior restoration is already broad: only the six Phase 2 pilots use the common installer, while 91 existing exact capabilities need implementation-pattern migration rather than new API design.
• Rebased restoration into four migration-value families and added explicit P3.2 through P3.4 completion bars so ledger work, kernel adoption, and evidence closure cannot be conflated.
• Preserved every compatibility-only input and lifecycle claim; no missing or unsupported registration is hidden from collection.
• Marked P3.1 DONE and advanced P3.2 as the sole project-level NEXT item.

2026-08-01 — P3.2 generated migration-evidence ledger

• Added a deterministic 161-entry ledger spanning every registry operator, creation/static function, value, type, and standalone function.
• Joined the registry to 2,271 distinct source-pinned evidence cases and kept seven uncovered entries visible with no implied pass.
• Recorded the required import, evidence, Next surface, lifecycle, cancellation, classification, type, migration, adapter, and decision fields in machine-readable JSON plus a readable generated Markdown view.
• Added generation and freshness commands and updated the compatibility policy to name the ledger as the Phase 3 evidence authority.
• Marked P3.2 DONE and advanced P3.3 as the sole project-level NEXT item.

2026-08-01 — P3.3 complete extension-kernel adoption

• Migrated the remaining 91 exact capabilities in nine small family commits; all 97 current exact public Symbols now use transactional installation.
• Preserved existing implementation bodies, public Symbols, declarations, subpaths, and lifecycle behavior while making paired static/instance installation atomic.
• Added a permanent source audit to the ordinary unit and package gates so a future direct public constructor/prototype assignment fails immediately.
• Passed all family suites, then the complete 107-file/756-test focused suite, the public type consumer, the installation audit, and the multi-dialect package build.
• Marked P3.3 DONE and advanced P3.4 as the sole project-level NEXT item.

2026-08-01 — P3.4 classified restoration evidence

• Corrected three cold claims at the accepted hot-to-platform lifecycle boundary and passed their 63 focused registrations.
• Re-ran complete 2,338-case audits with zero pending registrations: cold records 2,299 passes and fallback records 2,316 passes.
• Classified every remaining ordinary failure by exact case ID. Cold retains 24 intentional lifecycle divergences plus 15 compatibility-only legacy-input claims; fallback retains seven intentional divergences plus the same 15 compatibility-only claims. No portable or harness-rewrite failure remains.
• Regenerated the migration evidence, recorded the reviewed cold pass baseline, marked P3.4 DONE, completed Phase 3, and advanced P4.1 as the sole project-level NEXT item.

2026-08-01 — P4.1 intentional Next API stabilization

• Recorded D-050 for the accepted cold, Subject-family, advanced subject-base, and exact Symbol composition imports and lifecycle contracts.
• Made the widened Observable<T> declaration boundary explicit when the runtime creation protocol preserves ColdObservable.
• Added and passed a published-declaration consumer for every intentional API, root/subpath identity, Subject view, factory/config type, advanced subclass, and static/instance [pipe] composition form.
• Marked P4.1 DONE and advanced P4.2 as the sole project-level NEXT item.

2026-08-01 — P4.2 completed prioritized migration ledger

• Closed every one of the 161 prioritized rows with direct, canonical-alias, or supplemental pinned/focused evidence; zero rows remain uncovered.
• Added per-row cold and fallback pass/failure accounting and made unexplained failure classifications a generation error.
• Finalized type status at 150 changed, seven preserved, and four compatibility-only rows with no deferred status.
• Regenerated and verified the machine-readable and Markdown ledgers, marked P4.2 DONE, and advanced P4.3 as the sole project-level NEXT item.

2026-08-01 — P4.4 accepted migration fixtures

• Added four source-linked fixtures for cold, shared platform, hot Subject, and unsupported scheduler/interop migration outcomes.
• Compiled the three accepted targets against public Next declarations and executed their lifecycle, cancellation, Subject-state, and negative controls.
• Verified all 166 migration-package tests, marked P4.4 DONE, and advanced P4.I1 as the sole project-level NEXT item.

2026-08-01 — Direct public Symbol installation decision

• Accepted D-051 and partially superseded D-048's transactional installation mechanism while preserving its exact module-owned Symbol identity, realm, duplicate-copy, and bundling policy.
• Recorded direct assignment as the target public-extension pattern. The accepted contract no longer includes a common installer, custom descriptors, exact-key conflict checks, repeat-install idempotence, extensibility preflight, rollback, or named errors for unsupported hardened targets.
• Added P4.I1 with explicit code-removal, audit, fixture, bundle-size, and verification gates. This decision-only session changes no production source or test baseline.
• Advanced P4.I1 as the sole project-level NEXT item after P4.4 completed.

2026-08-01 — P4.I1 direct public Symbol installation

• Migrated all 97 exact public capabilities from the common installer to ordinary direct assignment in ten small family commits without changing their Symbols, declarations, implementations, construction, or cancellation.
• Deleted the installer and its six transactional-only tests, then changed the permanent source audit and package fixtures to enforce the D-051 contract.
• Reduced the representative rxjs/map bundle by 1,279 minified bytes (8.1%), 340 gzip bytes (7.4%), and 307 Brotli bytes (7.4%); the root-only control was byte-identical.
• Passed the 106-file/750-test focused suite, all 97 direct-installation audit cases, public types, build, package/import/bundler fixtures, and all eight fallback/native Chrome kernel cases.
• Marked P4.I1 DONE, completed Phase 4, and advanced P5.1 as the sole project-level NEXT item.

2026-08-01 — P5.1 package-local migration guidance

• Added the task-oriented RxJS 7-to-9 guide under packages/rxjs, covering lifecycle selection, Symbol imports, cancellation, teardown, input conversion, Subjects, schedulers, testing, automation limits, and closeout.
• Moved the generated migration-evidence and unsupported-surface references into packages/rxjs/docs; their existing generation and freshness gates now own those package-local paths.
• Added a declaration consumer for the guide's public examples and updated the repository compatibility links without changing production behavior.
• Accepted RxJS 9 with 9.0.0-beta.0 under D-007 and recorded D-052 for package-local documentation and the explicit apps/rxjs.dev exclusion.
• Passed the public type, generated-doc freshness, parity freshness, link, format, diff, and no-rxjs.dev-change checks. Marked P5.1 DONE and advanced P5.2 as the sole project-level NEXT item.

2026-08-01 — P5.2 representative migration closure audit

• Mapped the P5.1 migration journey to the existing P0.M3 deterministic, P4.4 contract-fixture, and P0.M5 four-repository evidence and recorded that package-local map beside @rxjs/migrate.
• Confirmed the matrix covers applications and libraries, Vitest/Mocha/Jest, strong and weak coverage, cold, shared platform, mixed, Subject, and unsupported contracts, with three completed migrations and one timely safe stop.
• Re-ran all 17 migration suites and 166 tests. Exact transforms, compilation/idempotence, behavior, lifecycle negative controls, seed oracles, 14 semantic gate families, all four captured records, all 20 artifact hashes, and mutation controls pass.
• Passed the migration package build, public types, ESM/CommonJS imports, and publication dry run. No new live-model or cross-harness claim was added.
• Marked P5.2 DONE, left the two explicitly deferred Phase 5 ideas deferred, and advanced P6.1 as the sole project-level NEXT item.

2026-08-01 — P6.1 release identity and support policy

• Accepted D-053: Node 22.13+ and 24 are blocking, Node 26 is advisory, and latest Chrome/Firefox, desktop and Mobile Safari, Webpack 5, Deno, and Bun form the blocking non-Node matrix.
• Kept RxJS 9 prereleases on npm next, RxJS 7 on latest during beta, and RxJS 7 maintained without assigning a sunset date.
• Synchronized the four-package train, internal dependency edges, runtime identities, fixtures, lockfile, and installed migration Skill adapter to 9.0.0-beta.0; historical qualification evidence remains immutable.
• Replaced ESM/CommonJS/browser/webpack copies with one dist/esm build. Browser, Webpack, import, and Node require(esm) conditions share that implementation; the Node bridge and shared exact-Symbol identity pass.
• Added tested ESM-distribution and release-coherence checks, passed all four builds, declaration consumers, and import fixtures on Node 24.12.0, and confirmed that apps/rxjs.dev was neither changed nor executed.
• Marked P6.1 DONE and advanced P6.2 as the sole project-level NEXT item.

2026-08-01 — P6.2 release gates and environment evidence

• Added blocking package lanes for Node 22.13+/24 and advisory Node 26, plus package-built Deno 2.8.0 and Bun 1.3.14 runtime contracts.
• Added one cross-browser contract for current Chrome, Firefox, and WebKit and separate SafariDriver lanes for branded desktop and Mobile Safari; the Mobile Safari request requires an actual iOS simulator.
• Added Webpack ESM-resolution and 22,000-byte bundle checks plus measured map and cancellation budgets. Local Node 24 evidence was 17,502 bytes, about 50.4M map values/second, and about 140.5K cancellations/second.
• Passed the four-package train, focused product suites, supported-runtime imports, browser-engine contract, and pinned 52-URL/525-subtest WPT closure. Retained the reviewed 39 cold and 22 fallback migration divergences as non-blocking classified evidence.
• Extended release coherence to reject environment-lane or documentation-site drift. Marked P6.2 DONE and advanced P6.3 as the sole project-level NEXT item.

2026-08-01 — P6.3 package-local documentation and README refresh

• Replaced the repository README with a current RxJS 9 beta introduction, prominent “What happened to RxJS 8?” explanation, first-use example, package map, support boundary, and contribution path.
• Added package-owned READMEs for the runtime, polyfill, and test packages; refreshed the migration package; and added RxJS API, release-gate, and package-contributor guides.
• Moved the migration tooling design beside @rxjs/migrate, kept all public package guidance inside its owning container, and published the applicable README/docs paths from each manifest.
• Added a tested documentation gate that rejects missing published paths, broken or cross-package local links, and documentation-site references. Four npm dry runs include the intended local documentation and Skill content.
• Marked P6.3 DONE and advanced P6.4 as the sole project-level NEXT item; the separate documentation application remained untouched and unused.

2026-08-01 — P6.4 packed prerelease adoption and beta approval

• Added a tarball-content and size auditor plus an isolated offline consumer that exercises ESM, Node require(esm), strict types, @rxjs/test, and a browser bundle from the actual package artifacts.
• Passed adoption on Node 22.13.0, 24.12.0, and advisory 26.5.0. The bundle was 20,222 bytes; all four tarballs stayed below their checked-in ceilings with package-local documentation and no duplicate dialect or source spec.
• Passed the Nx first-release version dry run and retained the three-completion, one-safe-stop migration adoption evidence from P5.2.
• Accepted D-054: the synchronized 9.0.0-beta.0 train is approved for npm next after blocking CI is green; this does not publish, promote to latest, or approve stable 9.0.0.
• Marked P6.4 DONE and advanced P6.5 as the sole project-level NEXT item for the terminal plan and no-documentation-site closure audit.

2026-08-01 — P6.5 terminal closure audit

• Completed the frozen-lockfile audit and verified the resulting lockfile offline. The only native-build warning came from a legacy optional dependency outside the four release packages and did not fail installation.
• Closed the @rxjs/test lint false positive caused by the intentional repository test-graph cycle, and declared its previously masked polyfill import as an exact test-only dependency.
• Passed all four package lint and package gates, release/documentation checks, packed-consumer adoption, Webpack size, performance floors, workflow parsing, and diff hygiene from the cleanly installed workspace.
• Confirmed that apps/rxjs.dev has no worktree change and that no site build, test, publication, or deployment command ran.
• Marked P6.5 DONE. The execution queue is complete with no remaining NEXT item; P5.3 and P5.4 retain their accepted DEFERRED status.

2026-08-01 — P6.6 centralized source subscriptions

• Consolidated source notification forwarding, setup/callback error handling, and destination/local cancellation into the positional D-049 helper across every semantically compatible operator and source module.
• Preserved the 24 reviewed raw subscription boundaries required by root-core, Subject-like, retained-lifecycle, async-generator, host-error, and terminal ordering semantics; removed every obsolete helper identifier and type.
• Passed focused, lint, type, package, import, kernel, Webpack, and performance gates and retained the existing cold/fallback audit classifications.
• Recorded byte-identical root control, non-growing map, 6.99–8.45% representative compression reductions, and 13.32–16.50% catalog compression reductions. Marked P6.6 DONE with no remaining NEXT item.

2026-08-01 — P6.7 direct derived construction

• Replaced the remaining derived-construction wrappers in map, scan, switchMap, timeout, and timer with direct D-037 [create] calls, then deleted both wrappers, their obsolete options type, and their dead tests.
• Passed the complete source, lint, public-type, package/import, native/fallback kernel, Webpack, and performance gates while retaining the reviewed cold and fallback audit classifications.
• Recorded a byte-identical root control and reductions in every encoding for map, the representative operator set, and the complete runtime catalog. Marked P6.7 DONE with no remaining NEXT item.

2026-08-01 — P6.8 complete CI ownership

• Added D-055 and made exact cold/polyfill pass-set comparison, migration manifest and document freshness, Symbol-installation inventory, bundle-analysis, SafariDriver, and active-workflow checks blocking in main CI without changing ordinary test:unit semantics.
• Retained blocking Node 22.13 and maintained Node 24 package gates, made Node 26 advisory, ran TypeScript-latest for pull requests and master, and made pinned WPT and release readiness unconditional on master with complete pull-request path coverage.
• Extended release coherence to guard commands, runtime lanes, clean build prerequisites, Safari simulator selection, advisory status, and unfiltered master triggers.
• Validated draft PR #7611: all blocking main CI, TypeScript-latest, pinned WPT, browser, bundle, performance, adoption, Deno, Bun, desktop Safari, and Mobile Safari jobs passed. Node 26 passed as advisory and scheduled latest-Chrome remained advisory.
• Classified and corrected CI configuration failures separately from product behavior; no product-test failure was found. Marked P6.8 DONE and returned the completed queue to no NEXT item.

2026-08-02 — P6.9 README status and security automation

• Reopened the completed queue at the user's direction and added authoritative CI, release-readiness, WPT, Scorecard, npm-channel, download, license, and architecture badges without adding unsupported coverage or size claims.
• Corrected prepublication copy so npm next can truthfully display the older prerelease without implying that RxJS 9 is published or installable there.
• Added SHA-pinned Scorecard publication/SARIF and pull-request dependency review, with only the required narrow permissions and no license enforcement or automated PR comments.
• Passed YAML parsing, active-workflow and Markdown formatting, all 24 release checks, documentation links, release coherence, badge image/target probes, and diff hygiene. P6.9 remains NEXT for PR and post-merge live validation; no runtime, package API, publication artifact, or apps/rxjs.dev file changed.

2026-08-02 — P6.10 secure staged release implementation

• Accepted D-057 and added the irreversible-publication runbook, deterministic scenarios, exact-tarball boundary, npm-link validation, recovery, and bootstrap.
• Replaced private Nx release APIs and NPM_TOKEN publishing with a GitHub-App release PR, protected-commit qualification, artifact attestations, exact-tarball environment gates, stage-only npm OIDC, TFA approval order, and registry-integrity-driven GitHub Release finalization.
• Added release policy/candidate/staging tests, the read-only doctor, CodeQL, full action pin enforcement, and action Dependabot. No npm, tag, release, or external-account mutation was performed locally.
• Corrected the first PR CI findings by making source-commit binding explicit only at the privileged release boundary and replacing incomplete HTML-comment sanitization with non-rendering placeholder detection covered by edge tests.
• P6.9 is blocked only on its first GitHub-side security runs. P6.10 is the sole NEXT item until administrator setup and the disposable-package rehearsal prove the external npm/GitHub controls.

2026-08-02 — P6.10 single-maintainer hardening

• Replaced every team/reviewer assumption with the accepted single-maintainer invariant. Ordinary successful merges still automatically create or refresh the release PR; self-merging it starts read-only qualification.
• Split qualification from staging. Two fresh exact-toolchain builds must match byte-for-byte, all blocking gates use the canonical tarballs, and the run stops after publishing its ID, version, and manifest SHA-512. Only benlesh can later type those values. A no-OIDC authorization job validates them before the protected staging job can start and revalidate for stage-only npm OIDC; tested rejection covers actor, run, branch, commit, version, digest, expiry, bytes, and replay.
• Added checked npm 11.18.0 integrity, CycloneDX, exception-free release OSV, stable attestation assets, registry/signature/attestation finalization, and consumer verification documentation. Moved Scorecard into that contextual assurance document and accepted Code-Review 0.
• Upgraded or overrode every dependency path reachable from release, build, or test tooling. The remaining 260 npm advisory paths are isolated to excluded apps/rxjs.dev, represented by 177 reviewed OSV IDs expiring 2026-10-31, and guarded by a path-aware checker. Added bounded PR and scheduled extended fast-check properties for release controls and Observable lifecycle.
• Local rehearsal produced two byte-identical four-package candidates, hydrated the retained tarballs, and passed 1,044 package/source tests, all four type and import suites, migration packing/contracts, the 45-test release/security suite, workflow formatting, package preparation, and a frozen install. The broader strict cold compatibility command reproduced the reviewed 2,299-pass, 39-failure baseline rather than introducing a new failure.
• P6.10 remains the sole NEXT item because WebAuthn, GitHub/npm rules and trusted-publisher configuration, live workflow evidence, and the sole-account disposable-package rehearsal are external and have not yet been verified.

2026-08-03 — P6.10 dependency-advisory response

• Responded to GHSA-rgw5-rvv9-x895 by overriding every affected brace-expansion major in the workspace to its patched release: 1.1.18, 2.1.4, and 5.0.9. The frozen lockfile contains no affected version.
• Re-ran OSV Scanner 2.3.8 and the npm path classifier. Four newly reported advisory IDs were confined to excluded apps/rxjs.dev; three obsolete IDs were removed. The baseline generator now canonicalizes linked GHSA aliases so regeneration cannot restore duplicate, unused exceptions. The configured root scan has no unreviewed findings or unused exceptions, while P6.10 remains the sole NEXT item.

2026-08-02 — P6.10 Copilot review follow-up

• Classified indented multi-line BREAKING CHANGE and BREAKING CHANGES footers as breaking while retaining the non-rendering pull-request-template placeholder behavior.
• Ensured that an unchanged template placeholder cannot mask a later populated breaking footer in the same squash-commit body.
• Replaced the release doctor's workflow-wide runner substrings with exact release-PR, qualification, authorization, and staging job runner validation, including a false-positive test where an unrelated job still uses the expected runner.
• Scoped privileged runner discovery to the workflow's jobs mapping and rejected duplicate privileged job definitions so top-level lookalikes cannot satisfy the audit.
• Kept P6.10 as the sole NEXT item; these local review fixes do not satisfy the remaining administrator setup or disposable-package rehearsal gates.

2026-08-04 — P6.10 review-branch conflict resolution

• Rebased the Copilot follow-up onto the merged single-maintainer hardening and dependency-advisory work, dropping the two release commits already present on master instead of replaying or reverting them.
• Adapted runner validation to the split release-PR, qualification, authorization, and staging workflows while preserving exact Ubuntu 24.04, macOS 15, Node 24.12.0, OSV, property-test, and checked-toolchain controls.
• Passed all 53 release/security tests, 177 OSV exception validations, release coherence and doctor checks, workflow formatting, and diff hygiene. P6.10 remains the sole NEXT item pending its external setup and rehearsal gates.

2026-08-04 — P6.10 first-beta staging policy and dry-run evidence

• Replaced the public disposable-package rehearsal with exact-tarball npm pack --dry-run, npm publish --dry-run, and npm stage publish --dry-run checks plus stage-only GitHub trusted-publisher previews. All sixteen dry runs passed over a local four-package 9.0.0-beta.0 candidate with the pinned npm 11.18.0 CLI and no registry credentials. The local manifest SHA-512 was d8977e21abb704df0c72ed2ba5663e282f81b6e4be24d3ff6a87f9922eadfc22c09de1b04adf998e4c8817f37ce15e669cb714b82c2f983400d9417430526585; it is diagnostic evidence, not the future canonical Ubuntu qualification digest.
• Split the exact release policy into blocking master checks, pull-request-only branch-protection checks, and advisory checks. Added regression coverage that rejects PR-only or advisory checks in RELEASE_REQUIRED_CHECKS, malformed or duplicate configuration, weak branch protection, and drift from the exact check set.
• Verified PR #7613's successful Dependency review and the successful post-merge Scorecard run 30919216705, completing P6.9. P6.10 remains the sole NEXT item: GitHub App/ruleset/environment administration, npm trusted publishing, canonical Ubuntu qualification, private staging, Ben's WebAuthn approvals, and public registry verification still have to succeed.

2026-08-04 — P6.10 interactive release simplification

• Rejected the staged design after confirming npm's initial-publication prerequisite: rxjs has a public registry record, while @rxjs/observable-polyfill, @rxjs/test, and @rxjs/migrate return 404 and therefore cannot use npm staged publishing or trusted-publisher configuration for their first version. The earlier dry runs did not contact the registry and did not reveal that blocker.
• Superseded D-057 with D-058. Deleted the GitHub App before any stage or publication, then removed its repository private-key secret and App-ID variable plus the obsolete NPM_TOKEN. The unrelated Firebase and Nx Cloud secrets remain unchanged; no ReactiveX organization setting was changed.
• Removed five privileged release workflows and the App, release-PR, two-builder, candidate, OIDC, staging, typed-authorization, doctor, and finalizer scripts. Retained ordinary CI, release readiness, CodeQL, dependency review, OSV, Conventional Commit validation, and package gates.
• Added pnpm release:beta <9.0.0-beta.N> with clean synchronized-master, interactive-terminal, and no-environment-token guards; synchronized package checks; local package gates; exact tarball SHA-512 display; npm publish dry runs; exact-version confirmation; supporting-package-first/rxjs-last publication; resumable integrity matching; and next/latest verification.
• Passed 37 release, documentation, coherence, Conventional Commit, and OSV tests; all four package build/type/import gates; workflow formatting; and four exact-tarball npm publication dry runs. Nothing was published. P6.10 remains the sole NEXT item until the command publishes and verifies 9.0.0-beta.0, package access disallows automation tokens, and the immutable GitHub Release is recorded.

2026-08-05 — P6.11 callback receiver removal

• Audited the complete RxJS production source and removed callback thisArg overloads and .call dispatch from every, filter, find, findIndex, map, and partition; no platform-polyfill protocol receiver binding was changed.
• Reworked focused and migrated tests to use ordinary closures or bound callbacks, added negative public-type assertions, and classified the six source-pinned RxJS 7 receiver cases as intentional divergences while preserving their historical IDs.
• Passed 750 source tests, both 76-test affected migrated suites, exact complete cold/polyfill audit baselines, public types, lint, package/build/import and migration freshness gates, the packaged fallback/native Chrome kernel, and the release performance floor.
• Recorded D-059, marked P6.11 DONE, and retained P6.10 as the sole NEXT item.