[QA-CAMPAIGN] Reconcile 32 retained original HTTP MCP predicates and current release gates

Current completion register — final beta.49 owner evidence

Original historical acceptance: 21 PASS / 32 OPEN of 53. The original 518-context universe and 17 planned exclusions remain unchanged. The final beta.49 assessment has zero new original advances. Earlier 37-open and 53-open checkpoints below are historical.

Current feature acceptance: the supplied owner evidence completes the bounded criteria of #399, #394, #390, #382, #361, #312 and #141; each is closed with its exact source/partner limits. The policy Web parent #398 stays open for #414 (active OIDC renewal crash) and #415 (Inspect focus restoration), now being repaired for beta.50.

Measured observation: beta.49 completed 240/240 actual successful calls over 3,600 seconds through the declared observable published bridge protocol consumer. Selected native Codex preflight was separate. This is not opaque AI-host scheduling or a retroactive pass for the original historical hour.

The 32 original gaps comprise original AI (10), protected producer/reader (10), classification/attenuation (3), QA identity (2), service/progress (2), immutable snapshots (2), authority restoration (1), original hour (1) and original cleanup (1). Preserve immutable owner/runtime/prompt requirements and unknown-delivery state. Current successors remain separately versioned.

The next laptop campaign verifies the beta.50 Web delta, retains unaffected prior passes and cleans up its newly owned state. A targeted pass and the original historical register are separate outcomes. Do not recreate missing predecessors or repeat the hour solely because this Web release number changes.

Historical completion register — beta.47 delivery, beta.46 owner credit

Owner acceptance remains OPEN / NO-GO. The supplied beta.46 owner checkpoint advanced 16 of the 53 previously retained original contexts. The unresolved original register is now 37 rows: 12 P0, 23 P1 and two P2. The 17 planned exclusions remain inside the original 518-context universe. These 37 rows are unresolved assertions and prerequisites, not 37 newly demonstrated product defects.

The older 53-row count and its source/release evidence below remain historical. Beta.47 source work and publication do not advance an original row without its actual acceptance receipt.

Credited beta.46 outcomes

Retain the exact attributed native AI connection/control/readback results, original durable readbacks, three original ordinary-reader redaction passes, both supported browse-continuation results, and the supported parameterized prompt-message workflow. The separate Linux beta.45→beta.46 automatic update also proves the repaired beta.45 coordinator's activation path: same enrolled identity, Accepted result, acknowledged delivery and matching running executable hash, without unzip. #361 now retains only its still-unproven legacy-helper obligation.

Real daemon smoke turns and a local-account Web turn remain useful observations within their original scope. They do not substitute for the missing exact conversation vectors, real uncertain-delivery predecessor or genuine SSO first turn. Disposable snapshot/command regressions and a Windows service Start/Stop regression do not silently replace their historical target, progress-association or original fault/fallback assertions.

Original unresolved register

Group	Priority	Remaining original rows
Original AI create/send/approval/reconcile/archive	P0	10
Original QA identity recovery and start preview	P0	2
Owned event retry/disable	P1	4
Protected output producers and reader matrix	P1	11
Policy/classification/attenuation	P1	3
Windows service fault branch and selected-native progress	P1	2
Job definition/input snapshot integrity	P1	2
Historical temporary authority restoration	P1	1
Isolated measured native hour	P2	1
Final owned cleanup/restoration	P2	1
Total		37

Beta.47 source delivery

Services refresh correlation: selected-native deployed acceptance passed #399: durable request-to-terminal Services correlation, current-session fencing, coalescing/throttling and explicit legacy attribution. The new capability requires an upgraded client and a fresh request; the old uncorrelated request stays unproven.
Bounded MCP policy Web UI: beta.50 refresh and focus acceptance pending #398: bounded administrator policy Web lifecycle with exact preview/confirmation, current admission inspection, authoritative version/audit readback and original replay receipts. Real browser acceptance remains due.
Windows updater: supported recovery and controlled automatic progression accepted #312: a supported checksum/attestation-bound Windows recovery artifact with guarded state-preserving cutover. The actual LocalSystem recovery and subsequent automatic beta.46→beta.47 chain remain due.
The prior callback (Durable callback rejection and explicit replay: deployed acceptance passed #394), paired authoring (Paired request-definition authoring and input sync: deployed acceptance passed #390), SSO authority (SSO investigation authority: genuine first turn and continuation accepted #382), Linux updater (Linux updater: legacy installed-helper repair and automatic progression accepted #361) and monitoring/flow/connector (Owned monitoring-to-RatelDesk two-window incident acceptance completed #141) corrections are retained with their own remaining live gates. The requested dependency changes are consolidated in the release PR.

Historical evidence and prospective successors

Keep original IDs, assertions, predecessors, attempts, exclusions and earlier credited passes unchanged. Where a historical runtime/backup, immutable caller-owner tuple, literal question or pre-correlation refresh is permanently unavailable, define the precise versioned successor before execution and link it to the original coverage anchor. Retain the unresolved original and record successor coverage separately. A new enrollment is not identity recovery; an empty reconciliation is not recovery of an uncertain delivery; a newer inventory is not proof of an old refresh. Do not rewrite ownership, revive expired/revoked credentials or invent missing inputs.

The next laptop handoff supplies explicit successor specifications only for these demonstrated unavailable prerequisites. The final verdict must report unresolved historical evidence and current-release coverage separately, without reducing the original denominator.

Remaining owner release gates

Verify the deployed matching image digests, migration and selected native consumer against the completed publication. Complete the original functional/security assertions and any required versioned successors, genuine SSO turn, paired companion create/sync/execution, callback terminal-rejection/replay/transient behavior, supported Windows recovery and automatic progression, legacy Linux-helper repair, and one real monitoring occurrence → retained incident with duplicate suppression and recovery/rearm.

Obtain a supported off-host backup and actual isolated restore/decryption continuity for the matching database, Data Protection and required application/native identity material. Storage inventory or a local copy is not restore proof; retain private key material and operational evidence outside this public tracker.

Run the original selected-native endurance sequence only after its functional/security prerequisites: 240 serial slots over 3,600 seconds, 15-second cadence, concurrency one and the declared four-operation rotation, with no interleaving, replacement slots or catch-up. Finish owned cleanup and authority restoration in dependency order, retaining audit/updater history and unrelated state.

Continue independently ready work while a precise blocker is resolved. Keep #294 and the issue-specific live acceptance trackers open until their evidence is present. Use the exact-source campaign guide; the owner laptop handoff is for live execution, not another source implementation/release assignment.

Delivered in the consolidated beta.47 change at source f587d8f050ac991e126f3d6ab9d3375191a220e6. Distribution evidence: beta.47 release, all-target release build, and completed publication. These references establish source/build/distribution delivery; live acceptance is recorded separately.

Current completion register — beta.46 source, release and owner acceptance

Owner acceptance remains OPEN / NO-GO. The supplied beta.45 checkpoint advanced four original log-gap/resynchronization contexts and retains 53 unresolved original contexts: P0 22, P1 27, P2 4. The 17 planned exclusions remain within the 518-context universe. The remaining rows are unattempted or prerequisite-gated, not 53 newly demonstrated product defects. Private report and receipt identities remain outside this public tracker.

Beta.46 source delivered and independently validated

PR #397 is merged at exact source 4dcf0a4d09d05c350143039c8652be9ba8471df8. It delivers the following corrections:

Authenticated Netclaw diagnosis: connection tests now identify bounded connection, DNS, TLS, authentication, endpoint/protocol, timeout and protected-token failures with safe repair guidance in API responses and the saved Web view. Tenant/current-revision checks and caller cancellation remain enforced. A real bug is fixed: transport disposal failure after a successful connection can no longer leave a successful result paired with a failure code.
Notification disable race: atomic state/lease claims and conditional completion preserve an accepted disable when a worker selected the notification earlier or publication is already in flight. A stale worker can no longer overwrite Disabled with its late success/failure result; publication claims recheck current state and lease ownership. An already admitted publication can finish. Disabling a notification controls its outbox record and MCP retry; it does not cancel the originating workflow or the separate durable callback reconciler.
Exact owned-event discovery: bounded exact eventType, correlationId and notification status filters now run from MCP through the authenticated API to the query. Callback failure/rejection notifications may be Published: publication status does not describe the callback outcome. Redacted replay metadata, disabled-event retry-preview rejection and deterministic equal-timestamp paging make the owned event assertion discoverable without scanning unrelated pages.
Working development Netclaw preparation: the integration guide uses supported daemon lifecycle commands and requires isolated tenant daemon contexts for native SignalR sessions. The laptop campaign agent is assigned actual setup and repair, described below.
Current residual preparation: the existing offline checker validates the latest remaining-context ledger against exact original IDs/counts/exclusions, pins its hash, separates durable QA readbacks from runtime recovery, distinguishes platform prerequisites and supports the applicable owner decision path. Earlier ledgers and their evidence remain unchanged.

The two source bugs above have direct regression coverage. The final PR receipts confirm that all four new PostgreSQL notification concurrency cases executed and passed. The change also adds 18 Netclaw service cases, nine event adapter/API/database cases, seven preparation checks and assertions in existing schema/component tests. No retained assertion or required workflow gate was removed.

Independent validation	Source	.NET result	Source/release contracts	Complete run
Final PR validation	`be2d9594f8310940cfa040ed08e54511597a5c14`	5,152 passed; 0 failed; 0 skipped	175 passed	16m50s
Merged-main validation	`4dcf0a4d09d05c350143039c8652be9ba8471df8`	5,152 passed; 0 failed; 0 skipped	175 passed	16m12s

Each .NET execution comprises 4,524 Core + 36 API integration + 592 component tests. These are separate executions of the same suites, not additional live campaign passes. Both complete validation runs satisfy the under-20-minute requirement. Main's conservative elapsed time uses the completed run's final update timestamp; its final job completed one second earlier.

Beta.46 release and distribution

Release completion: COMPLETE — source and build/distribution integrity verified; functional acceptance remains pending-owner-testing.

Beta.46 tag build: SUCCESS on attempt 1 at the exact merged source; all 8 jobs passed, 5,152 regressions passed with zero failed/not-executed, 175 contracts passed, and 15 Windows MCP bounded synthetic smoke checks passed (20m30s).
Beta.46 publication: SUCCESS on attempt 1 at the exact merged source; all 7 jobs passed, including five image jobs and final anonymous-pull, metadata, disclosure-scan and public-asset verification (27m28s including the tag-build wait).
Independent public distribution verification, exact tag/source and publication bindings, public assets/checksums and immutable image references: PASS: 15 uploaded assets, 14 checksum entries, 13 authenticated build inputs, five immutable images, all five SPDX files and the derived Compose bundle match the exact source and completed publication. The public Windows MCP archive hash matches the verified synthetic smoke receipt. Independent selected public downloads were hashed; the successful hosted Finish job performed all-asset download checks, anonymous image pulls, OCI metadata/license checks and repeated disclosure scans. These scans do not establish a container-vulnerability or live-readiness pass.

Release references: v0.1.1-beta.46, publication record and checksums. Source validation, completed distribution, owner deployment and live acceptance require their own evidence. No release/source work advances the 53 original contexts without their exact acceptance receipts.

Deployment remains an owner action using the completed matching publication and canonical migration sequence. Verify actual running image digests and successful migration before the deployed assertions. Publication does not establish functional readiness.

Retained beta.45 evidence and credit

Beta.45 was published and distribution-verified: release, exact source b3e935d706b2b461b143c643fc6172b0edad0e42, merged PR #396. Required main CI and the all-target release build each passed 5,121 tests with zero failures or skips. All eight build jobs and all seven publication jobs succeeded. The release has 15 uploaded assets, verified public checksums and five immutable image references; independent public record and Compose checks matched the exact source and image pins. Its publication record retained pending-owner-testing.

The beta.45 source/distribution work advanced zero exact original contexts. Its subsequent owner campaign supplied the four original log-gap/resynchronization passes and passed the signed-in target/profile/policy/audit route gate. Retain those passes and their exact scope. Preserve the 53-row residual register and the 17 existing exclusions until further exact owner receipts change them.

This issue is the single remaining campaign register for the overlapping work formerly split across #282, #288, #304, #305, #306, #308 and #310. Their criteria are retained below and in their immutable histories. Closing those tracking tickets as superseded does not mark their unexecuted assertions passed.

Exact remaining groups

Group	Priority	Original contexts	Required completion
Authenticated AI	P0	18	Reachable paired daemon, genuine authenticated turns, required approval continuation/recovery and selected-system attenuation.
Historical QA identity	P0	4	Complete original durable client/job readbacks independently; recover the original runtime or identity-preserving backup for its separate runtime/start assertions.
Event retry/disable	P1	4	Original owned receiver/predecessor, actual delivery/retry/disable effects and authoritative readback.
Protected output readers	P1	14	Fresh fixed-native standalone/job synthetic secrets, required ordinary/no-reveal/restricted readers, stream/error/export and server-enforced untrusted-context outcomes.
Policy and attenuation	P1	3	Original classification, revision/race, authority/expiry/revocation and target-boundary assertions.
Windows service / native progress	P1	2	Owned inert Windows service timing and original observable selected-consumer progress requirements.
Job snapshot integrity	P1	2	Exact reviewed definition/input snapshot, post-edit execution and stale-start protection.
Native prompt consumer	P1	1	Preserve already-attributed prompt-message workflow passes; complete the remaining original consumer assertion under its declared evidence contract.
Temporary authority restoration	P1	1	Actual original auxiliary/role restoration with authorized readback.
Browse continuation	P2	2	Real supported Linux/Windows continuation, exact values/order, no gaps/duplicates, and relevant stale/current-authority controls. Historical Unsupported evidence remains.
Isolated endurance	P2	1	Exactly 240 selected-native calls over 3,600 seconds, 15-second cadence, concurrency one, original four-call rotation, no interleaving/replacement slots/catch-up.
Final owned cleanup	P2	1	Original and new owned fixtures removed/retired in dependency order with authoritative readback; retained audit/history and unrelated state preserved.

Load the original private definitions and latest appended ledger once. Operation names alone are not full assertions. Preserve exact predecessor/effect requirements, all attempts and attributed earlier passes; do not rerun unaffected work solely because a beta number changed. Newly discovered operations have separate added rows and do not inflate original progress.

Active development Netclaw assignment

The owner has assigned the laptop campaign agent to set up, repair and run development Netclaw until the required connections and real turns work. Prepare the real daemon and model provider, verify reachability from the actual API runtime, pair each required tenant once using its current saved revision, test the saved pairing, and complete genuine authenticated sessions, a real pending-approval continuation and the required interrupted-session reconciliation. The separate SSO first-turn/continuation gate remains required.

Use a separate isolated daemon context for each intended NetRatel tenant, with only that tenant's bounded MCP registration and permitted tools available in its context. Native SignalR sessions share the daemon's Personal tool audience; distinct session names or pairing tokens do not provide tenant isolation. Follow the canonical Netclaw integration guide and AI operations guide. Preserve genuine approval decisions, current authority, selected-system attenuation and tenant boundaries.

Continue the available supported repair paths until the checks pass or a precise nondelegable dependency is demonstrated. Record the exact affected context IDs and continue independent ready groups while that dependency is resolved. A generic unavailable-daemon result does not complete the setup assignment. Retain the working development setup after campaign-owned temporary objects are cleaned up.

Additional release and seven issue-specific obligations

Read actual deployed API/Web/MCP image digests and successful migration receipt against the completed publication. Runtime version alone does not prove running image pins.
Prove supported off-host Data Protection key backup and isolated restore with identity/decryption continuity. Keep key material outside this tracker.
Windows updater: supported recovery and controlled automatic progression accepted #312: recover the designated Windows canary through a supported path that preserves identity and historical state, reconcile its historical result, then prove a subsequent automatic update without ACL or state-file manipulation. If recovery first installs beta.45, a subsequently eligible published beta.46 offer can supply the next update; its eligibility and result must be observed. Protected original clients remain excluded.
Linux updater: legacy installed-helper repair and automatic progression accepted #361: retain the evidenced clean no-unzip first-upgrade pass; complete applicable legacy installed-helper recovery and unattended activation under the repaired coordinator, with final accepted result/hash. An incoming update into beta.45 used the previously installed coordinator. Proving the repaired beta.45 activation path requires beta.45 already active and a later eligible published offer. A manual service start remains qualified recovery.
SSO investigation authority: genuine first turn and continuation accepted #382: genuine SSO first turn and continuation against the reachable intended daemon; source tests using a test daemon do not replace this.
Paired request-definition authoring and input sync: deployed acceptance passed #390: paired request-definition create/input-sync/replay/binding cases after deploying source that includes Add paired request-definition authoring for RatelDesk beta.9 #392, with an owned execution fixture.
Durable callback rejection and explicit replay: deployed acceptance passed #394: an owned terminal peer rejection produces one durable rejected phase and no repeat over multiple worker intervals; transient failure recovers and audited replay is explicit. Notification publication status and notification disable are assessed under their own outbox contract.
Owned monitoring-to-RatelDesk two-window incident acceptance completed #141 and Flows: RatelDesk incident connector with tenant mapping and idempotent delivery #146: real monitoring occurrence → retained RatelDesk incident, duplicate suppression, recovery and fresh-occurrence behavior. Obsolete dual-consent ceremony is not reinstated.
Retain the remaining owner-mode current/stale/revoked/expired/cross-owner/cross-tenant controls. Beta.40 create/protected delivery/revoke/replay and beta.44 reconcile/effective dispatch are credited within their recorded scope.
The credited beta.45 signed-in route pass does not discharge the original configured login-mode or notification assertions; retain their independent requirements.

Consolidated preparation and completion

Complete one consolidated preparation and preflight across deployment, login/daemon, disposable executors, original recovery data, readers and owned sinks. Continue the required setup and repair until the checks pass or the exact nondelegable dependency is evidenced; recheck the prerequisites affected by each repair. Record exact affected context IDs for blockers and continue ready independent groups. Avoid repeating unchanged denied calls.

Run the original isolated hour only after its functional/security predecessors are satisfied and competing traffic stops. Reconcile cleanup last, including administratively retiring MCP-owned objects whose original credential was revoked; owner-filtered 404 is not absence. Preserve original identities, earlier attempts, current-authority controls, required human decisions and retained audit/history throughout.

Use the exact-source completion guide together with this updated preparation instruction. Keep all private identities, infrastructure values, credentials and raw receipts outside this public issue. Source/distribution completion and owner readiness remain separately reported.

Retained historical issue body (earlier candidates, instructions and evidence)

Current candidate: beta.31; final beta.29 owner audit is the retained baseline

Owner acceptance remains OPEN / NO-GO. Continue the full retained native HTTP MCP campaign on v0.1.1-beta.31, delivered through merged #340 at f89593ccef6fa7385779baf1ba18328426218c07. The tag resolves to that exact source. Its source delivery and public distribution do not establish owner deployment or fresh campaign acceptance.

Corrected PR validation and merged-main validation each passed 4,973 tests (4,352 Core + 36 API + 585 Components), with zero failures/skips; complete runs took 11m54s and 16m57s. These are separate executions of the same suites.

Historical tag/publication evidence: tag build attempt 1 failed one Core case (4,351 passed, one failed, zero skipped); API and Components did not execute. The recorded mechanism review supports a pre-existing TestServer/gRPC request-flush race, and the failed execution remains retained. Publication attempt 1 stopped on that prerequisite before registry writes or public asset uploads; it produced no publication journal or release asset. The one documented failed-jobs recovery on the unchanged tagged source completed successfully as build attempt 2, followed by successful publisher attempt 2. No source, test, assertion or workflow gate was changed for this recovery. The 168 passing release contract tests from build attempt 1 remain retained successful evidence.

Tag build: attempt 2 succeeded with all eight required jobs; fresh source validation passed 4,973 tests (4,352 Core + 36 API + 585 Components), zero failed/skipped, in 17m34s. All four unsafe-frame cases passed the original assertions; Windows MCP passed 15 bounded synthetic checks; three native Client archive integrity checks and all five package provenance-upload steps passed. The 168 contract tests are the retained successful attempt 1 execution. Publication: attempt 2 succeeded with all seven required jobs in 7m29s; five anonymous digest pulls/rescans, all 15 public assets and 14 public-download checksum matches passed. Independent distribution verification: 2026-10-19T12:37:08.811855+00:00; 15 completed-publication files matched public asset metadata, with 14 checksum entries, 13 artifact hashes, exact source and authenticated build attempt 2, all five Compose image pins, and the published Windows MCP ZIP matching its 15-check smoke receipt. Public downloads and anonymous image pulls were verified by the publication workflow; the independent local check used the completed artifact and public digest metadata. Functional acceptance remains pending owner testing. Use the release's publication record, checksums, and exact-source campaign guide.

The initial PR run built successfully but failed 13 existing ordinary-delegation assertions. The production boundary was corrected while retaining those assertions. Keep that failed execution separate from the corrected passing PR/main receipts.

Final beta.29 evidence supersedes the earlier phase1 counts

The owner-tested source was 4b148ebe824b3de445a12d30e95b8f5c30d53655. The final checkpoint preserves these distinct populations:

Evidence population	Final beta.29 result
Bounded named assertions	268
Functional assertions	135/161 passed; 26 failed assertions across five P2 defect families
Raw native receipts	811
Retained contexts invoked	283/518
Genuine native prompt workflows	0/8
Qualifying endurance	0/240 calls; 0/3600 seconds

All 518 retained context IDs and histories remain: 438 original, 60 earlier additions and 20 beta.27 additions. Append beta.31 attempts without resetting IDs or replacing earlier failures. Current dispositions are 225 bounded operation passes, 3 catalog/schema-only, 218 untested, 28 admission denied, 25 failed, 14 prohibited (all uninvoked), 2 negative-enforcement-only, 2 unsupported and 1 deferred; total 518. New contexts need a separately reported denominator.

Retain the 20 policy receipt mappings across nine operation contexts: scope comes from nested/persisted selectors. The earlier labels did not establish positive control-plane mutation coverage. No aggregate readiness percentage is valid across these different evidence populations.

Five replay families and the separate source-found regression

Final-audit family	Delivered correction / fresh acceptance
DEF-B29-01 / #333	Exact definition replay: update, parameter add/update/delete and step add/update/reorder/delete in both applicable tenants; 16 originally failed contexts. Preserve original result identity/version after later changes.
DEF-B29-02 / #336	Script deletion replay through retained scoped ownership, without another delete or ordinary tombstone visibility.
DEF-B29-03 / #337	Policy replace/disable/revoke replay before mutable lifecycle/version checks, preserving the original normalized result.
DEF-B29-04 / #338	Artifact cleanup replay without repeated cleanup; cleaned status/download remain unavailable.
DEF-B29-05 / #339	Cryptographically receipt-bound deleted job/run replay through actual local HTTP exchange, repeated current admission and business confirmation.

#334 remains an additional source-review run cancel/delete regression, separate from the five owner-observed defect families. Rerun it on beta.31 with valid current authority, exact original-result equality and no repeated action. Each family also needs altered payload/identity/target, current policy/credential revocation and original-expiry denial controls. Fresh stale/invalid requests remain denied and unconsumed. Direct handler tests alone do not prove the deployed delegation path.

Required owner preparation and next campaign order

Owner deployment and recovery: verify the released API/Web/HTTP MCP source and immutable image digests, actual migration receipt and protected key/storage continuity. Retain the original QA recovery and data-preservation obligations. The owner performs deployment, recovery and infrastructure changes.
Actual fixed native execution and selected consumer: provide independently registered disposable Windows/Linux executors; prove their actual loaded implementation and protected identity separation. The owner performs supported consumer replacement/reconnect. Verify the active selected process, implementation hashes, complete loaded schemas and effective native reads in both original tenants. Archive availability or a separate diagnostic process is insufficient.
Exact current authority and fixtures: preserve the original primary scope and exact approved expiry, supported no-reveal auxiliary constraints, original protected assets and exact registered fixture boundaries. Use actual independent review only where currently required; no fabricated human decision, administrative bypass, narrowed scope or silent expiry extension. Keep existing Prohibited contexts uninvoked.
Fresh functional and security acceptance: complete all five replay families plus [beta.30][P2] Reconcile successful job-run cancel and delete confirmations after state transitions #334, confirmation/race/stale-state controls, persisted profile/tag/revision and isolated rollback cases, fresh standalone/job synthetic-secret reader matrices, native inert-service timing, values/paging and server-enforced untrusted-context tests. Complete live monitoring/flow/event/delivery through a verified owned synthetic sink. Keep diagnostics, denied admission and positive business effects distinct.
Native protocol and all eight real prompt workflows: inspect_client, diagnose_job_run, controlled_remote_execution, create_workflow, manage_operator_access, safe_file_operation, safe_terminal_operation and onboard_client. Record actual native invocation/effects and required human checkpoints. The owner performs onboarding installation/enrolment. Prompt retrieval, rendered text and tools-only approximations do not complete these workflows. Prove progress before the final response and actual in-flight protocol cancellation separately from application-job cancellation.
Exact isolated endurance: after other diagnostics/review/fixture traffic ends, issue 240 selected-native calls at offsets 0 through 3585 seconds, observe through 3600, cadence 15 seconds and concurrency one. Repeat tenant A flows/list, auth/status, tenant B flows/list and system/version 60 times. Report 120 core and 60 reads per tenant separately. No interleaving, replaced retry slots, overlap or catch-up bursts; preserve actual failures, cadence and latency. No numerical latency SLA was adopted.
Original dependency-ordered cleanup and preservation: reconcile original/new owned fixtures, policies/roles/auxiliaries, notification sinks/receiver artifacts and saved update intent. Preserve historical runs, terminal/request/execution audits and unrelated owner state. The owner performs infrastructure disposal and restoration; testing records actual postconditions. Earlier bounded cleanup does not discharge unresolved original obligations.

Keep #288/#304/#305/#306/#308/#310 open for their remaining criteria, and #282 open with this parent. Autofill remains owner-deferred and unexecuted. Continue independent admitted work while concrete owner prerequisites remain outstanding; record blocked/partial steps honestly. Final readiness requires actual dispositions and evidence for the complete retained scope. Private identities, deployment topology, credentials and raw evidence remain outside public threads.

Historical release delivery: beta.30

Beta.30, source 139a25f0bfcf237b9313282e46eaedbd2eabfe04 through #335, completed distribution verification at 2026-10-19 10:19:34 UTC. #333 and #334 were closed as delivered repairs, as recorded in the earlier delivery comment. Preserve that immutable release and closure history. Beta.31 adds the final-audit corrections and becomes the next owner test candidate; none of these release receipts supplies new owner acceptance.

Retained history: beta.29 first checkpoint and earlier owner campaigns

The prior issue body below is retained verbatim as historical evidence. Its older counts, candidate versions and prerequisite snapshots do not override the current final-audit scope above.

Latest owner checkpoint: beta.29 tested; continue on beta.30

Beta.29 at 4b148ebe824b3de445a12d30e95b8f5c30d53655 was deployed and its four server-component digests matched publication. An independent read-only migration check confirmed the expected migrations. Retained key-ring volume continuity was observed; key-file/backup equivalence was not established.

The owner supplied 33/35 fresh bounded functional assertions, with both failures attributable to DEF-B29-01 (#333). Historical remediation (12/12), metadata-only projections (2/2), access-control assertions (7/7) and new-fixture cleanup (3/3) are separate evidence populations.

All 518 prior context IDs and history remain intact. Invocation coverage is 110/518. Current dispositions: 104 bounded passes, 2 failed contexts, 4 exact admission denials, 391 Untested, 14 Prohibited, 2 Unsupported and 1 Deferred. Prohibited contexts remained uninvoked. Native prompt workflows completed: 0/8. Qualifying endurance: 0/240 calls and 0/3600 seconds.

Keep this acceptance parent open. Beta.30 must repair exact definition replay and retain the following order: owner verifies deployment and fixed isolated executors; completes original QA recovery and selected-consumer identity; exact legitimate authority review where required; remaining fresh functional/security/prompt cases; the isolated exact hour; original dependency-ordered cleanup. A separate RatelDesk campaign is not this campaign's acceptance evidence.

#315–#318 stay delivered. #321 has current bounded-browse owner proof. #288/#304/#305/#306/#308/#310 retain their remaining criteria. Autofill remains owner-deferred.

Retained previous checkpoint and acceptance criteria

Latest owner checkpoint: beta.27 results accepted; continue on beta.28

The checkpoint provides valid bounded owner results and does not assert a newly confirmed product defect. The new source follow-up is #321's explicit bounded-browse metadata and #322's prerequisite guides, delivered through #323. This owner acceptance issue remains open / NO-GO.

Beta.28 distribution complete

v0.1.1-beta.28 is published and independently verified at source 5079f000fb577fc55925c2701c22fb7221fd7821, through merged #323. Release build passed 8/8 jobs and publication passed 7/7 jobs, both first attempt. Independent checks matched all 15 public asset hashes/sizes, 14 checksum entries, 13 publication bindings, three native package identities and all five immutable image references.

The publication record is complete for build-distribution-integrity; functional acceptance is pending-owner-testing with no executed owner smokes claimed. Use the release's campaign guide for continuation. This does not add beta.28 owner passes or close the production-acceptance criteria below.

Completed and credited

[beta.25][B25-01][P0] Apply reviewed Production MCP admission and preserve exact intended expiry #301 minimal Production reconciliation/admission is closed on the combined owner evidence: distinct authenticated reviewer, applied policy state, preserved primary grants and exact expiry, both native tenant reads, negative controls and idempotent effective result.
[beta.27][P1] Protect secret-designated script and task output before persistence and ordinary reads #315 historical output remediation passed in six ordinary UI views, ordinary API projections and same-owner no-reveal MCP task/log reads.
[beta.27][P1] Admit job-start script validation under the exact delegated execution contract #316 exact Execute-only job completed; confirmation replay returned the same run. A multistep run preserved reviewed content after a live definition edit; a stale new start was refused.
[beta.27][P1] Bind policy mutations to trusted target classification and profile revision #317 intended classification-constrained create/replace/disable/revoke and persisted-class spoof rejection passed.
The supported no-reveal auxiliary attenuation was applied without changing its other grants or original expiry. Temporary review roles were removed.
Canonical file values/bytes and actual link-boundary negatives passed within their recorded contexts. New beta.27 definitions were removed and owned receiver incidents were resolved.

Exact remaining order

Verify beta.28 public distribution and actual deployed component image digests, source/version, migration receipt and durable Data Protection key ring/backup.
Preserve/recover the original unavailable QA runtime and its historical data. Use registered fixed-release disposable Linux and independently enrolled Windows executors. A second Windows executable directory does not isolate canonical credentials.
Complete supported selected-consumer reconnect; prove loaded bridge process/version/assembly hashes, current operation definitions and actual caller/both tenant reads. Beta.27 preparation did not replace the active beta.26 bridge; protected executors remained beta.25.
Revalidate exact profiles, authority, revisions and fixture cleanup entries. Finish [beta.30][P1] Finish admitted MCP functional, runtime and policy-race acceptance #305/[beta.30][P1] Finish fresh fixed-native secret-reader and untrusted-context acceptance #306: fresh standalone/job secret matrix and adjacent values, supported stream/error/export surfaces, profile/tag/revision/race cases, fixed-native Windows service timing, remaining injection/monitoring/flow/event/delivery cases.
Complete [beta.30][P2] Finish native prompt workflows and value/paging acceptance; autofill deferred #308's eight actual native prompt workflows with required explicit human decisions and [beta.30] Complete selected native consumer identity, metadata and progress acceptance #288's actual progress/in-flight cancellation. A tools-only host, prompt retrieval or Inspector rendering cannot substitute for prompt invocation.
Validate the new bounded-browse response contract under [beta.30][P2] Finish native prompt workflows and value/paging acceptance; autofill deferred #308. Continuation remains unsupported for browse; test actual paginated operations against their own supported contracts. Do not turn bounded results into pagination passes.
Only after all other functional/review/alternate-identity traffic ends, run [beta.30][P2] Run the exact isolated 240-call native HTTP MCP acceptance hour #310: 240 selected-native calls at offsets 0..3585, 15-second cadence, observation through 3600 seconds, concurrency one, four-operation sequence repeated 60 times. No replacement retries, interleaving or catch-up bursts.
Complete original dependency-ordered cleanup, effective auxiliary revocation, fixture/account/role/policy/sink reconciliation and update-intent restoration. Reconcile any uncertain background receiver artifacts and retain terminal audit.

Evidence boundaries

All 518 IDs are preserved: 438 original, 60 previously appended and 20 new beta.27 contexts. Current dispositions include 66 bounded passes, 426 untested, 14 prohibited and individually recorded partial/dependency/unsupported/deferred states. These are not a valid combined pass-rate denominator.

The unavailable QA start preview returned capability-unavailable without dispatch; it is neither an additional start-admission pass nor a new source defect. Fresh output and service acceptance cannot be attributed to a fixed native version that was not executing. Actual deployed digests/migration/key-ring verification remains distinct from published metadata. Separate authenticated review is evidenced; independent human judgment is not asserted.

Autofill remains owner-deferred and untested. Preserve private bindings/raw captures outside GitHub and retain every earlier evidence row unchanged.

Historical beta.25 acceptance plan (retained)

Current acceptance iteration: final beta.24 evidence to beta.25

The final corrected campaign dated 18 October 2026 establishes missing persisted Production MCP operator policies for an otherwise configured local credential. Both original tenant flow reads fail before business dispatch. The supported independently reviewed repair works in isolated Production testing; no live repair, positive tenant business execution or admitted endurance hour was achieved. This issue remains the owner acceptance parent under #282.

Corrected baseline and retained evidence

Baseline source/main is 623a75a, released as v0.1.1-beta.24; catalog 2026-10-17.2 exposes 30 tools, 231 operations, 6 resources and 8 prompts. Record any beta.25 catalog additions explicitly.

Native ledger: 533 calls = 307 Passed / 223 Blocked / 3 invalid tester inputs across 114 distinct operations; 239 schema attempts are included, with 231 selected schemas verified. Context ledger: 438 rows = 24 direct diagnostic/control successes, 171 direct denials, 229 uninvoked dependency links and 14 prohibited contexts. The 24 Passed / 400 Blocked / 14 Prohibited matrix is not a reproduced-defect count. Keep diagnostic/source/browser/transport/native workflow evidence separate.

Protected Windows preparation and authenticated diagnostic runtime now work with a complete safe path selection. Actual selected-consumer switch/reconnect remains open in #288. Supplied owner health receipts complete #289; retained UUID, schema and host-identity successes remain bounded positive evidence. Historical disclosure/autofill source fixes are not reopened without current reproduction.

Linked execution work

Order	Work	Required evidence
1	#301 / B25-01	Real independent review, authoritative applied receipt, exact original expiry, exact Production admission and native flow reads in both protected tenants
2	#288 / B25-04	One supported selected-consumer switch/reconnect, actual loaded catalog/caller and admission; finish runtime criteria with #305 evidence
3	#304 / B25-05	Registered disposable targets, current server-owned profile revisions, reviewed scoped auxiliaries, bounded roots, inert service and verified owned sinks
4	#305 / B25-06	Valid admitted positive and negative confirmation/binding/idempotency/readback/runtime/monitoring/flow/delivery/lifecycle cases
5	#306 / B25-07	Fresh synthetic designated secret, all required reader surfaces/identities and product-side untrusted-context controls
6	#308 / B25-09	Eight actual native prompt workflows, synthetic saved-dummy autofill, false/zero/Unicode/ID round trips, populated paging under controlled change
7	#310 / B25-11	Isolated admitted 60-minute native workload, exact 240-call order/cadence/concurrency and separate latency distributions
8	#304, #310	Owned cleanup, effective revocation, session/schedule closure and protected-state preservation

Source work #302, #303, #307 and #309 improves the next release and its reproducibility. Those issues and the beta.25 publication can finish independently of live owner admission. Do not close this campaign because the source work or artifact integrity is complete.

Acceptance discipline

Preserve all 438 context IDs, 21 retained NR/NR23 finding IDs, 14 gap IDs, G01–G12 and earlier case IDs. Add fresh executions without editing historical raw evidence or inventing assessed scores.
Keep both original tenants and all eight existing live agents protected. Historical jobs/runs remain inspect-only. Mutation, commands, files, terminal, service/lifecycle and onboarding tests require separately registered disposable targets.
Reinspect exact current identities, canonical values, original expiry intent and durable request state before review. The staged minimal reconciliation was never submitted and covers only two flow-read targets; it does not authorize the full suite.
Use legitimate independent review, exact operation/target/profile bindings and the supported 32-target proposal limit. Do not use an administrative API helper, direct database grants, fabricated reviewers or broader delegation.
A missing fixture or admission is an unexecuted prerequisite; a denied call is only evidence for the boundary actually invoked. Prove positive effects with authoritative readback and reconcile uncertainty before retry.
Native prompt retrieval, official launcher diagnostics, synthetic bridge controls and source protocol tests cannot substitute for actual selected-consumer workflows, streaming/cancellation or business reads.
Finish other diagnostics/fixture/review/auxiliary phases before [beta.30][P2] Run the exact isolated 240-call native HTTP MCP acceptance hour #310. No parallel traffic, catch-up bursts, substituted denial hour or invented 1700/1900 ms SLA.
Report release integrity, configured/effective access, native integration, functional/security coverage and endurance as separate states. Keep this issue and [beta.30] Epic: HTTP MCP readiness follow-up, verified release and owner campaign #282 open for evidenced owner disposition.

The authoritative private archive begins with PLANNER-HANDOFF.txt, root-cause.json and campaign-summary.json; root-cause evidence refines corrected evidence, which supersedes retained-first-pass assertions. Public follow-ups contain sanitized reproductions and source references only.

Historical beta.24 and earlier acceptance record

Priority: P1
Classification: Acceptance prerequisite / coverage gap
Campaign finding: NR-H1 and acceptance gaps G01–G12

Parent: #282

Purpose and baseline

Make the next native HTTP MCP campaign executable after the corrective issues, with owned fixtures, reviewed access and reproducible evidence. Extend existing optional acceptance workflows, fixtures and canonical MCP guides; do not introduce a parallel broad test framework or turn the hour-long campaign into required PR/main CI.

The assessed beta.22 catalog contained 30 tools, 223 operations, 6 resources and 8 prompts. All tool families were invoked, but only 68 distinct operations were invoked natively. The 430 operation/context rows were 3 Passed, 5 Failed, 408 Blocked and 14 Prohibited. Four core positive paths worked; the auth aggregate failed a separate schema control. Discovery, denials, previews and positive business execution must remain distinct.

Local evidence: operation-coverage.json, case-results.json, coverage-summary.json, ai-quality-scores.json, acceptance-gaps.json, endurance-summary.json, fixture-register.json, checkpoint.json. Keep deployment-specific files and original raw evidence private. Public examples use Tenant A/Tenant B, https://netratel.example.test/mcp, and a synthetic exact string ID such as "638000000000000321".

Prerequisites and controlled fixtures

First verify effective native read access independently in both existing tenants and resolve each exact target through supported discovery. An authenticating credential with configured grants is insufficient. Never substitute one tenant's agent identifier for another. Reviewed bootstrap, restricted/revoked/expired auxiliary identities and exact UTC expiry must use the supported MCP lifecycle from #283 / #286, with accurate failure mapping from #284. Do not request an administrative API token or reactivate the abandoned helper. NR-005 remains historical diagnostic evidence only.

Register every fixture's ownership, tenant, exact ID, permitted operations, current revision, expiry and cleanup state. Reuse designated disposable agents/environments, a bounded working directory, an inert service, disabled schedules, synthetic jobs/flows/policies and controlled notification/connectivity destinations. Confirm a destination is a test sink before delivery. Keep fixtures separated from existing live assets. Preserve the retained primary and existing synthetic run; adjust retained expiry only through the exact reviewed decision, without silently extending it or re-executing historical work.

Gap-complete matrix

Gap	Required completion evidence
G01	Reviewed setup receipt binds identity, scopes, roles, both tenants, exact targets and UTC expiry; effective native admission succeeds without self-escalation.
G02	Successful tenant/client reads, identity resolution, freshness, bounded pagination and restricted-identity isolation; explicit denied controls must follow working positive controls.
G03	Owned policy, disposable-tenant and object lifecycles: preview, exact confirmation binding/reuse/expiry, stale revisions, idempotency, read-after-write, cancellation and reconciled uncertain outcomes.
G04	Benign execution only on designated disposable targets; bounded files, hashes, traversal rejection, terminal cancellation and an explicitly designated inert service.
G05	Owned monitoring/workflow/event/notification lifecycles and controlled sinks; real AI approval decisions are supplied explicitly by the human operator.
G06	Synthetic secret redaction across authorized/denied UI/MCP readers, errors, resources, logs, exports and relevant persistence; #287 owns the product correction.
G07	Untrusted file/log/script/AI content cannot expand authority, alter approved targets or bypass confirmation; retain actual server enforcement results, not merely an AI refusal.
G08	Admitted tenant endurance and pagination while owned fixtures change; preserve the original paced baseline as a separately reported phase.
G09	Reuse native saved-dummy autofill acceptance from #279/#280 and collect deployment-campaign positive-control/form evidence; do not create a duplicate autofill defect.
G10	Restricted, expired and revoked MCP credential sequences, exact approved expiry, inspectable revocation and immediate effective-access behavior, without raw tokens in receipts.
G11	Record source/version/catalog, all deployed component identities and baseline configuration revisions; distinguish registry manifest digests from local image configuration hashes.
G12	Retrieve prompts and complete supported native prompt workflows; observe real streaming and in-flight cancellation independently of loopback simulations from #288.

Schema, health and prompt corrections are dependencies: #289, #292, #290, #293, #291. Supported local connectivity from #285 must use controlled sinks and separately audited probe execution.

Same-parameters endurance and evidence

Run the original 60-minute phase at one call every 15 seconds, maximum concurrency one: Tenant A flows/list, server auth/status, Tenant B flows/list, server system/version; repeat for 240 calls. Do not attach fabricated tenant arguments to server operations. The earlier phase had 120 core successes and 120 tenant denials, 60 per tenant, with zero transport exceptions; it did not establish tenant reliability.

Start the comparable admitted phase only after both tenant reads succeed. Require all expected tenant reads to complete successfully and report successes, denials, transport failures, cadence, versions and latency separately by workload/tenant. Preserve the proposed core-read p95 of 1700 ms as an explicitly optional pilot objective, not an accepted production SLA. Measure admitted tenant latency independently. Keep additional lifecycle, pagination-under-change and transport-fault phases separately identified.

Rediscover the new catalog and compare against the 30/223/6/8 baseline; added lifecycle operations require new matrix rows, not a hard-coded old count. Preserve original case IDs and evidence provenance. Every operation/context needs Passed, Failed, Blocked or Prohibited with its observable postcondition and evidence. Re-score the seven existing AI-tooling dimensions from actual new evidence; null remains unassessed, and no score may imply successful blocked backend execution. Diagnostic prompt retrieval cannot be relabeled a native workflow; simulation cannot be relabeled a live outage.

Execution and completion boundaries

Never delete, update, disable, reboot or otherwise disrupt live agents or alter the two named production tenants. Exercise lifecycle paths on separately designated disposable fixtures; protected rows remain Prohibited. Reconcile actual state before retrying an uncertain mutation; do not automatically replay it. Do not fabricate human AI-approval responses.

Use existing integration-validation.yml selections and JobsAutofillAcceptanceTests. #280 records a persistent-profile native Chrome control and hosted test evidence; this campaign did not repeat that deployed check. Keep those evidence levels separate. Link existing #281 for image-scanner coverage and release integrity rather than opening a duplicate.

Completion requires a fixture/cleanup receipt, updated coverage and gap matrix, exact release/source identity and a runnable owner retest guide. Required fast CI remains under 20 minutes. A published beta can be distribution-verified with owner campaign acceptance pending; do not claim production acceptance or close unexecuted acceptance gaps.

Final beta.23 campaign reconciliation (authoritative over earlier draft)

Assessed source 86babb9d94a65c10d5b88e391ea1cba73c5c1fea; correct beta.23 component identities verified. Final catalog: 30 tools, 229 operations, 436 operation/context rows: 17 Passed, 2 Failed, 403 Blocked, 14 Prohibited. Native invocation cases are a different denominator: 382 cases (178 Passed, 194 Blocked, 3 Failed, 7 invalid tester inputs), with 65 distinct operations invoked. Full schemas and new access operations exist.

Four named prior regressions passed their current controls: closed envelopes (#290), string IDs/empty search (#291), prompt argument failures (#292), and prompt discovery guidance (#293). Retained ordinary run UI now redacts secret material (#287 partial owner evidence); fresh execution, alternate/unauthorized readers, MCP/log/export/persistence matrix remains blocked. These gaps do not reopen every source defect without a reproduced failure.

Reviewed request/receipt/cancel/resume and exact six-decimal proposal validation work, but actual approval/application and tenant admission remain unmet. Review requests ended cancelled/expired with no applied grants. A filename containing “approved receipt” still records pending state. Beta.23 already includes protected access prepare/resume and profile workflows; NR23-006 is a native campaign migration/authority prerequisite, not evidence of wholly absent setup code. #296/#297 correct UUID and approval diagnostics; #288/#298 supply supported migration/metadata paths. NR23-008 is tracked in #299.

The fresh hour completed 240 calls: 120 core successes, 120 tenant denials (60 each), zero tenant successes and zero transport exceptions. Core p95 948.35 ms; denied-tenant p95 967.55 ms. Cadence deviations reached 85.28 seconds and independent diagnostics occurred concurrently, so it is not an isolated strict-cadence SLA measurement. The original 1700 ms and subsequent 1900 ms pilot proposals are both unadopted.

Next campaign admission and retained state

Before the measured phase: verify both approved tenants' effective native reads, independently resolve every positive target, complete a real independent operator decision, inspect the exact applied UTC expiry, and prepare/reconnect supported protected primary/restricted/revoked/expired profiles. The later campaign's newly proposed expiry differs from the earlier retained target and was never applied. Preserve the original owner-intended UTC instant as the desired value unless the operator explicitly chooses another; it was not an applied or approved receipt. A real independent operator decision is required before application. Do not reset the access window from a new campaign start.

The eight existing live clients remain protected. The latest baseline contains mixed versions, so do not state all eight are beta.19. No current positive agent ID was verified in this campaign; historical run references are not current execution targets. No business mutation, issued auxiliary credential, active campaign session or external delivery was completed. Full unrelated-object equality is not proven by denied inventory. Keep existing synthetic history inspectable without replay.

Repeat the original ordered 240-call hour at 15-second intervals, concurrency one, after admission. Isolate concurrent diagnostics from that measured phase; report actual cadence and all deviations. Expected admitted mix is 120 core successes and 60 successful reads in each tenant. Expand the matrix from the newly advertised catalog rather than hard-coding 229 or 436. Run all existing blocked safety/confirmation/revision/idempotency/isolation/redaction/injection/pagination/fixture/cleanup controls with owned inert objects. Actual human AI-approval decisions and the controlled native autofill positive control remain owner actions.

Keep #282 and this acceptance issue open until owner evidence gives every required context an explicit disposition. Source completion, distributed beta integrity, native migration and production acceptance are separately reported.

Beta.24 candidate ready for owner deployment and retest

The correction is merged in #300 at 623a75a7807cf9ac305a99740ad359df609ee5bb, catalog 2026-10-17.2. Source follow-ups #296–#299 are closed; #288 and #289 retain their actual owner/native verification gates.

Release v0.1.1-beta.24 points to the exact merged SHA. PR validation, merged-main validation, and tag validation/build each passed 4,714 executed tests with zero failures/skips; the respective .NET/source-validation jobs took 16m34s, 11m08s, and 16m55s. All eight tag-build jobs succeeded.

All seven publication jobs succeeded, including final distribution verification. All 15 public release files were independently downloaded and matched their GitHub SHA-256 digests; all 14 SHA256SUMS entries and all 13 publication artifact bindings matched. Five immutable image references are recorded in publication.json, SHA-256 f9cc6fb716e9ea833bbc26f2e3df8a3416e5b9ada5f2ef59134bbf42cc80bdd5. The hosted completion gate verifies anonymous pulls, source/version metadata, inventory and credential-disclosure scans before marking build/distribution integrity complete; this is not a CVE audit. Functional acceptance remains explicitly pending-owner-testing.

The tagged Windows archive passed 10 synthetic checks and 68 file-digest checks. The separate reviewed-tree Linux integration passed 2 actual HTTP protocol cases (independent reviewed admission in two synthetic tenants and configured gateway health) plus 16 bridge controls. These do not establish owner deployment, primary approval, applied exact expiry, consumer reconnect or successful owner-tenant endurance.

Proceed after owner rollout with the exact-source campaign guide and the private handoff: verify component/source/image identities; prepare and reconnect the supported protected client; obtain an actual independent operator review of the exact intended proposal/UTC instant; verify effective reads in both original tenants. Then provision reviewed restricted/revoked/expired auxiliary identities and registered inert fixtures, complete G01–G12 and all retained gap contexts, and run the isolated 240-call / 15-second / concurrency-one hour. Keep diagnostics and fault simulations outside that measured phase.

The retained primary, live agents, original tenants and synthetic history stay protected. No administrative API-token workaround or fabricated human decision. Preserve original expiry intent without labelling an unapproved proposal as approved. This issue and #282 remain open for the owner campaign's evidenced disposition.

The enabling source changes and owner retest guide are merged in #295 as 86babb9d94a6. Required PR source CI passed 4,641/4,641 with no failures/skips (.NET 13m48s). Optional disposable protocol/bridge checks passed 2 + 16 cases, and native Chromium saved-account acceptance passed one disposable case. The refreshed source catalog is 30/229/6/8; retain the original baseline and add rows for the six new access operations.

Keeping this acceptance issue open. After verified distribution and owner deployment: record installed transport/provenance and component manifest digests; preserve the retained credential's previously approved exact UTC expiry; prove effective native reads independently in both original owner tenants; register designated fixtures/sinks/cleanup and execute G01–G12 with real human AI decisions where required. Then run the unchanged admitted 60-minute sequence: 15-second cadence, concurrency one, 240 calls, with separate tenant/core outcomes and latency. Native owner-host workflows, live streaming/cancellation, controlled delivery, retained-record/browser checks and cleanup remain unexecuted owner gates. Protected rows remain Prohibited.

Merged-main verification: Run 39044037939 passed all 4,641 tests with zero failures/skips: 17m00s overall, .NET 16m52s; under 20 minutes, above the 10–15-minute target.

Release/build/publication and independent distribution verification: beta.23; tag build and publication succeeded. Independent verification passed all 13 public downloads, 12 checksum entries, 11 original attestations, the exact two-file Compose transformation, four downloadable SBOMs, three native Client archives, all five immutable image manifests/anonymous pulls/nonempty actual-layer scans and embedded runtime SBOM file checks. Reviewed exact upstream benign matches are preserved. publication.json binds source, files and images. Owner acceptance remains pending.

Final beta.29 audit checkpoint supersedes the earlier first-checkpoint counts. Tested public source: 4b148ebe824b3de445a12d30e95b8f5c30d53655. Private package integrity was verified; preserve all 518 context IDs and histories. The final populations are 268 bounded assertions; 135/161 functional passes with 26 failed assertions across five P2 replay defects; 811 raw native receipts; 283 invoked contexts. Current dispositions: 225 bounded operation passes, 3 catalog/schema-only, 218 untested, 28 admission denied, 25 failed, 14 prohibited (all uninvoked), 2 negative-enforcement-only, 2 unsupported and 1 deferred. Twenty policy receipt mappings across nine operation contexts are corrected to their nested/persisted tenant scope; there is no positive control-plane mutation claim.

#333 now covers eight definition mutation operations in both tested tenants (16 contexts). Four additional final-audit replay defects are tracked as #336 (script delete), #337 (policy lifecycle), #338 (artifact cleanup) and #339 (deleted-job local delegation). #334 remains the separate source-review run-action regression. Beta.30 was already tagged from the first-checkpoint source; retain its immutable history. Beta.31 is the next corrective release integrating all final-audit findings.

This is a source/release continuation, not acceptance closure. #282/#288/#304/#305/#306/#308/#310 remain open for their actual criteria: fixed native and selected-consumer identity, exact authorized admission, profile/secret/service/live monitoring cases, eight actual native prompt workflows (0/8), native progress/in-flight cancellation, the unchanged isolated 240-call/3600-second endurance campaign (0/240; 0/3600), and original dependency-ordered cleanup. Deferred autofill remains unexecuted. Keep private identities, deployment details and raw evidence out of public threads.

Beta.30 is published from merged #335 (139a25f0bfcf237b9313282e46eaedbd2eabfe04). Release and public distribution verification completed at 2026-10-19 10:19:34 UTC; #333 and #334 are closed as delivered repairs.

Carry fresh #333 definition replay and #334 cancellation/deletion replay retests into beta.30 using currently authorized owned fixtures and valid confirmations. Preserve retained owner accounting until actual campaign receipts establish new outcomes.

The remaining order is unchanged: verified owner deployment/recovery; fixed isolated executors and selected-consumer identity; legitimate current authority; remaining functional/security cases and eight actual native workflows; the isolated 240-call/3,600-second hour; dependency-ordered cleanup. Autofill stays deferred and protected contexts stay uninvoked. This acceptance parent remains open.

Beta.31, source f89593ccef6fa7385779baf1ba18328426218c07 through merged #340, is the corrective delivery for the final beta.29 audit. It adds #336 script deletion, #337 policy lifecycle, #338 artifact cleanup and #339 receipt-bound deleted-job/run delegation, while retaining #333 definition replay and the separate source-found #334 run-action correction. #333/#334 stay closed for their earlier beta.30 delivery; fresh acceptance carries forward to beta.31.

Corrected PR validation and merged-main validation each passed 4,973 tests (4,352 Core + 36 API + 585 Components), with zero failures/skips; complete runs took 11m54s and 16m57s. These are separate executions of the same suites. The initial 13-failure run remains historical failure evidence; existing relay assertions were preserved in the correction.

Historical tag/publication evidence: tag build attempt 1 failed one Core case (4,351 passed, one failed, zero skipped); API and Components did not execute. The recorded mechanism review supports a pre-existing TestServer/gRPC request-flush race, and the failed execution remains retained. Publication attempt 1 stopped on that prerequisite before registry writes or public asset uploads; it produced no publication journal or release asset. The one documented failed-jobs recovery on the unchanged tagged source completed successfully as build attempt 2, followed by successful publisher attempt 2. No source, test, assertion or workflow gate was changed for this recovery. The 168 passing release contract tests from build attempt 1 remain retained successful evidence.

Tag build: attempt 2 succeeded with all eight required jobs; fresh source validation passed 4,973 tests (4,352 Core + 36 API + 585 Components), zero failed/skipped, in 17m34s. All four unsafe-frame cases passed the original assertions; Windows MCP passed 15 bounded synthetic checks; three native Client archive integrity checks and all five package provenance-upload steps passed. The 168 contract tests are the retained successful attempt 1 execution. Publication: attempt 2 succeeded with all seven required jobs in 7m29s; five anonymous digest pulls/rescans, all 15 public assets and 14 public-download checksum matches passed. Independent distribution verification: 2026-10-19T12:37:08.811855+00:00; 15 completed-publication files matched public asset metadata, with 14 checksum entries, 13 artifact hashes, exact source and authenticated build attempt 2, all five Compose image pins, and the published Windows MCP ZIP matching its 15-check smoke receipt. Public downloads and anonymous image pulls were verified by the publication workflow; the independent local check used the completed artifact and public digest metadata. Functional acceptance remains pending owner testing. publication.json and SHA256SUMS bind the delivered artifacts.

Keep #294 and #282 open, together with #288/#304/#305/#306/#308/#310 for their actual remaining criteria. The final beta.29 checkpoint remains the owner baseline: 268 bounded assertions, 135/161 functional passes, 26 failed assertions across five P2 families, 811 native receipts and 283/518 invoked contexts. Preserve all 518 IDs/history, the 20 policy receipt mappings across nine operation contexts and 14 uninvoked Prohibited contexts. Source/release checks add no beta.31 owner passes.

Next: owner deployment/recovery, fixed isolated executors and selected-consumer/schema proof; legitimate current admission with original authority/expiry preserved; all five fresh replay families plus #334 and remaining native functional/security/profile/service/monitoring/synthetic-delivery cases; eight genuine prompt workflows with actual human decisions and live progress/in-flight cancellation; the unchanged isolated 240-call/3,600-second hour; original dependency-ordered cleanup and receiver reconciliation. Prompt workflows remain 0/8 and qualifying endurance 0/240 and 0/3600 until actual new evidence exists. Autofill remains deferred. Production acceptance remains NO-GO pending its required owner evidence.

Fix acceptance regressions and prepare beta.32 (#342)

Beta.38 owner campaign checkpoint (2026-10-20 22:09 UTC): published/deployed API and MCP host report 0.1.1-beta.38+aa0c01a69dfa57551e393b9eab7b441927bd1615; owner-direct tenant 1/2 connector reads passed and two fail-closed negative controls passed. The disposable no-unzip Linux canary's first beta.37→beta.38 attempt failed on its installed helper (unzip absent, exit 127); hash-checked helper repair on that canary and a fresh offer then yielded beta.38 Accepted and online. Temporary tenant 9 workflow credential was revoked; owned job 24/script 25 cleanup remains blocked by exact policy admission. Eight current first-prerequisite B1 calls remain denied, including connectivity's missing mcp.connectivity.read. The original Ubuntu-26.04 QA WSL distro now starts, but its historical QA container and volume are absent from the current default Docker daemon, so QA runtime/backup recovery is still unproved. No disposable Windows VM is available locally (11.5 GiB free, no Hyper-V tooling/elevation); the live CCTV VM was untouched. Conservative ledger: 198/518 original contexts unresolved; 0/240 and 0/3600s endurance. Production readiness remains NO-GO, with original-tenant update holds intact. Full local receipt: outputs/netratel-beta32/handoff-beta38.md.

Beta.40 owner campaign checkpoint: NO-GO, with two previously open disposable-client gates now passed.

Published beta.40 is deployed; exact source/version and public assets/images were independently verified, and live API/MCP health is healthy.
A clean disposable Ubuntu 24.04 client automatically accepted beta.39 to beta.40; its installed executable matches the verified public archive.
A disposable Windows client automatically accepted beta.39 to beta.40 from the canonical installation after preserving a stale Accepted result left by an earlier manual beta.39 fixture repair. The server and local attempt are Accepted; the running executable matches the public archive. No beta.40 updater was invoked manually. This qualification remains part of the result.
Fresh Windows Application Event 902 / record 236 appeared in the Web Live log stream, closing the published retest for Fix Windows Event Log live subscriptions #369.
Protected original tenants retained update holds and the live CCTV VM was untouched.

The append-only private campaign record still has 518 original context IDs and 186 unresolved required contexts. Synthetic receiver delivery, original QA/backup recovery, dependent workflow isolation, remaining contexts, owned artifact cleanup, the isolated 240-call/3600-second endurance phase, and complete owner-direct gate removal acceptance under #356 remain open. Eight prompt workflows passed earlier. No production-readiness pass is claimed.

Follow-up to the beta.40 checkpoint above: four more original contexts passed after that comment. Exact owner-direct admission preserved the main credential's 64 tenant grants, 30 instance permissions and original expiry. netratel_clients/preview_ping and confirmed ping succeeded for both original tenant clients; the tenant 2 CCTV VM received only one ping, with no configuration or update-policy change. The original ledger now has 182 unresolved required contexts of 518.

Both preview_software_update calls remained denied. In the published route, that preview also requires the software_update action policy. I deliberately withheld that action from protected original clients and retained their update holds. The campaign verdict remains NO-GO. Private evidence and the append-only beta.40 ledger record these bounded outcomes.

Follow-up to the beta.40 NO-GO checkpoint: the published Discord receiver gate now passed. In disposable tenant 9, an exact-agent monitoring rule raised clients:offline for canary 94844233-78c3-42bf-ac66-872871f6de24; immutable flow version 767005a5-8610-4bfb-b835-1aaa119fa013 run 2245bb23-5621-490b-8016-85f1b236e882 completed its single Discord action with notification-delivered in one attempt. Its delivery reference 1562344403908042794 was fetched from the user-owned receiver by exact ID; webhook identity and synthetic rule/client content matched, then the test message was deleted. The canary service is active/Online again. The test rule was removed, and the flow and connector are disabled. The connector credential remains encrypted in its disabled record; the immutable flow history remains for audit. Local evidence: outputs/netratel-beta32/evidence/beta40-discord-published-flow-campaign.json and beta40-discord-published-flow-readback.json. Overall verdict remains NO-GO, 72 original contexts unresolved; QA recovery, AI workflow, accepted endurance, owned artifact cleanup, #356 acceptance and other recorded defects remain open. Slack remains deferred by the user.

Beta.40 campaign checkpoint (2026-10-21): published Discord monitoring-flow delivery and exact receiver readback passed on disposable tenant 9; flow/target disabled, temporary rule removed, client restored Online. The ledger remains 518 original contexts, 72 unresolved required, 17 exclusions; readiness is NO-GO. Fresh tenant 1 disposable CPU tracing established issue #375: an observation 35.4374 ms after gateway receipt was classified future_clock and then displayed stale. The trace rule was deleted with original configuration intact. Source corrections are open in #376 (large onboarding referral), #377 (target-profile replay), #378 (owned library script admission), #379 (Windows private temp ACL), and #380 (bounded client clock lead). These are not counted as production passes before CI, merge, deployment and exact retest. #356 reviewer-gate removal remains in the handoff. Other open gates: original QA runtime recovery, AI isolation, owned job/script cleanup, historical Windows delivery reconciliation, and a protocol-compliant isolated 240-call/3600-second endurance run. Protected original clients and their update holds remain unchanged.

Beta.40 continuation after the previous checkpoint: 518 original context IDs and order remain intact. Three appended B26 Linux contexts advanced from fresh native beta40 command, cancellation and terminal effect/readback evidence; two more appended B26 script/job contexts advanced from exact already recorded beta40 disposable receipts. The latest private ledger is context-ledger-beta40-b26-script-job-reuse.json: 67 required contexts unresolved, 17 planned exclusions, NO-GO. No protected original client changed. PR #377 passed validation and was merged; #376 and #378-#380 remain in validation. Source changes are not credited as live acceptance, and the remaining QA/AI/cleanup/reviewer/endurance gates remain open.

Beta.41 production campaign checkpoint (2026-10-21): Web/API/MCP HTTP run 0.1.1-beta.41+3dae073583b41f378ccc053b6c49b4aad9f404ce. The tagged GitHub release and all 15 public assets passed independent publication/hash verification. The beta.41 client pack is now imported, Published, and Enabled for win-x64/linux-x64/osx-arm64 in the prerelease catalog.

Fresh live retests passed: version-pinned large onboarding collateral referrals and exact 42,859,780-byte downloads on both original tenant scopes, with zero-use temporary codes revoked; exact target-profile confirmation replay returned persisted revision with replay=true on an isolated tenant 1 agent; caller-owned netratel_tasks/run_library_script completed with exit 0 and exact stdout logs on isolated tenant 1 and 2 Linux agents, followed by script deletion/absence; and beta.41 CPU monitoring accepted 31–33 ms client clock leads, raised client-specific occurrences in both tenants, and completed MCP acknowledge/clear. Temporary monitoring rules were removed; original rules and zero bypasses remained. The append-only ledger retains all 518 original context IDs and 17 planned exclusions; 59 required contexts remain unresolved. Verdict: NO-GO.

Still open: native beta.41 Windows canary/ACL correction retest after client auto-update, original tenant 3 QA runtime or backup recovery, dependent AI workflow isolation, remaining original contexts, supported cleanup of historical tenant 9 job 24/script 25 (#304), a protocol-compliant 240-call/3600-second endurance rerun, and protected original Windows Accepted delivery proof. Reviewer-gate removal #356 remains required in the next handoff. The protected original agents retain update holds. Evidence and handoff: outputs/netratel-beta32/handoff-beta40.md (beta.41 heading), readiness-extension-beta41-monitoring.json, and linked beta.41 receipts under outputs/netratel-beta32/evidence/.

beta.41 campaign checkpoint — NO-GO (21 Oct 2026)

Production Web/API/MCP and the published client catalog run beta.41. The append-only campaign ledger retains all 518 original contexts; 59 required contexts remain unresolved and 17 planned exclusions remain. Fresh beta.41 onboarding referral, exact target-profile replay, caller-owned library-script tasks, and client-specific monitoring clock/acknowledge/clear retests passed on isolated original-tenant agents. The disposable Windows private-temp safe-inherited ACL fix also passed through a real client-executed task.

The disposable Windows client naturally accepted beta.40 → beta.41 and its running executable matches the independently verified public package, but only after the preceding beta.40 Accepted result was preserved outside the active state directory. The new beta.41 Accepted result again has no delivery record; a read-only ACL trace identified legacy writable service SID and TrustedInstaller entries rejected by the current protected-file validator. Issue #312 carries the sanitized finding. A clean second automatic update and the historical original Windows result remain unproved.

The original QA runtime/backup, dependent AI isolation, remaining contexts, supported owned job/script cleanup, and a new protocol-compliant 240-call/3,600-second endurance run remain open. Reviewer-gate removal #356 is explicitly carried forward. Protected original clients, including the live CCTV VM, remain held at beta.31 with auto-updates disabled and no configuration change. Current private handoff: outputs/netratel-beta32/handoff-beta41.md.

Current completion register — beta.47 delivery, beta.46 owner credit

Owner acceptance remains OPEN / NO-GO. The supplied beta.46 owner checkpoint advanced 16 of the 53 previously retained original contexts. The unresolved original register is now 37 rows: 12 P0, 23 P1 and two P2. The 17 planned exclusions remain inside the original 518-context universe. These 37 rows are unresolved assertions and prerequisites, not 37 newly demonstrated product defects.

The older 53-row count and its source/release evidence below remain historical. Beta.47 source work and publication do not advance an original row without its actual acceptance receipt.

Credited beta.46 outcomes

Retain the exact attributed native AI connection/control/readback results, original durable readbacks, three original ordinary-reader redaction passes, both supported browse-continuation results, and the supported parameterized prompt-message workflow. The separate Linux beta.45→beta.46 automatic update also proves the repaired beta.45 coordinator's activation path: same enrolled identity, Accepted result, acknowledged delivery and matching running executable hash, without unzip. #361 now retains only its still-unproven legacy-helper obligation.

Real daemon smoke turns and a local-account Web turn remain useful observations within their original scope. They do not substitute for the missing exact conversation vectors, real uncertain-delivery predecessor or genuine SSO first turn. Disposable snapshot/command regressions and a Windows service Start/Stop regression do not silently replace their historical target, progress-association or original fault/fallback assertions.

Original unresolved register

Group	Priority	Remaining original rows
Original AI create/send/approval/reconcile/archive	P0	10
Original QA identity recovery and start preview	P0	2
Owned event retry/disable	P1	4
Protected output producers and reader matrix	P1	11
Policy/classification/attenuation	P1	3
Windows service fault branch and selected-native progress	P1	2
Job definition/input snapshot integrity	P1	2
Historical temporary authority restoration	P1	1
Isolated measured native hour	P2	1
Final owned cleanup/restoration	P2	1
Total		37

Beta.47 source delivery

Services refresh correlation: selected-native deployed acceptance passed #399: durable request-to-terminal Services correlation, current-session fencing, coalescing/throttling and explicit legacy attribution. The new capability requires an upgraded client and a fresh request; the old uncorrelated request stays unproven.
Bounded MCP policy Web UI: beta.50 refresh and focus acceptance pending #398: bounded administrator policy Web lifecycle with exact preview/confirmation, current admission inspection, authoritative version/audit readback and original replay receipts. Real browser acceptance remains due.
Windows updater: supported recovery and controlled automatic progression accepted #312: a supported checksum/attestation-bound Windows recovery artifact with guarded state-preserving cutover. The actual LocalSystem recovery and subsequent automatic beta.46→beta.47 chain remain due.
The prior callback (Durable callback rejection and explicit replay: deployed acceptance passed #394), paired authoring (Paired request-definition authoring and input sync: deployed acceptance passed #390), SSO authority (SSO investigation authority: genuine first turn and continuation accepted #382), Linux updater (Linux updater: legacy installed-helper repair and automatic progression accepted #361) and monitoring/flow/connector (Owned monitoring-to-RatelDesk two-window incident acceptance completed #141) corrections are retained with their own remaining live gates. The requested dependency changes are consolidated in the release PR.

Historical evidence and prospective successors

Keep original IDs, assertions, predecessors, attempts, exclusions and earlier credited passes unchanged. Where a historical runtime/backup, immutable caller-owner tuple, literal question or pre-correlation refresh is permanently unavailable, define the precise versioned successor before execution and link it to the original coverage anchor. Retain the unresolved original and record successor coverage separately. A new enrollment is not identity recovery; an empty reconciliation is not recovery of an uncertain delivery; a newer inventory is not proof of an old refresh. Do not rewrite ownership, revive expired/revoked credentials or invent missing inputs.

The next laptop handoff supplies explicit successor specifications only for these demonstrated unavailable prerequisites. The final verdict must report unresolved historical evidence and current-release coverage separately, without reducing the original denominator.

Remaining owner release gates

Verify the deployed matching image digests, migration and selected native consumer against the completed publication. Complete the original functional/security assertions and any required versioned successors, genuine SSO turn, paired companion create/sync/execution, callback terminal-rejection/replay/transient behavior, supported Windows recovery and automatic progression, legacy Linux-helper repair, and one real monitoring occurrence → retained incident with duplicate suppression and recovery/rearm.

Obtain a supported off-host backup and actual isolated restore/decryption continuity for the matching database, Data Protection and required application/native identity material. Storage inventory or a local copy is not restore proof; retain private key material and operational evidence outside this public tracker.

Run the original selected-native endurance sequence only after its functional/security prerequisites: 240 serial slots over 3,600 seconds, 15-second cadence, concurrency one and the declared four-operation rotation, with no interleaving, replacement slots or catch-up. Finish owned cleanup and authority restoration in dependency order, retaining audit/updater history and unrelated state.

Continue independently ready work while a precise blocker is resolved. Keep #294 and the issue-specific live acceptance trackers open until their evidence is present. Use the exact-source campaign guide; the owner laptop handoff is for live execution, not another source implementation/release assignment.

Delivered in the consolidated beta.47 change at source f587d8f050ac991e126f3d6ab9d3375191a220e6. Distribution evidence: beta.47 release, all-target release build, and completed publication. These references establish source/build/distribution delivery; live acceptance is recorded separately.

Beta.47 owner campaign observation on a hosted disposable tenant-2 agent (diagnostic native stdio consumer; selected Codex chat schema remains unloaded): a new AI investigation was persisted Idle with no active message. The exact T2/v1 message was admitted once, reached AwaitingApproval for nr-campaign-t2/netratel_flows, and offered approve_once. A planned netratel_ai/set_enabled(false) on that tenant's saved development connection changed revision 5→6 and moved the same investigation/message to DeliveryUnknown. The SAME connection was re-enabled at revision 7 and test_connection returned authenticated-signalr-verified. One confirmed netratel_ai/reconcile added recovery_requested and recovery_admission_confirmed for the original message; no message was resent. Three subsequent exact get reads still showed DeliveryUnknown, lastSequence 11, no pending approval and no turn_completed. The NetRatel read model exposes the pending tool name/call ID/options but not the call arguments, so I did not approve an uninspectable call. The connection is restored; the investigation is retained for diagnosis. This is partial recovery evidence, not the completed AI vector or original row credit. The beta.47 worker source appears to move DeliveryUnknown back to AwaitingApproval only if the resumed session emits tool_interaction for the saved call ID; please check whether the daemon replays that event after session join, or expose a supported exact-call review/recovery path. No general defect attribution to either side is asserted yet.

Additional beta.47 Linux Services campaign finding: two independent hosted Debian 12 agent containers produced the same 14 systemctl list-units --all service names whose systemctl show LoadState is not-found; neither list-unit-files nor the batch query had a nonzero exit, and no state-property block was missing. A bounded systemctl daemon-reload on one disposable container left the same 14 names while its NetRatel client service stayed active. In beta.47 LinuxSystemdServiceInventoryCollector.CollectInventoryAsync unions installed and loaded names and calls QueryServicesAsync(..., allowMissing:false), which marks each discovered not-found name systemd-query-partial. This independently corroborates a stable host/template condition behind the previously observed Partial inventory, but is not that original request's per-unit trace and does not establish a product defect or a Complete result. Please adjudicate whether these list-units --all optional missing entries should count as incomplete inventory, or specify a Linux fixture expected to yield Complete. Original 16/53 owner credit is unchanged.

Follow-up to the Linux Services adjudication request: independent read-only replays on both hosted Debian fixtures excluded only LOAD=not-found rows from systemctl list-units --all discovery. Each replay retained 113 installed/loaded candidates, with three zero-exit systemctl show batches, no missing state blocks, and no remaining not-found candidate. Draft PR #403 proposes that narrow discovery filter while preserving installed unit-file candidates and exact watched-service lookups. Two focused local tests passed; Public PR validation run 40198809267 passed 5,192 tests with zero failures or skips in about 18 minutes. This is proposed source behavior, not a beta.47 deployment or a retroactive Complete result for the original Partial request. The campaign remains NO-GO.

Final beta.49 assessment — 27 October 2026

The original ledger is 21 PASS / 32 OPEN of 53, with zero new original advances and unchanged SHA256 38bc31ba1e410fc8e9ddff94e7d8e654dbe1f17427103dfe60a2d7547be0d370. The original 518-context universe and 17 planned exclusions remain intact. Earlier 37-open checkpoints are historical.

Current bounded successors include 15 complete passes, one retained partial protected-output matrix and two observed Web failures (#414, #415). The real measured hour completed 240/240 actual successful calls over 3,600 seconds, using the declared observable published live-bridge protocol consumer. Its selected native Codex preflight is separate; no opaque AI-host scheduling or original historical hour credit is inferred.

The supplied owner evidence closes the current bounded criteria of #399, #394, #390, #382, #361, #312 and #141 with version/partner limits recorded on each. #398 remains open for the two Web repairs. The 32 original immutable input/owner/runtime/unknown-delivery predicates stay OPEN. Missing QA ownership/predecessor diagnosis records are distinct from the Windows recovery receipts actually retained in the packet.

The beta.50 campaign is a finite targeted Web-delta acceptance with prior passes carried at their exact versions. No new full hour or recreated historical identity is required merely to increment the release. This issue remains open; a new current pass will not rewrite missing historical evidence.

Published fix-forward: v0.1.1-beta.50, source 8c5fb3b8169335af233c281a36a9e5afe9113d2c, merged through PR #416.

Main validation passed all 5,221 Core/API/component tests with zero failures or skips. Focused browser verification executed and passed all three cases. Tag build and publication completed successfully. The final release contains 15 uploaded assets and five public image digests; source/asset/checksum bindings, the downloaded Compose and Windows MCP archives, and anonymous image manifest/config labels were independently verified. This is build/distribution integrity; deployed owner acceptance remains pending.

Historical ledger remains 21 PASS / 32 OPEN of 53, SHA256 38bc31ba1e410fc8e9ddff94e7d8e654dbe1f17427103dfe60a2d7547be0d370; the 518-context universe and 17 planned exclusions are unchanged. New source/browser tests and the upcoming targeted beta.50 campaign do not supply absent immutable historical predicates.

The final beta.49 input archive was independently checked: ZIP CRC passed and all 1,379 manifest entries matched size and SHA256. Seven current feature issues (#399, #394, #390, #382, #361, #312, #141) were closed using that retained evidence and their exact version/consumer/partner boundaries. #414/#415 contain the new observed Web defects and remain open for beta.50 deployed acceptance through #398.

The next laptop prompt records every original missing predecessor once, carries prior passes at their actual identities, preserves unresolved DeliveryUnknown investigations and cleanup/rollout state, and ends after the bounded Web campaign. No beta.49 hour is relabelled as beta.50 or as the original historical hour. Keep #294 OPEN while the original predicates remain unmet.

[QA-CAMPAIGN] Reconcile 32 retained original HTTP MCP predicates and current release gates #294

Description

Current completion register — final beta.49 owner evidence

Historical completion register — beta.47 delivery, beta.46 owner credit

Credited beta.46 outcomes

Original unresolved register

Beta.47 source delivery

Historical evidence and prospective successors

Remaining owner release gates

Current completion register — beta.46 source, release and owner acceptance

Beta.46 source delivered and independently validated

Beta.46 release and distribution

Retained beta.45 evidence and credit

Exact remaining groups

Active development Netclaw assignment

Additional release and seven issue-specific obligations

Consolidated preparation and completion

Current candidate: beta.31; final beta.29 owner audit is the retained baseline

Final beta.29 evidence supersedes the earlier phase1 counts

Five replay families and the separate source-found regression

Required owner preparation and next campaign order

Historical release delivery: beta.30

Latest owner checkpoint: beta.29 tested; continue on beta.30

Retained previous checkpoint and acceptance criteria

Latest owner checkpoint: beta.27 results accepted; continue on beta.28

Beta.28 distribution complete

Completed and credited

Exact remaining order

Evidence boundaries

Historical beta.25 acceptance plan (retained)

Current acceptance iteration: final beta.24 evidence to beta.25

Corrected baseline and retained evidence

Linked execution work

Acceptance discipline

Historical beta.24 and earlier acceptance record

Purpose and baseline

Prerequisites and controlled fixtures

Gap-complete matrix

Same-parameters endurance and evidence

Execution and completion boundaries

Final beta.23 campaign reconciliation (authoritative over earlier draft)

Next campaign admission and retained state

Beta.24 candidate ready for owner deployment and retest

Activity

Proxicon commented on Oct 17, 2026

Proxicon commented on Oct 19, 2026

Proxicon commented on Oct 19, 2026

Proxicon commented on Oct 19, 2026

Proxicon commented on Oct 20, 2026

Proxicon commented on Oct 21, 2026

Proxicon commented on Oct 21, 2026

1 remaining item

Proxicon commented on Oct 21, 2026

Proxicon commented on Oct 21, 2026

Proxicon commented on Oct 21, 2026

Proxicon commented on Oct 21, 2026

Proxicon commented on Oct 21, 2026

beta.41 campaign checkpoint — NO-GO (21 Oct 2026)

Proxicon commented on Oct 24, 2026

Current completion register — beta.47 delivery, beta.46 owner credit

Credited beta.46 outcomes

Original unresolved register

Beta.47 source delivery

Historical evidence and prospective successors

Remaining owner release gates

Proxicon commented on Oct 25, 2026

Proxicon commented on Oct 25, 2026

Proxicon commented on Oct 25, 2026

Proxicon commented on Oct 27, 2026

Final beta.49 assessment — 27 October 2026

Proxicon commented on Oct 27, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development