release: freeze Nexus One AI 1.0.0-rc1

specs/001-enterprise-experience/tasks.md

@@ -11,6 +11,104 @@ deployment, recovery, and artifact-evidence tasks are included.
 **Organization**: Tasks are grouped by user story. Phase 1 product delivery consists of Setup,
 Foundational, and User Story 1; later story phases remain separately acceptable increments.
 
+## Installer Architecture Planning Tasks (Docs Only)
+
+- [x] T074 Extend this feature with the four-artifact release matrix and Workstation-first sequence.
+- [x] T075 Define zero-network, payload, signing, update, rollback, interruption, sizing, and evidence contracts in `contracts/installer-artifact-contract.md`.
+- [x] T076 Run and record fresh consistency analysis in `analysis/installer-release-consistency.md`.
+- [ ] T077 Approve signing ownership, custody, rotation, revocation, and expiry policy (B1).
+- [x] T078 Define canonical manifest/signature/compatibility schemas and test vectors (B2); named approval remains under T086.
+- [ ] T079 Approve representative Workstation hardware and immutable starter-model inventory (B3).
+- [x] T080 Define Online Universal Server acquisition endpoint classes, trust, resume, and failure contract (B4); production values and owner approval remain under T100.
+- [ ] T081 Resolve update rollback, offline license, payload inventory, lab, and evidence ownership decisions (B5-B8).
+- [x] T082 Re-run consistency analysis and pass Architecture-ready Gate 1; record explicitly that this does not authorize installer implementation or media building.
+- [ ] T083 After Offline Workstation passes Gates 1-5, amend the shared contract from evidence and authorize or defer Offline Server S/M planning.
+- [x] T084 Draft canonical artifact-manifest and detached-signature JSON Schemas.
+- [x] T085 Record repository-derived Workstation baseline and recommended signing/update/license decisions in `contracts/architecture-decisions.md`.
+- [ ] T086 Obtain named owner approval of the schemas, compatibility evaluation and completed immutable manifest/signature positive and negative test vectors.
+- [x] T087 Remove circular manifest self-digest and define deterministic compatibility evaluation and reason codes.
+- [x] T088 Add conspicuously non-production structural and compatibility test vectors.
+- [x] T089 Add dedicated Ed25519 test-key signature vectors, validator evidence, and examples for model, update, and trust artifacts.
+- [x] T090 Add a real Workstation Ed25519 positive vector, tamper/unknown-key negatives, and structural examples for model, update, and trust artifacts without retaining the private test key.
+- [x] T091 Add an automated Draft 2020-12 schema, Ed25519, and compatibility fixture validator.
+- [x] T092 Run the automated suite and record passing schema (Draft 2020-12), signature (4/4), and compatibility (12/12) evidence.
+- [x] T093 Sign all four artifact examples with one transient test key, verify each signature and tamper case, then delete the private key.
+- [x] T094 Verify official Workstation hardware architecture/specifications and immutable starter-model candidate revisions, digests, and licenses.
+- [x] T095 Confirm universal hardware-adaptive Workstation media per CPU architecture; GPU model/class does not create separate media or commercial tiers.
+- [ ] T096 Approve representative machines, firmware/Secure Boot, driver/CUDA/runtime pins, model redistribution/attribution, and measured file inventory.
+- [ ] T097 Decide later ARM64 DGX Spark media, prove a multi-architecture image, or remove DGX Spark support; validate AMD64 CPU-only, 8 GB, 24 GB, and 48 GB-or-higher profiles.
+- [x] T098 Exclude AMD GPU/ROCm/Strix Halo acceleration from the first Workstation release while retaining truthful CPU-only fallback on AMD64 hosts.
+- [x] T099 Draft the Online Universal Server endpoint, trust, acquisition, resume/cache, failure, rollback, and evidence contract.
+- [ ] T100 Obtain platform/security decisions for production endpoints, credential issuer, mirrors, proxy/customer CA, release channels, and retry/cache/SLA values.
+- [x] T101 Draft shared offline update ownership, update classes, checkpoints, activation, retention, health, interruption, and rollback policy grounded in current backup/migration behavior.
+- [ ] T102 Approve atomic switch/snapshot mechanisms, full backup inventory/SLA, prior-version window, disk quota, driver/kernel recovery, health window, and break-glass authority.
+- [x] T103 Draft zero-network license import, activation, staging, clock, expiry, binding/rehost, revocation, interruption and evidence policy.
+- [ ] T104 Approve staging/expiry matrices, clock tolerance, device identity/rehost, revocation ages and license/artifact trust ownership; remove optional binding enforcement and authorization use of staging tier defaults during implementation.
+- [x] T105 Produce repository-derived four-artifact payload boundaries, role matrix, exact-digest deduplication rules, sizing worksheet and media-capacity evidence policy.
+- [ ] T106 Resolve every payload object/version/digest/license, build and measure each artifact, reconcile extracted media, and approve the Server L/Max split justification.
+- [x] T107 Define offline lab isolation, attempted-network capture, physical hardware, interruption/security matrix, evidence package, independent review and stop/retry rules.
+- [ ] T108 Assign named lab/reviewer owners, exact hardware/topology/tools/schedule/evidence storage, then execute and independently review the required clean and failure runs.
+- [x] T109 Draft artifact naming/version/channel, SPDX 3.0.1 SBOM, in-toto v1.2/SLSA v1.2 provenance, reproducibility, retention, support cadence and withdrawal policy.
+- [ ] T110 Approve lifecycle formats/toolchain, publication/archive locations, retention/privacy periods, supported-version/backport window, cadence/EOL and named owners.
+- [x] T111 Define model-pack quarantine, content-addressed sharing, dependency, activation, uninstall, garbage-collection, entitlement-change and redistribution-license behavior (B10).
+- [ ] T112 Approve model-store mechanisms, quotas/retention, customer-data handling, license continuity/rehost rules, withdrawal authority, GC policy and named owners; execute Gate 5 lifecycle vectors.
+- [x] T113 Define release trust domains, key roles, dual-control custody, signing handoff/receipts, rotation, offline trust distribution, revocation and emergency recovery contract (B1).
+- [ ] T114 Name signing/trust owners and alternates; approve hardware/custody/backup, key and overlap lifetimes, revocation age, offline time, compromised-root recovery, emergency authority and ceremony evidence; execute signing acceptance vectors.
+- [x] T115 Define installer release assets, adversaries, trust boundaries, threat/control register, security invariants, abuse/interruption tests and residual risks for Gate 1.
+- [ ] T116 Obtain Security/Platform/Release/Product/Support approval of the threat model, assign every residual risk and execute/link the minimum control evidence.
+- [x] T117 Consolidate all Gate 1 decisions, accountable roles, acceptable outcomes, dependencies and evidence requirements into an approval dossier.
+- [x] T118 Obtain Cezen Technologies approval-record IDs for G1-01 through G1-30, retain authorized human attribution in restricted evidence, allocate open evidence to its proper downstream gate, and record Gate 1 pass without authorizing implementation implicitly.
+- [x] T120 Adopt Cezen Technologies as the visible accountable organization while preserving authorized-person attribution and separation-of-duties proof in restricted approval evidence.
+- [x] T121 Record Cezen Technologies approval `CT-G1-2026-001` for G1-12 AMD64/NVIDIA/CPU-fallback scope and G1-17 later separate ARM64/DGX Spark identity; retain G1-13–G1-16 as evidence-open.
+- [x] T122 Record Cezen Technologies approval `CT-G1-2026-002` for G1-08–G1-11 artifact schemas, compatibility, identity/provenance and immutable publication/withdrawal direction; retain production tooling, archive and retention evidence gaps.
+- [x] T123 Record Cezen Technologies approval `CT-G1-2026-003` for G1-01–G1-07 trust domains, custody/rotation/revocation/recovery direction, threat controls and required platform-security baseline; retain operational values, assignments and executed evidence gaps.
+- [x] T124 Record Cezen Technologies approval `CT-G1-2026-004` for G1-18–G1-24 online acquisition, update/checkpoint/rollback and offline license/binding/rehost direction; retain production values, mechanisms and executed evidence gaps.
+- [x] T125 Record Cezen Technologies approval `CT-G1-2026-005` for G1-25–G1-30 model lifecycle, customer-data preservation, lab/evidence methods, acceptance inventory and Server L/Max deferral; retain operational values, assignments and executed evidence gaps.
+- [x] T126 Record Cezen Technologies approval `CT-G1-2026-006` for the G1-13 CPU-only/sub-threshold, NVIDIA 8 GB, 24 GB and 48 GB-or-higher profile matrix without claiming unavailable physical-machine evidence.
+- [x] T127 Derive a coherent G1-14 runtime candidate from repository and official evidence; identify the driver 550/CUDA 12.4, declared CUDA 12.6 and vLLM CUDA 12.8.1 contradiction; propose R570/CUDA 12.8.1-compatible branch baseline with immutable-lock and physical-test gates.
+- [x] T128 Record Cezen Technologies approval `CT-G1-2026-007` for the G1-14 Ubuntu 22.04.5/R570/CUDA 12.8.1/vLLM 0.10.2 runtime direction and CPU independence; retain exact locks and physical/rollback evidence gaps.
+- [x] T129 Record Cezen Technologies approval `CT-G1-2026-008` for the G1-15 pinned Phi-3 Mini, Nomic Embed Text v1.5 and Microsoft Phi-3 Mini starter set and license/hardware boundaries; retain packaged file/digest/notice and redistribution evidence gaps.
+- [x] T130 Record Cezen Technologies approval `CT-G1-2026-009` for the G1-16 Offline Workstation payload inclusion/exclusion boundary; retain resolved signed inventory, exact size/digest/license, reconciliation and measurement evidence gaps.
+- [x] T131 Audit Gate 1 closure against the normative gate definition and allocate every open item and C1-C7 contradiction to Gates 2-6 or deferred Server planning.
+- [ ] T132 Under separately authorized scope, complete the Gate 2 immutable inventory, concrete mechanisms/policy values, restricted operator/lab assignments and implementation plan for Offline Workstation.
+- [x] T133 Inspect build host 172.16.10.144, confirm no active legacy build, quarantine the 30 GB legacy workspace, create a clean workspace, synchronize the current source/spec snapshot and record deterministic source/host evidence without starting a build.
+- [x] T134 Add an immutable Offline Workstation payload-lock contract and fail-closed preflight, block the legacy mutable acquisition path, add dependency-free tests, and prove on the build host that unresolved input creates no staging/artifact output.
+- [x] T135 Add and run connected resolver increment 001 for a deterministic source archive and six Linux/AMD64 OCI manifests, prove source reproducibility, record registry drift/unresolved licenses and keep artifacts empty.
+- [x] T136 Add and run an isolated empty-status APT resolver, expand 26 requested packages to 487 exact measured `.deb` objects, retain repository-snapshot/license/NVIDIA closure gaps, and prove no package installation or artifact creation occurred.
+- [x] T137 Add and run a Python 3.11 wheel-only resolver against combined inputs; record its fail-closed bcrypt/Chroma result and prove the failed run produced no candidate lock, wheel payload or artifact.
+- [x] T138 Reconcile the apparent dependency conflict against the real Ansible/runtime boundaries, preserve separate backend/Chroma/platform environments, resolve 363 exact wheel objects without changing either constraint, and retain the candidate lock and zero-artifact evidence.
+- [x] T139 Derive the CUDA 12.8 Python ML pins from authoritative vLLM 0.10.2 wheel metadata, resolve Torch 2.8.0/Torchaudio 2.8.0/Torchvision 0.23.0 into a fresh 361-object candidate, prove zero CUDA 13 objects and retain zero-artifact evidence.
+- [ ] T140 Validate the pinned stack on approved R570/CUDA 12.8.1 GPU profiles and an NVIDIA-independent CPU-only profile; complete index-origin/resolver-report approval, license/notices, profile-aware installation and clean disconnected Python 3.11 install/import evidence.
+- [x] T141 Prove network-disabled Python 3.11 installation/import for the backend, ChromaDB and platform wheelhouses; prove CPU tensor execution with CUDA hidden and supplemental CUDA 12.8 execution on the build host RTX 3050/Open Driver 595.58.03, retaining hashed logs and zero-artifact evidence.
+- [x] T142 Define exactly-one CPU/NVIDIA Python runtime profile selection, add exact CPU/cu128 wheel identities and a fail-closed validator, assemble a 196-wheel CPU-only selection with zero GPU objects, and prove disconnected CPU-only installation/import/tensor execution.
+- [ ] T143 Integrate signed profile membership with the future installer hardware detector, pre-mutation journal, resume and rollback path; test unsupported/failed detection, interrupted selection, attempted profile change and tampered/ambiguous membership.
+- [x] T144 Add a non-mutating fail-closed profile-selection planner and test hardware fallback, NVIDIA readiness, structural membership, start/resume, active-operation exclusion, rollback-before-change and new-operation-after-rollback semantics; retain build-host vectors and zero-artifact evidence.
+- [x] T145 Define a closed profile-membership verification-receipt schema, replace the planner's bare verification boolean with manifest/envelope/key/trust/policy/namespace/profile/verifier binding, and prove production rejection plus explicit structural-test acceptance.
+- [ ] T146 Implement the production Ed25519/trust-policy verifier and atomic fsync/rename profile journal/checkpoint writer; prove stale-generation, torn-write, power-loss, tamper and recovery behavior before installer integration.
+- [x] T147 Implement and test the isolated atomic profile journal writer with locking, generation CAS, digest chaining, immutable identity, checkpoint-before-install transition, mode 0600, file/directory fsync and atomic rename; retain stale-writer/tamper/transition evidence without claiming checkpoint creation or power-loss proof.
+- [x] T148 Implement the production-shaped Ed25519 profile-membership verifier with canonical manifest digest, namespace, key purpose/environment/status, fingerprint and policy enforcement; prove positive/tamper/wrong-environment behavior with ephemeral test keys without claiming production signing evidence.
+- [x] T149 Implement and test an isolated hashed Python-environment checkpoint with original-absence handling, file/directory/symlink inventory, same-filesystem fsync/atomic replacement, pre-restore verification and failed-tree quarantine; retain exact restore evidence without claiming crash or live-path coverage.
+- [x] T150 Add gated test-only crash injection and deterministic restore recovery; prove pre/post checkpoint publication and post-quarantine/post-activation forced-process termination outcomes on disposable build-host trees without claiming physical power-loss durability.
+- [x] T151 Compose receipt, planner, exact profile validation, atomic journal and checkpoint into a non-mutating preparation flow; prove invalid payload leaves no state and valid strict-CPU preparation reaches idempotent `checkpoint_ready` without package mutation.
+- [x] T152 Implement the isolated offline Python profile executor with checkpoint-ready enforcement, no-index install, profile health checks, fsynced atomic activation and rollback transitions; prove strict-CPU install to `installed` plus dependency-failure rollback on disposable targets.
+- [x] T153 Add gated post-activation executor termination and recovery-only handling; prove a full disconnected CPU environment crash restores the exact checkpoint, quarantines the activated environment and records generation-5 `rolled_back`.
+- [x] T154 Strengthen NVIDIA health to require CUDA availability plus synchronized device execution, then prove full disconnected `nvidia-cu128` install/health/atomic activation to `installed` on the build-host RTX 3050 without claiming the R570 or 24/48 GB matrix.
+- [x] T155 Run fresh consistency audit Pass 4, reconcile C1-C7 against executed Gate 2 evidence, correct stale authority/TBD statements, and publish exact remaining readiness blockers without claiming an artifact or Gate 2 pass.
+- [x] T156 Replace the stale offline README build sequence with one authoritative operator flow that classifies resolver/profile/checkpoint/install stages, marks bundle/ISO assembly blocked, defines mandatory stop conditions and forbids release claims from disposable evidence.
+- [x] T157 After external-action quota reset, synchronize the complete current source/spec snapshot to the clean build host, run all 27 remote deployment tests including Ed25519, hash the suite log and reconfirm zero artifact entries.
+- [x] T158 Prove full disconnected `nvidia-cu128` post-activation process-crash rollback on the RTX 3050: restore the exact checkpoint, preserve the executable CUDA environment in quarantine, record generation-5 `rolled_back`, and retain zero-artifact evidence.
+- [x] T159 Add the boot-safe profile recovery dispatcher and prove a strict-CPU post-activation crash is recovered by a separate fresh network-disabled container invocation; verify terminal idempotence, exact checkpoint restore, quarantine retention and zero artifacts without claiming an actual reboot or power-loss test.
+- [x] T160 Convert the build host from DHCP to the same observed persistent static address, then prove a fresh strict-CPU interrupted operation recovers during a controlled OS reboot; retain the exact-path mismatch negative vector, remove the temporary boot unit, and keep physical power-loss and production integration claims open.
+- [x] T161 Prove bounded strict-CPU target byte and inode exhaustion in separate executable network-disabled tmpfs vectors with independent reserved journal/checkpoint capacity; require real `ENOSPC`, unchanged original target, generation-5 terminal rollback, read-only re-verification and zero artifacts.
+- [x] T162 Add a direct bounded post-activation service-health probe boundary and prove a full disconnected strict-CPU activation whose probe observes the new environment then exits non-zero restores the checkpoint, quarantines the activated environment and records generation-5 rollback with zero artifacts.
+- [x] T163 Model the intended model/vector/API/proxy dependency graph with uniquely named disposable systemd units; prove explicit ordering/requirements, a 3-sample 12-check semantic health window and vector-failure `OnFailure` rollback, then remove every temporary unit and record the non-compliant real-unit wiring as an open blocker.
+- [x] T164 Design candidate production API/Chroma/activation-health/rollback units plus nginx ordering drop-in and active-operation environment contract under the spec; validate syntax and dependency cycles on systemd 249 using `/tmp`-only executable stubs without installing/reloading units, and retain missing real executables/nginx/security scoring as blockers.
+- [x] T165 Implement the isolated deterministic loopback-only activation health-window tool with canonical atomic evidence and stable exits; prove consecutive ready, explicitly allowed optional degradation, timeout, malformed response, non-loopback rejection and HTTP-503 rollback-trigger behavior in unit tests and a network-disabled integration container.
+- [x] T166 Add a monotonic total-window deadline and signed per-component systemd restart ceilings; prove deterministic deadline/storm tests plus real isolated systemd vectors that reject an impossible next sample and eight-restart storm with exit 23 before HTTP, then remove all temporary units.
+- [x] T119 Run fresh consistency audit pass 2, reconcile task/decision counts, verify JSON parsing and document that remaining payload-size TBDs require build evidence outside docs-only scope.
+
+T077-T083 are planning/evidence gates, not authorization to change installer code or build media.
+
 ## Format: `[ID] [P?] [Story] Description`
 
 - **[P]**: Can proceed in parallel because it targets independent files after prerequisites pass.