[M6-09] Qualify backup, restore, upgrade, configuration, and process

[vicotrbb]

Release 1 source: #1. The master specification is referenced, not modified by this ticket.

Execution signal: ready-for-agent means this issue is implementation-ready once every native GitHub blocker below is closed.

Outcome

Execute exact local and external Repository identity/CAS/multipart/checksum/interruption/throttle/purge cells, online incremental backup, key recovery, fresh restore, query validation, patch/minor/format upgrades, every failure phase, configuration parity/reload/migration/drift, and native/systemd/Docker/Kubernetes lifecycle.

Scope and semantic ownership

• Milestone: M6
• Global sequence: 142 / 149
• Semantic owner: Recovery and Lifecycle
• Highest useful behavior seam: Qualify backup, restore, upgrade, configuration, and process
• Affected registered scopes: positron-backup
• Expected project size / estimate: XL / 13
• Completion status rule: An exact cell may become Qualified only when the frozen candidate passes the complete real-target gate and immutable evidence is retained. Any failed, missing, stale, unsupported, inconclusive, or exceptional required cell remains release-blocking.

No weaker document, implementation convenience, partial target, emulator, local diagnostic, or green subset may reinterpret the binding contract.

Native blockers

• [M6-01] Freeze the exact Release Candidate #135 — M6-01

The native blocked-by graph is authoritative for execution order. Do not start implementation while a blocker is open, while an affected scope remains unlawfully scaffold-only, or while a required decision/review is missing.

Binding product and architecture references

• Frozen Positron product contract
• Binding domain language
• Release 1 qualification and M6 milestone contract
• Whole-application module and flow design
• Project contract §3.12 Tenant Lifecycle
• Project contract §3.14 Retry-Safe Administration
• Project contract §3.19 Multi-Provider Backup Repositories
• Project contract §19.1 Backup and Restore
• Project contract §3.25 Release 1 Scope and Qualification
• Application Design §4.13 Backup and Restore
• Application Design §6.6 Backup and restore
• Application Design §6.7 Tenant Purge
• Application Design §7 State ownership
• Application Design §10.1 Module contract tests

Accepted ADRs

• ADR-0007 — limit v1 deletion to segment retention and tenant purge
• ADR-0026 — use verified incremental full instance backups
• ADR-0036 — require independent key recovery before backup
• ADR-0039 — make tenant purge cryptographically complete within managed scope
• ADR-0040 — sign backup control state and audit checkpoints
• ADR-0051 — make single instance upgrades declarative fenced and recoverable
• ADR-0052 — make operator backups visible application verified and restore only
• ADR-0063 — make administrative mutations idempotent and long work durable
• ADR-0068 — qualify native backup repository adapters by provider behavior
• ADR-0074 — freeze release 1 through an evidence backed qualification matrix

If implementation would change caller knowledge, semantic ownership, a public or durable format, compatibility, a non-waivable invariant, or Release 1 scope, stop and land the required accepted superseding ADR together with its contract, migration, tests, and gates.

Engineering invariants and gates

• Mandatory engineering invariants: ARC-02, ARC-03, ARC-05, SAFE-03, SAFE-04, CON-02, CON-03, ERR-01, ERR-04, SEC-01, SEC-02, SEC-03, SEC-04, TEST-01, TEST-02, TEST-03, TEST-04, TEST-05, TEST-07, PERF-01
• Authoritative gates: EG-00, EG-ARCH, EG-BUILD, EG-DEPS, EG-DOCS, EG-ERROR, EG-EVIDENCE, EG-POLICY, EG-RUST, EG-SAFETY, EG-SECRETS, EG-SUPPLY, EG-TEST, EG-CONCURRENCY, EG-CORRECT, EG-CRYPTO, EG-DYNAMIC, EG-FAULT, EG-INTEGRITY, EG-MATRIX, EG-RESOURCE, EG-SECURITY
• Canonical runner: cargo xtask quality
• Policy sources: Engineering standards, Quality gates, Contribution and activation policy, Application scope registry, Artifact scope registry, Allowed dependency graph

Do not weaken, skip, rename, path-filter, delete, reclassify, or silently bypass a gate, target, threshold, owner, test, corpus, fixture, hook, workflow, or evidence field.

Affected Qualification Cells

• Q-BACKUP-001
• Q-BACKUP-002
• Q-UPGRADE-001
• Q-TENANT-001
• Q-CONFIG-001
• Q-PROCESS-001

Each cell remains independently reportable. A passing child cannot hide a failed, unsupported, missing, unresolved, or inconclusive sibling.

Required implementation contract

• Deliver the outcome exactly as stated above, through the named semantic owner and highest useful behavior seam.
• Represent backup, restore, purge, and long recovery work as restartable Durable Operations with explicit irreversible boundaries.
• Create verified incremental full-instance snapshots online and restore only into fresh storage.
• Bind manifests, objects, keys, repository identity, signatures, audit, and Purge Tombstones.
• Preserve interruption, cancellation, retry, retention, provider failure, and recovery truth.
• If any affected code or artifact scope is still scaffold, activate it atomically with its owner, exact edges, risk gates, test commands, measured coverage/mutation baselines, and required threat model or format/API decision before adding behavior.
• Preserve the registered acyclic dependency graph and keep provider/protocol/distribution types behind their adapters.
• Return closed typed outcomes with safe details, retry class, completion state, and source context. Never use strings for control flow.
• Declare every externally reachable input, work, memory, allocation, copy, I/O, task, queue, cache, lease, retry, and latency bound plus overload behavior.
• Add no dependency without the complete review required by CONTRIBUTING.md.
• Keep generated output generated and prove clean deterministic regeneration.
• Update user/operator compatibility, migration, failure, recovery, and release-note documentation where behavior is visible.

Required tests

• Snapshot consistency, incremental reuse, interruption, cancellation, crash, restart, retention, key recovery, signature, and registry recovery.
• Fresh-target restore with empty-target proof, wrong identity, missing key/object, corruption, purge tombstone, and query validation.
• Exact real-repository target tests; emulators may accelerate development but cannot qualify.
• Include positive, boundary, negative, and adversarial cases at the lowest interface that proves the contract.
• Assert returned outcomes, durable externally readable state, published artifacts, or public behavior—not private fields, helper calls, queue layout, or incidental ordering.
• Keep tests hermetic, deterministic, parallel-safe, bounded, and retain seeds, clocks, entropy, schedules, fault plans, fixtures, and minimized failures.
• A reproducible defect fix begins with a failing regression test and retains the reproducer.
• Run all scope-selected unit, compile-fail, contract, property, integration, end-to-end, compatibility, platform, real-target, fuzz, model, fault, sanitizer, and soak suites that apply.

Evidence and completion

• Run the authoritative cargo xtask quality profile required by the change and retain its exact revision-bound evidence.
• Record exact source revision, artifact digests, target/environment, configuration, fixtures, dataset/workload, fault schedule, toolchain, command, duration, raw measurements, safe logs/metrics, result, owner, and verifier as applicable.
• Retain passing, failing, inconclusive, exceptional, timeout, cancellation, and missing-tool attempts. A later pass must not erase a failure.
• Verify no panic/unwrap/expect, unchecked indexing, ignored result, detached task, unbounded resource, ambient authority, prohibited unsafe code, secret leak, or hidden best-effort path was introduced.
• Verify the working tree and generated artifacts are clean after the authoritative runner.
• Do not close this issue from an ad hoc command, mock-only proof, emulator-only proof, local dirty evidence, or partial Qualification Cell.

Explicit non-goals

• No continuous PITR, selective-tenant restore, individual-record recovery, merge restore, or in-place restore.
• No CSI VolumeSnapshot treated as a Positron Backup Snapshot.
• No Release 1 Metrics or Profiles implementation.
• No replication, consensus, HA, automatic failover, follower reads, clustering, or live shard split/merge.
• No hidden feature flags, placeholder APIs, empty adapters, speculative modules, or disabled deferred behavior.

Agent handoff checklist

• Re-read the referenced contracts and ADRs before changing code.
• Identify the exact semantic owner, caller-visible seam, durable publication owner, and affected Qualification Cells in the implementation PR.
• Explain the lifecycle and failure boundaries before editing.
• Keep implementation, qualification, publication, deployment, and release as separate states and authorizations.
• Leave an evidence-backed status if blocked; never fabricate Implemented, Qualified, or merge eligibility.