{"schema_version":1,"research_id":"eoa_inverse_innovation_exp03_external48_20260801","source_assessment_id":"eoa_inverse_innovation_exp03_opportunity320_20260801","cell_id":"layer_decay_and_expiration_management__computer_science","selection_stratum":"DEPLOYABLE_PRIORITY","search_queries":["site:docs.gitlab.com container registry reduce storage cleanup policy tags protected","site:docs.aws.amazon.com Amazon ECR lifecycle policies reference count images","site:docs.aws.amazon.com Amazon ECR archive restore archived images","site:cloud.google.com artifact registry cleanup policies dry run delete keep tags","site:goharbor.io docs garbage collection untagged artifacts dry run","OCI distribution specification referrers deletion manifest digest","site:kubernetes.io containers images digest pull policy","site:docs.aws.amazon.com ECS resources block task definition deletion"],"sources":[{"source_id":"S1","title":"Reduce container registry storage","publisher":"GitLab","url":"https://docs.gitlab.com/user/packages/container_registry/reduce_container_registry_storage/","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"","accessed_at":"2026-08-02","claims_supported":["Registry growth can make image and tag listing slower and consume substantial storage.","GitLab cleanup policies select tags by name, age and retained count while excluding protected and immutable tags.","Deleting tags and reclaiming unreferenced layers are separate operations, and shared layers complicate storage accounting.","Maintainer, Owner or administrator authority is required for relevant registry-management operations."]},{"source_id":"S2","title":"Automate the cleanup of images by using lifecycle policies in Amazon ECR","publisher":"Amazon Web Services","url":"https://docs.aws.amazon.com/AmazonECR/latest/userguide/LifecyclePolicies.html","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"","accessed_at":"2026-08-02","claims_supported":["ECR lifecycle policies support preview followed by archive or expiration actions and log lifecycle actions in CloudTrail.","Policies select images by tag, push age, pull age or retained count.","An image referenced by a manifest list cannot be archived or expired until the manifest list is handled.","Reference artifacts associated with a subject image are archived or expired with that image."]},{"source_id":"S3","title":"Archiving an image in Amazon ECR","publisher":"Amazon Web Services","url":"https://docs.aws.amazon.com/AmazonECR/latest/userguide/archive_restore_image.html","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"","accessed_at":"2026-08-02","claims_supported":["ECR provides an archival storage class for rarely accessed images retained for compliance or long-term reference.","Archival can be driven by push age, last-pull age or repository image count.","Archived images can later be restored, while lifecycle rules may permanently delete them after another interval."]},{"source_id":"S4","title":"Restoring an image","publisher":"Amazon Web Services","url":"https://docs.aws.amazon.com/AmazonECR/latest/userguide/restore-image.html","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"","accessed_at":"2026-08-02","claims_supported":["An archived ECR image can be returned to standard storage by console or API.","An archived image cannot be pulled while restoration is pending, and restoration typically takes up to 20 minutes.","Restoration changes image state and can trigger scanning and replication behavior."]},{"source_id":"S5","title":"Cleanup policy overview","publisher":"Google Cloud","url":"https://docs.cloud.google.com/artifact-registry/docs/repositories/cleanup-policy-overview","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"2026-07-17","accessed_at":"2026-08-02","claims_supported":["Artifact Registry provides conditional delete and keep policies for repositories containing many artifact versions.","A matching keep policy takes precedence over a delete policy.","Cleanup policies support dry runs and active runs, and attachments are deleted with their subject image."]},{"source_id":"S6","title":"The OpenContainers Distribution Spec, version 1.1.1","publisher":"Open Container Initiative","url":"https://specs.opencontainers.org/distribution-spec/?v=v1.1.1","source_class":"STANDARD","publication_date":"2025-11","accessed_at":"2026-08-02","claims_supported":["OCI Distribution is a standards-track specification for registry distribution operations.","OCI manifests can be addressed by tag or content digest, and the specification defines discovery of artifacts referring to a subject digest."]},{"source_id":"S7","title":"Images","publisher":"Kubernetes","url":"https://kubernetes.io/docs/concepts/containers/images/","source_class":"OFFICIAL_GUIDANCE","publication_date":"","accessed_at":"2026-08-02","claims_supported":["Kubernetes workload specifications may identify container images by immutable digest.","A digest-pinned workload can require a future registry pull depending on image-pull policy and local cache state.","Using the latest tag makes production tracking and rollback harder than meaningful tags or digests."]},{"source_id":"S8","title":"Amazon ECS task definition states","publisher":"Amazon Web Services","url":"https://docs.aws.amazon.com/AmazonECS/latest/developerguide/task-definition-state.html","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"","accessed_at":"2026-08-02","claims_supported":["ECS uses a visible DELETE_IN_PROGRESS state rather than immediately deleting an inactive task definition.","ECS blocks final deletion while active tasks, deployments or task sets depend on the task-definition revision.","Operators can identify the resources blocking deletion."]           
    }
    ],
  "problem_evidence": {
    "support": "STRONG",
    "rationale": "GitLab directly documents that unmanaged registry growth consumes storage and slows image or tag listing, while Google describes cleanup policies as useful where many versions exist but only selected production versions need retention. Shared layers and separate tag-versus-blob deletion confirm that apparently simple cleanup has reference-safety complications. These sources establish the general problem, not its magnitude or incident rate in the proposed adopter's registry.",
    "source_ids": ["S1", "S5"]
  },
  "stakeholder_evidence": {
    "support": "MODERATE",
    "rationale": "Multiple registry vendors expose lifecycle, retention, archival and dry-run controls, demonstrating operational relevance. GitLab assigns relevant actions to Maintainers, Owners and administrators. Evidence of expressed demand, budget commitment or a named adopter for this specific cross-system composition was not found.",
    "source_ids": ["S1", "S2", "S3", "S5"]
  },
  "prior_art": {
    "proximity": "SUBSTANTIAL_COLLISION",
    "closest_analogues": [
      {
        "name": "Amazon ECR lifecycle policies plus ECR Archive",
        "similarity": "This directly combines policy-based selection, preview, archive or expiration, pull-age and push-age criteria, manifest-list dependency blocking, associated-reference handling, CloudTrail records and later restoration.",
        "remaining_difference": "The reviewed documentation does not establish a unified veto spanning external deployment definitions, rollback and provenance catalogs, authority-controlled legal or investigation holds, reversible quarantine with tombstones, and periodic sampled restore verification. That documentation boundary is not proof that no such integration exists.",
        "source_ids": ["S2", "S3", "S4"]
      },
      {
        "name": "Google Artifact Registry cleanup policies",
        "similarity": "Google provides ordered delete and keep policies, keep-over-delete precedence, immutable-tag protection, dry-run evaluation and audit-log inspection.",
        "remaining_difference": "The reviewed feature is registry-policy driven; it does not establish external dependency discovery, archival restoration tests, a distinct quarantine state or authoritative cross-system hold workflow.",
        "source_ids": ["S5"]
      },
      {
        "name": "Amazon ECS dependency-gated task-definition deletion",
        "similarity": "ECS visibly stages deletion and blocks final removal while known tasks, deployments or task sets still depend on a revision, closely matching the dependency-veto mechanism.",
        "remaining_difference": "It governs ECS task definitions rather than registry artifacts and does not provide artifact tiering, archive restoration, tombstones or cross-platform reference coverage.",
        "source_ids": ["S8"]
      }
    ],
    "distinctive_claim_remaining": "Against a registry-native lifecycle preview and archive baseline, adding cross-system deployment, rollback and provenance reference checks plus authority-controlled holds will identify materially more unsafe disposition candidates; reversible quarantine, tombstones and periodic digest-verified restores will then reduce obsolete discoverability without exceeding preset false-block, recovery-latency or operator-effort bounds.",
    "confidence": "HIGH"
  },
  "implementation_evidence": {
    "support": "MODERATE",
    "rationale": "ECR supplies lifecycle preview, archive, audit events, registry-local dependency handling and restoration; Google supplies dry-run keep/delete evaluation; Kubernetes exposes digest-pinned workload references; and ECS demonstrates dependency-blocked staged deletion. These make the composition technically plausible. The difficult unverified part is joining registry metadata to complete external deployment, rollback, provenance and hold inventories.",
    "source_ids": ["S2", "S3", "S4", "S5", "S7", "S8"]
  },
  "scores": {
    "meaningful_impact": {
      "score": 4,
      "rationale": "Official documentation confirms storage and listing burdens and shows that retained versions require differentiated treatment. Protecting rollback and digest-pinned workloads is operationally important, although local stale-byte volume and incident frequency remain unknown.",
      "source_ids": ["S1", "S5", "S7"]
    },
    "stakeholder_pull": {
      "score": 3,
      "rationale": "Lifecycle, retention, archive and dry-run features across major registries indicate a recognized need, but no named adopter, incident history or budget commitment supports this exact composition.",
      "source_ids": ["S1", "S2", "S3", "S5"]
    },
    "incremental_advantage": {
      "score": 3,
      "rationale": "Cross-system dependency and hold vetoes could improve on registry-native criteria, but much of the proposed archive, restore, preview, audit and reference-aware lifecycle is already available in ECR. Advantage must therefore be tested against native lifecycle tooling, not merely age-only deletion.",
      "source_ids": ["S2", "S3", "S4", "S7", "S8"]
    },
    "distinctiveness_plausibility": {
      "score": 2,
      "rationale": "The broad lifecycle composition substantially collides with ECR Archive and Google cleanup policies. Only the integrated external-dependency, authoritative-hold, quarantine/tombstone and periodic-restore-verification layer remains plausibly distinctive.",
      "source_ids": ["S2", "S3", "S4", "S5", "S8"]
    },
    "technical_implementability": {
      "score": 4,
      "rationale": "Registry APIs already expose previews, lifecycle state, archival and restoration, while workload definitions provide inspectable digest references. Completeness across hidden consumers and heterogeneous rollback or provenance systems remains the main technical constraint.",
      "source_ids": ["S2", "S3", "S4", "S5", "S7"]
    },
    "adoption_authority_feasibility": {
      "score": 3,
      "rationale": "Registry administrators clearly have implementation authority, but the remaining claim requires service-owner, deployment-platform, security, compliance and legal coordination beyond ordinary repository administration.",
      "source_ids": ["S1", "S5", "S8"]
    },
    "evidence_readiness": {
      "score": 4,
      "rationale": "Native lifecycle previews and dry runs permit a bounded comparison without deletion, and archive/restore APIs permit isolated fidelity testing. Local acceptance thresholds and adjudicated dependency ground truth still need definition.",
      "source_ids": ["S2", "S4", "S5"]
    },
    "safety_net_benefit": {
      "score": 5,
      "rationale": "A read-only shadow comparison plus restoration testing on copied artifacts avoids production deletion. Existing preview, dry-run, staged-deletion and restore mechanisms provide strong reversible safety primitives.",
      "source_ids": ["S2", "S4", "S5", "S8"]
    },
    "scalability": {
      "score": 3,
      "rationale": "Vendor lifecycle engines demonstrate scalable registry-side processing, but organization-wide dependency discovery, exception adjudication, hold synchronization and recurring restore verification can scale in coordination cost and remain unproven.",
      "source_ids": ["S2", "S5", "S7", "S8"]
    }
  },
  "score_confidence": "MODERATE",
  "costs": {
    "first_evidence": {
      "band_2026_usd": "10K_TO_50K",
      "scope": "Thirty-day read-only shadow study of at most 500 superseded artifacts in one nonproduction namespace, including native lifecycle preview, one deployment and rollback-reference scan, hold matching, manual adjudication of a stratified sample, isolated archive/restore tests on copied artifacts, analysis and stakeholder review.",
      "confidence": "MODERATE",
      "assumptions": [
        "Existing registry inventory, preview or dry-run APIs are usable.",
        "One registry engineer and limited release, security and service-owner review are sufficient.",
        "No production artifact is quarantined, archived or deleted.",
        "Software, log storage and test-repository charges are minor relative to labor.",
        "This is a resource-equivalent band, not a vendor quote."
      ],
      "source_ids": ["S2", "S4", "S5", "S7"]
    },
    "initial_deployment_startup": {
      "band_2026_usd": "50K_TO_250K",
      "scope": "Build or configure the external-reference and hold connectors, lifecycle state store, quarantine/tombstone controls, access control, audit integration, restore-test harness, monitoring and policy documentation for a limited production scope.",
      "confidence": "LOW",
      "assumptions": [
        "Native registry lifecycle and archive capabilities are reused where available.",
        "One or two deployment or rollback systems are integrated initially.",
        "Ambiguous references default to retention.",
        "Labor, security review, compliance coordination, software and test infrastructure are included.",
        "No registry replacement or enterprise-wide dependency graph is included."
      ],
      "source_ids": ["S2", "S3", "S4", "S5", "S7", "S8"]
    },
    "operational_launch": {
      "band_2026_usd": "50K_TO_250K",
      "scope": "Backfill lifecycle metadata, validate permissions and holds, train operators, execute production shadow mode followed by reversible quarantine in selected namespaces, test recovery procedures and evaluate launch outcomes; hard deletion remains a separately authorized phase.",
      "confidence": "LOW",
      "assumptions": [
        "The differentiation study meets preset benefit and safety thresholds.",
        "Launch is limited to selected namespaces and artifact classes.",
        "Archive restores meet the locally agreed recovery objective.",
        "Labor, coordination, compliance review, training, software and evaluation are included.",
        "Unresolved dependencies and holds remain active vetoes."
      ],
      "source_ids": ["S2", "S3", "S4", "S5", "S8"]
    },
    "annual_recurring": {
      "band_2026_usd": "50K_TO_250K",
      "scope": "Operate connectors and policies, review holds and exceptions, maintain tombstones and audit records, investigate dependency ambiguity, perform periodic restore tests, update integrations, monitor outcomes and conduct governance reviews for a limited multi-team deployment.",
      "confidence": "LOW",
      "assumptions": [
        "Routine registry-side evaluation is automated.",
        "Cross-system integrations and exception reviews require continuing engineering and stakeholder time.",
        "Archive storage, retrieval, logging, software and evaluation resources are included.",
        "The band excludes a very large enterprise, intensive litigation holds or major certification work.",
        "Storage savings are not netted against recurring resources."
      ],
      "source_ids": ["S2", "S3", "S4", "S5", "S7", "S8"]
    }
  },
  "verified_pipeline_gates": {
    "externally_supported_problem": {
      "status": "YES",
      "reason": "Independent official registry documentation confirms that accumulated images and versions create storage and listing burdens and motivate differentiated cleanup policies.",
      "source_ids": ["S1", "S5"]
    },
    "externally_credible_adopter_or_authorizer": {
      "status": "YES",
      "reason": "Registry Maintainers, Owners and administrators are documented authorities for usage, cleanup and garbage-collection controls; external holds would still require locally designated legal or compliance authority.",
      "source_ids": ["S1", "S5"]
    },
    "distinct_testable_incremental_claim": {
      "status": "YES",
      "reason": "The surviving claim can compare native registry lifecycle results with an enriched external-dependency and hold gate on validated unsafe candidates, false blocks, stale exposure, restore performance and operator effort.",
      "source_ids": ["S2", "S4", "S5", "S7", "S8"]
    },
    "bounded_next_evidence_step": {
      "status": "YES",
      "reason": "Lifecycle preview and dry-run capabilities support a 30-day, no-deletion comparison, while copied artifacts can be archived and restored in an isolated repository.",
      "source_ids": ["S2", "S4", "S5"]
    },
    "no_unresolved_safety_or_authority_stop": {
      "status": "YES",
      "reason": "For the proposed evidence step, production state is unchanged, hard deletion is prohibited, ambiguous references are retained and restore testing is confined to copied artifacts. Hidden dependency coverage remains a study outcome rather than a safety stop.",
      "source_ids": ["S4", "S5", "S7"]
    },
    "credible_cost_scope_and_range": {
      "status": "YES",
      "reason": "Official product primitives bound the likely build-versus-configure scope, while broad bands explicitly include engineering, integration, compliance coordination, software, storage and evaluation. Production estimates remain low-confidence because organizational scale is unknown.",
      "source_ids": ["S2", "S3", "S4", "S5", "S7", "S8"]
    }
  },
  "next_evidence_step": "Run a 30-day read-only shadow study in one nonproduction namespace containing at most 500 superseded artifacts. Before starting, set minimum useful additional-dependency detection, maximum false-block rate, operator-time ceiling, digest-fidelity requirement and recovery-time bound. Compare the registry's native lifecycle preview with the same candidate set enriched by one deployment inventory, one rollback or provenance catalog and registered holds. Manually adjudicate a stratified sample of at most 50 candidates, and archive then restore at most 20 copied artifacts in an isolated test repository. Perform no production quarantine or deletion. Falsify the incremental claim if enriched gating finds no preset-material increase in validated dependencies or holds, exceeds the false-block or effort ceiling, fails to reduce the unsafe candidate shortlist, produces any digest mismatch or exceeds the recovery bound.",
  "blocking_evidence": [
    "A local inventory has not established material superseded-artifact volume, obsolete discoverability or storage and maintenance burden.",
    "No preset acceptance thresholds exist for additional validated dependency detection, false blocks, operator effort, restore fidelity or recovery time.",
    "Coverage of deployment definitions, rollback catalogs, provenance systems and external digest consumers is unknown.",
    "The accuracy, ownership and synchronization method for legal, investigation, compliance and release holds are unverified.",
    "No isolated restore sample has demonstrated digest fidelity and acceptable recovery latency for local artifact classes.",
    "Incremental benefit over native registry lifecycle preview and archival—not merely age-only cleanup—has not been demonstrated."
  ],
  "research_disposition": "PRIOR_ART_DIFFERENTIATION_STUDY",
  "world_novelty_boundary": "This bounded search of official AWS, Google, GitLab, Kubernetes, OCI and related documentation found substantial prior-art collision, especially Amazon ECR lifecycle policies plus ECR Archive. It did not find, in the reviewed sources, one documented system combining external deployment and rollback dependency vetoes, authority-controlled holds, reversible quarantine with tombstones and periodic digest-verified restoration. That absence is limited to this search and is not a world-novelty claim."
}