All MicroEvals
You are a distributed systems architect specializing in mult...
Create MicroEval
Header image for You are a distributed systems architect specializing in mult...

You are a distributed systems architect specializing in mult...

Prompt

You are a distributed systems architect specializing in multi-party approval workflows and versioned protocol migration. I am upgrading "AI Visual Director Production" v7.8.0 Release 6 (the merged, verified tree) β†’ v8.0.0. WHAT ALREADY EXISTS IN VERSIONING (verified β€” this machinery is PROVEN and must be extended, never replaced): - schemaVersion on 35 shipped JSON artifacts under one convention (field at line 2). KNOWN LIVE DEFECT your design must catch and fix: templates/SEMANTIC_REVIEW.json still carries "7.4.0" inside a 7.8.0 tree. - templates/MIGRATION_RECORD.json β€” the shipped migration record template. - dev/UPDATE_WORKFLOW.md β€” the shipped version-update protocol. - release/V7_6_0_RELEASE_4_BASELINE_HASHES.json β€” the 293-file V7.6.0 Release 4 baseline: per-file SHA-256, immutablePaths, and retiredPaths where every retirement carries a reason and a successor; enforced by validators/validate_upgrade_compatibility.py, which FAILS CLOSED on undeclared path loss (CANDIDATE_BASELINE_PATH_MISSING) and on unjustified retirement (empty reason). - templates/ROLLBACK_PLAN.json + validators/validate_rollback_plan.py. - Retired-names-as-provenance doctrine: historical docs carry HISTORICAL banners pointing at their successors; they are never rewritten as current. THE REAL GAPS: - Approval: the rule 35 gate receipt has NO role fields β€” single approval authority per gate. templates/HUMAN_REVIEW.md, HUMAN_EVIDENCE_CAMPAIGN.json and HUMAN_EVIDENCE_SESSION.json exist for evidence campaigns, not per-gate role-scoped approval. No veto hierarchy, no conflict resolution. - Versioning: the existing machinery is RELEASE/PACKAGE-level; there is no PROJECT-level mid-production rule migration, no producer_version in handoff packets, no agent-side version detection on incoming artifacts. YOUR TASK: PART A β€” STAKEHOLDER APPROVAL: 1. Role registry: creative_director, client, legal, technical_qa (+ extension mechanism for custom roles) 2. Per-gate approval matrix over the real gates (R0 Scope/rights/capability, R1 Brief/mode/taste, R2 Originality, R3 Causality/mechanism, R4 Assets/ model sheets, R5 Board/timing/feasibility, R6 Execution channels, R7 Sound/edit/delivery, R8 Evidence QA): which roles approve, in what order, with what evidence 3. Role-specific evidence schemas: legal needs rights docs (rule 21 rights_and_claims); technical QA needs format specs; client needs creative brief alignment (rule 24 client_delivery) 4. Conflict resolution: veto hierarchy, escalation paths, timeout/default behavior 5. Approval state tracking: partial approvals stored and resumed across sessions and model switches (rule 19 handoff resets execution profile to CONTROLLED β€” approval state must survive that reset) 6. Delegation/proxy rules for approval authority transfer PART B β€” VERSION MIGRATION (v7.8.0 Release 6 β†’ v8.0.0): 1. Version metadata: where schema_version, rule_version, producer_version live in every artifact β€” extending the existing schemaVersion convention 2. Migration registry: deterministic transformation functions as field mappings, extending MIGRATION_RECORD.json and UPDATE_WORKFLOW.md, and reusing the retiredPaths reason+successor pattern β€” not a new format 3. Version detection: how agents identify the source version of incoming artifacts (must catch the stale-7.4.0 SEMANTIC_REVIEW case) 4. Hybrid operation rules: allowed/blocked during the transition window 5. Lossless upgrade guarantee: proof anchored to the baseline-hash mechanism (no information lost; retirements declared with reasons) 6. Rollback protocol extending ROLLBACK_PLAN.json + its validator BINDING CONSTRAINTS: - Approval state survives session breaks and model switches. - Migration is deterministic and reversible. - Stakeholder evidence is segregated (legal docs never leak into creative review). - Both systems serialize cleanly into handoff packets (rule 19). - No breaking changes to v7.8.0 compliance guarantees. - EVIDENCE HONESTY: stakeholder-layer completeness never flips promotionEligible β€” that stays false until the five live-evidence tracks pass; X14 β€” no score accepted without evidence. - HARD FACTS carried into every approval/migration artifact that touches video: 720p native ceiling with R0/R1 flags; three-edit cap; two-repair budget. - RULE 03: no stale spec tables; no hardcoded model specs. - PRIME DIRECTIVE: YOU DO NOT BUILD FROM SCRATCH, YOU UPDATE THE ALREADY BUILT. FILES YOUR ANSWER MUST EXTEND (name each edit explicitly): templates/MIGRATION_RECORD.json, dev/UPDATE_WORKFLOW.md, release/V7_6_0_RELEASE_4_BASELINE_HASHES.json (pattern, not content), validators/validate_upgrade_compatibility.py, templates/ROLLBACK_PLAN.json, validators/validate_rollback_plan.py, templates/SEMANTIC_REVIEW.json (stale version β€” fix), templates/HUMAN_REVIEW.md, templates/HUMAN_EVIDENCE_CAMPAIGN.json, templates/HUMAN_EVIDENCE_SESSION.json, rules/detailed/35_reliability_gate_controller.md (Β§4 receipt), rules/detailed/19_production_memory.md, rules/detailed/21_rights_and_claims.md, rules/detailed/24_client_delivery.md, dev/run_v780_release6_tests.py. OUTPUT SHAPE: 1. EDITS per named file (exact section + replacement text). 2. NEW FILES only where justified (the role registry and migration registry JSON are expected), with justification. 3. REGRESSION CHECKS continuing the catalog at R651+ (must include: a stale artifact version is detected; an undeclared retirement fails closed). 4. FORMAT: recommendation β†’ tradeoffs β†’ failure modes β†’ what to cut β†’ named integration points. Do NOT ask clarifying questions. Make definitive architectural decisions. Output complete role matrices, migration transforms, and state schemas.