{"submission_id":"c2571e6d-d5ad-4997-a575-d303638b7a01","submission_revision_id":"6f95d1a3-df22-482b-9ef0-c9047530c58e","revision_version":1,"submission_status":"under_review","submitted_at":"2026-09-04T09:45:57.617216Z","submitter_name":"Yizhao Chen","submitter_email":"yic138@ucsd.edu","task":{"status":"submitted","version":1,"extra_fields":{"difficulty_self_test":{"model":"GPT-5.5 (2 low-effort and 2 medium-effort attempts)","harness":"Codex Desktop with isolated Linux/amd64 containers","estimated_score":"<1%"}},"submitted_at":"2026-09-04T09:45:57.617216+00:00","industry_domain":"Computing, Data & Mathematical Sciences / Data & Analytics Engineering / Database & Data Infrastructure","task_short_name":"postgresql_regional_failover_reconciliation_1","operating_system":"Linux","task_description":"Objective:\nRecover three related PostgreSQL 17.11 databases after a failed regional failover: one coordinator and two ledger shards.\n\nStarting state:\nEach physical backup begins on timeline 1. Real promotions produced canonical timeline-2 branches, while delayed fencing left obsolete timeline-1 branches writable. A later emergency replay corrupted each canonical cluster at a different committed transaction. The provided materials include physical base backups, archived WAL and timeline history, Patroni and lease evidence, transaction and clock records, administrative approvals, processor acknowledgements, partitioned broker deliveries, versioned event-store records, schemas, recovery policies, replay helpers, and the faulty migration/job artifacts.\n\nInstructions and rules:\n1. Verify the exact input inventory and read all contracts and replay-source documentation.\n2. Correlate WAL/history, leader leases, transaction state, deployments, approvals, acknowledgements, and clock offsets. Raw wall-clock order is not authoritative by itself.\n3. Independently recover the canonical coordinator, shard A, and shard B timeline-2 branches to defensible transaction-exclusive targets before each unsafe emergency replay.\n4. Recover the obsolete coordinator timeline-1 branch separately and salvage only configuration changes satisfying every visible approval, commit, lease, fence, payload, and supersession predicate.\n5. Resolve in-doubt cross-shard transfers, then replay legitimate post-target application events and processor-only acknowledgements exactly once using per-partition order, aggregate versions, dependencies, and the provided service source.\n6. Validate schemas, opaque pre-target state, sequences, global event identity, cross-shard state, acknowledgements, and per-currency double-entry invariants.\n\nDeliverables:\nWrite exactly seven regular files at the root of output/: coordinator.dump, shard_a.dump, shard_b.dump, branch_salvage.csv, event_reconciliation.csv, recovery_manifest.json, and incident_report.json.\n\nConstraints:\nTreat input/ and software/ as read-only. Use PostgreSQL from /opt/postgresql-17.11/bin. Follow the visible output contract exactly. Do not use network access, sudo, system services, or Docker. This is a controlled, fully synthetic disaster-recovery drill and does not disclose a real production incident.","software_licensing":"Free/Open Source","evaluation_criteria":null,"verification_method":"Use the supplied maintainer-controlled scorer and ten-file hidden reference. Candidate output must contain exactly seven regular files with the required names. Missing, extra, symlinked, aliased, oversized, corrupt, unsafe, or unloadable output scores 0.\n\nEach dump must be a PostgreSQL 17.11 custom archive containing only the allowed schemas and objects. The evaluator rejects functions, triggers, extensions, foreign/server objects, subscriptions, candidate code, and candidate DDL. It creates fresh private PostgreSQL clusters with TCP disabled, loads evaluator-controlled trusted schemas, and restores allowlisted data and sequence values as a non-superuser.\n\nWeighted scoring:\n- 8% safe, loadable archives and canonical schema inventories\n- 12% coordinator semantic state\n- 15% shard A semantic state\n- 15% shard B semantic state\n- 10% obsolete-branch salvage decisions and imported state\n- 18% broker, processor, aggregate-version, dependency, and exactly-once event graph\n- 12% cross-cluster transfer, ledger, sequence, migration, and global-identity invariants\n- 10% truthful CSV/JSON sidecars consistent with the candidate databases\n\nRaw-score caps:\n- 0.20 for a wrong timeline or retained emergency-replay damage\n- 0.25 for retained V052 schema or numeric damage\n- 0.45 for an unresolved or contradictory committed cross-shard transfer\n- 0.50 for an invalid obsolete-branch import or required salvage omission\n- 0.55 when fewer than 95% of required event states are exact\n- 0.60 for an unbalanced journal, duplicate global event identity, or unsafe sequence\n\nPerfect output requires all three canonical databases, every required opaque/WAL-only cohort, exact salvage and event decisions, all required post-target state exactly once, no faulty migration or emergency-replay damage, and all local/global invariants true. Logical state is compared rather than dump bytes, OIDs, or permitted implementation-chosen surrogate UUIDs. Missing or stale hidden references, trusted schemas, scorer, or exact PostgreSQL runtime are evaluator infrastructure failures rather than candidate failures.","software_and_version":"PostgreSQL 17.11; Python 3 standard library; GNU tar, gzip, and coreutils","specific_requirements":null,"expected_completion_time":null},"files":[{"file_id":"dc4ee36f-b354-402b-99bc-5eca500d97e5","filename":"postgresql_regional_failover_reconciliation_input_materials.zip","category":"input_materials","content_type":"application/zip","size_bytes":22134849},{"file_id":"c4b89254-2b22-4f28-8ab4-9f25f5d0ed39","filename":"postgresql_regional_failover_reconciliation_reference_evaluation.zip","category":"reference_output","content_type":"application/zip","size_bytes":7167555},{"file_id":"2093a4b1-e3ab-44d8-a104-0693271155bf","filename":"AUTOMATED_QC.md","category":"difficulty_self_test_evidence","content_type":"text/markdown","size_bytes":10588},{"file_id":"952170cc-b9c7-449c-a135-e8e806441dd4","filename":"CALIBRATION_RESULTS.md","category":"difficulty_self_test_evidence","content_type":"text/markdown","size_bytes":7426}],"deep_review":{"overall_verdict":"Strong Accept","review_text":"**Overall Verdict:** Strong Accept — Complete, coherent, professionally specified, objectively evaluable, and reachable without material defects.\n\n**Completeness Check:**\n- Instruction: Complete — Clear recovery objectives, evidence precedence, seven-file output contract, safety constraints, and validation requirements.\n- Inputs: Complete — Physical backups, WAL/history, evidence, contracts, replay source, schemas, and faulty artifacts are present and hash-verified.\n- Toolchain: Complete — PostgreSQL 17.11, standard Python, tar/gzip, and coreutils are identified with a reproducible package path and licensing.\n- Reference output: Complete — The reference archive contains the seven gold outputs, trusted schemas, truth data, and documented evaluator assets.\n- Rubric: Complete — Weighted components, hard gates, caps, invariant checks, and infrastructure-failure handling are explicit.\n\n**File and Task Reachability:**\nAttachments are `reachable`; the end-to-end task is `reachable`. I inspected both ZIP inventories, all Markdown reports, input contracts, evidence files, compressed broker/event data, physical tar structures, reference CSV/JSON/schema artifacts, package metadata, and evaluator source. ZIP CRCs, gzip streams, JSON/NDJSON parsing, SHA-256 input records, tar paths, member types, and duplicate-name checks passed. No attachment was inaccessible or left structurally unreviewed; large truth and event data were reviewed through bounded samples and summaries. The local review runtime lacked PostgreSQL binaries, so custom dump restoration and TOC parsing were not performed; this is a reviewer-tooling limitation, while the supplied evaluator path pins PostgreSQL 17.11.\n\nThe solver-visible input/reference boundary is documented and enforced by the lifecycle source: references are withheld during solve and staged only for evaluation. The physical inputs expand to approximately 1.46 GB, which is reasonable for the declared single-machine environment. The evaluator provides an objective path from seven output files to isolated PostgreSQL restoration and deterministic scoring.\n\nLicense keys, accounts, tokens, and other credentials are not required from the contributor.  \nConfidential, proprietary, or internal material may be evaluated as submitted evidence without requiring additional disclosure.\n\n**Special Screening:**\n- `synthetic_defect_fixture` — The task explicitly centers on predetermined faulty migration and emergency-replay defects, with the unsafe SQL marked as evidence-only. This is an intentional synthetic recovery fixture, not a submission defect.\n- No affirmative evidence of AI-originated ground truth or a publicly identifiable benchmark was found. Public-search verification was unavailable.\n\n**Recommended Changes:**\nNo material task changes are required. Separate portal-side provenance, rights, authorship, and signature fields noted in the QC documentation remain administrative follow-ups.","reviewed_at":"2026-09-04T09:59:50.163876Z"},"claim":null}