## v1 Update — 2026-09-27 This update consolidates the dataset, reference results, validation logic, and evaluation protocol developed since September 5, 2026. ### Canonical dataset - FrontierOR now contains **180 canonical tasks**, organized under `tasks//`. - The legacy `ostrowski2012` alias was removed because it duplicated `earl2005`. - `wang2018` was added to complete the canonical 180-task collection. - Paper-level metadata, including optimization direction, is consolidated in: - `metadata/paper_meta_info.json` - The canonical 50-task hard subset is published in: - `metadata/hardset/hardset.json` - Its selection criteria, frozen input snapshot, full 180-task ranking, and reproduction script are published in `metadata/hardset/` (see `metadata/hardset/hard_methodology.md`). ### Canonical Gurobi references Canonical Gurobi objectives, benchmark runtimes, time limits, and statuses are now distributed in compact machine-readable files: - `metadata/gurobi_references.parquet` - `metadata/gurobi_references.csv.gz` These files are the authoritative source for metric computation. They replace the need to reconstruct runtimes from heterogeneous solution fields, incumbent logs, or legacy per-instance CSV files. ### Hardened feasibility checkers The feasibility checkers were systematically audited and strengthened: - Checkers now validate submitted solutions against trusted instance data and the published solution contract. - Objective consistency checks reject missing, inconsistent, or fabricated objective values. - Malformed JSON, missing required fields, invalid field types, and variable-domain violations are handled as explicit infeasible results. - Capacity, flow, balance, linking, integrality, and task-specific constraints are checked where applicable. Geometric checks use numerically stable calculations to avoid overflow-based false acceptance. - Checkers no longer trust fixed problem parameters supplied by a candidate solution. - The affected solver-dependent checkers were converted to validate complete witnesses without invoking an optimization solver at evaluation time. - Checker execution limits and error handling were made more robust for large solutions. All published Gurobi reference solutions were revalidated using the updated checkers. ### Task and instance corrections We audited the task contracts, instances, reference implementations, and solutions. Notable corrections are grouped below by the underlying fix principle: - **Schema/checker contract alignment.** `carosi2019`, `kang2021`, `riley2025`, `vidal2011`, `luo2017`, `rothenbacher2018`, and `engineer2012` now expose the checker-facing solution conventions in `solution_schema.json`, including public numbering, virtual depot/sink nodes, expanded stop identifiers, time-indexing conventions, and required solution-key formats. - **Complete-witness variable naming.** `adulyasak2015`, `bertsimas2024`, `liao2020`, `rahmaniani2018`, and `rahmaniani2022` now document canonical complete-witness variable names or variable-name generation patterns in `solution_schema.json`, so submissions do not need to infer task-specific Gurobi variable names. - **Checker contract and robustness fixes.** `delage2022` now supports the open-ended, closed-ended, and knife-dependent setup variants with variant-specific fields and constraints. `bierwirth2017` now derives operation completion times from trusted instance processing times and no longer accepts an unsupported alternative solution schema. `belvaux2000` and `cordeau2019` received checker consistency fixes. `oliveira2020`, `ozturk2020`, `pecin2017`, and `spliet2015` received hardened malformed-solution handling. - **Task instance and reference artifact repairs.** `segundo2019` now includes the missing `large_3` and `large_4` instances and matching Gurobi artifacts. `wei2020b` has a restored non-empty `large_1` reference solution. `oliveira2020` has a regenerated tiny instance at an appropriate benchmark scale with updated derived metadata and reference result. `fragkos2021` has a corrected historical Gurobi reference affected by callback/runtime handling. `wang2018` was added to complete the canonical 180-task collection and then standardized with complete runtime metadata. - **Canonical instance-set cleanup.** Redundant second tiny instances and their derived Gurobi artifacts were removed from `armbruster2012`, `bierwirth2017`, `borndorfer2007`, `engineer2012`, `guo2021`, `huisman2005`, `nagy2015`, `oliveira2020`, `ozturk2020`, `pecin2017`, `potthoff2010`, `ropke2009`, `rothenbacher2018`, `ruiz2007`, `segundo2019`, `spliet2015`, and `vidal2011`, leaving the canonical one tiny plus five large instances per task. - **Gurobi runtime and fallback consistency.** `furini2016` and `segundo2019` have repaired zero-incumbent fallback handling in the reference Gurobi programs. Runtime start-time initialization was standardized in `bode2012`, `bodur2017`, `bragin2022`, `buchheim2018`, `chen1999`, `cote2018`, `furini2021`, `garcia2011`, `hoffman1993`, `kowalczyk2018`, `kreter2018`, `letelier2022`, and `roshanaei2017`. Associated reference solutions and feasibility results were regenerated or revalidated where required. ### Compatibility notes Users of earlier FrontierOR snapshots should: - update task paths to `tasks//`; - use `metadata/paper_meta_info.json` for paper metadata; - use `metadata/hardset/hardset.json` for the hard subset; - use `metadata/gurobi_references.parquet` or `.csv.gz` for Gurobi references; - rerun feasibility checks when relying on results produced by older checker versions. ### Future direction for v2: trusted anytime evaluation The current incumbent logs are intentionally treated as untrusted for formal anytime metrics. In many evaluated programs, the log records only an incumbent objective value and does not include the complete decision-variable assignment required by the task's `solution_schema.json`. As a result, if a program finds and logs an incumbent but later fails to terminate correctly, the host cannot verify that intermediate incumbent for feasibility or objective consistency. This also means AOCC computed from objective-only incumbent logs would not be checker-validated. This version still keeps these log files for inspection and debugging, but temporarily removes AOCC-style metric computation because incumbent logs are not yet trusted evidence of feasible intermediate solutions. A natural next step is to extend `SolutionLogger.log_solution(objective, solution)` so algorithms can report complete incumbent solution snapshots. This is nontrivial: not every algorithm can construct a complete feasible solution before termination; task schemas differ substantially; some search procedures maintain only internal state or objective values until final reconstruction; Gurobi callbacks may make full variable extraction expensive; repeated serialization of large solutions can add runtime, memory, and disk overhead; and oversized logs need bounded storage and validation policies. The planned approach is to support trusted snapshot logging first on a suitable subset of tasks. For those tasks, algorithms will log complete snapshots when a full incumbent solution is available, the trusted host will attach timestamps outside the candidate program, and each bounded snapshot will be checked with the task's feasibility checker and objective consistency checks. Only snapshots that pass validation will be used to reconstruct anytime curves or compute AOCC. Tasks or algorithms that cannot reliably emit complete intermediate solutions will remain eligible for final-solution feasibility, solution-quality, and binary QTE evaluation, but not for trusted snapshot-based AOCC.