Separate software, hardware and physics allocations and publication migration.
ZL512 Successor Corpus
Status: normative preview; allocation and migration specified, corpus unavailable
Implementation note: an unpublished ZL512 0.1.0 inventory and content-locked
ZL256 import now exist in benchmarks/zl512/. The zlbench successor command
validates their integrity and lists all six unavailable profiles. Executable
Core-2 cases, result/publication schemas and verified evidence remain future
work; inventory validation is not benchmark conformance.
C2-BENCH-001 — Separate successor. ZL512 contains 512 task slots: the existing 256 software tasks plus 128 hardware and 128 physics tasks. It is a new suite identity and publication namespace, not a rename or expanded version of ZL256. Freeze the imported ZL256 0.2.0 baseline by source catalog, task, fixture and authority digests at the first ZL512 dataset seal. Preserve original task IDs and bytes in that imported baseline; later maturity changes require a new dataset identity. Existing ZL256 versions, results and routes remain valid. This specification does not change the current catalog or publish a ZL512 seal.
C2-BENCH-002 — Hardware allocation. Category counts are normative. A task has one primary category, with secondary tags permitted without double counting.
| Category | Tasks | Required evidence families |
|---|---|---|
| syntax/elaboration | 16 | finite hierarchy, parameters, ports, rejection and modality |
| combinational/logic | 24 | truth tables, widths, resolution, arithmetic and order independence |
| clocked/reset | 24 | simultaneous state, initialization, reset priority and multiple clocks |
| CDC | 16 | valid bridges, incoherent bus rejection, latency and reset coordination |
| asynchronous | 16 | feedback, delay models, hazards, fork assumptions and progress |
| intrinsic | 8 | measured devices, validity envelopes, held-out population and invalidation |
| realization | 8 | observation relations, all software domains, claim rejection and promotion |
| synthesis | 8 | independent materialization, equivalence, target constraints and tamper |
| fabric | 8 | authority, configuration/readback, indeterminate actions and receipts |
| Total | 128 |
C2-BENCH-003 — Physics allocation. The physical task allocation is:
| Category | Tasks | Required evidence families |
|---|---|---|
| quantities/units | 12 | dimensions, affine conversion, shapes and frames |
| meshes | 12 | topology/orientation, boundaries, field placement and remeshing |
| equations | 16 | strong/weak forms, DAE/PDE constraints and analysis admission |
| electromagnetics | 12 | electrostatics, Maxwell, charge/energy and gauge |
| thermal | 8 | conduction, flux, transient energy and stability |
| mechanics | 8 | strain/stress, rigid modes, force and energy |
| fluids | 8 | continuity, pressure, momentum and flux |
| semiconductor | 8 | Poisson/carriers, contacts, positivity and conservation |
| coupled models | 8 | field projection, synchronization and coupling convergence |
| digital/physical bridges | 8 | sampling, thresholds, quantization, loading and uncertainty |
| SolverPlans | 16 | generated methods, resource bounds, independent comparison and promotion |
| evidence | 12 | residual/error distinction, convergence, uncertainty and stale evidence |
| Total | 128 |
C2-BENCH-004 — Task and case contract. The successor schema retains
software task identity and adds model-kind, preview, realization profile,
required features/capabilities, target/solver requirements, assumptions,
observation contract and evidence class. Physical models use model-kind
physical, not a fabricated callable domain. Cases bind complete source/model
closures, fixtures, exact expected diagnostics or typed observations with
tolerances, and expected terminal state. Each positive case has a boundary,
invalid or dependency-failure counterpart. Property cases cover canonical
ordering, truth-table laws, simultaneous updates, mesh projections and
conservation where applicable. Scaffold, executable and verified maturity
remain distinct; a parsed example is not an executed or verified task.
C2-BENCH-005 — Profiles and publication. Preserve the ZL256 schema and
current/roadmap views. Add separately versioned ZL512 catalog, result, public
projection and executor schemas when fixtures become executable. Its profiles
are hardware-reference, hardware-independent, intrinsic-device,
physics-reference, physics-independent and mixed-coupled. Each profile
declares target/solver identities, feature matrix and resource envelope;
unsupported tasks remain visible as unsupported, not silently omitted passes.
Do not compare simulator throughput with real-device throughput as one metric.
Publish software, hardware and physics completeness separately. Historical
results retain their original suite, dataset and toolchain identities.
Device evidence requires calibration, target identity, repeats and the stated environment; physical evidence requires applicable convergence/error checks. Statistical tasks use held-out fixtures/devices and declared confidence rules. Mutation evidence maps each non-equivalent mutant to the behavior and task that kills it. Dataset infrastructure changes must be tested through public executor/publication boundaries before publication is enabled.