SkillChain-Gym benchmark: Reskilling-aware production control

SkillChain-Gym, introduced on 15 June 2026 by Carlos Eduardo Sanoja, is a benchmark specification for reskilling-aware production-inventory control that models worker skills as decision variables and embeds training actions into the same per-worker time budget as production. The arXiv paper (arXiv:2606.17266) frames certification thresholds, skill forgetting, and capacity-consuming reskilling inside seed-controlled disruption scenarios and deterministic replay.

What is SkillChain-Gym?

SkillChain-Gym is a single-site environment with stylized worker skill-state dynamics, hard threshold certification, forgetting, and training actions that consume the same per-worker time budget as production. The benchmark specification includes seed-controlled disruption scenarios, three feasibility modes with projection diagnostics, deterministic replay, and metrics covering operations, resilience, capability growth, and training-access distribution.

The environment treats workforce capability as a decision variable rather than an exogenous input: certifications lapse when skills are not maintained, new products may require skills the current workforce lacks, and reskilling competes with production hours. That structure is intended to make workforce planning and production-inventory control interact directly inside a reusable testbed.

How did the benchmark evaluate policies?

The paper evaluates four policy classes—production-only, reactive adaptive, water-filling adaptive, and static-insurance—each tested with budget variants over 60-shift horizons using paired statistical tests. Evaluations run across seed-controlled disruption scenarios and compare outcomes on operations, resilience, capability growth, and training-access distribution, with deterministic replay and projection diagnostics to probe feasibility.

Results are "regime-dependent rather than a ranking," the paper states, meaning no single policy class dominates across all tested conditions. The authors report that training-capable policies dominate the production-only baseline, and that maintenance training is necessary under forgetting even without disruptions. Within training-capable approaches, adaptive training helps when bottlenecks are visible in the forecast, while a deliberately designed lean static cross-training plan acts as strong insurance under surprise shocks and absenteeism.

Why it matters

The benchmark forces decision-makers to trade off production hours for training because reskilling consumes worker time and certifications can lapse, explicitly linking workforce capability to inventory and production outcomes. That coupling changes which controls work best: the paper finds maintenance training is required under forgetting, and that the benefit of adaptive versus static cross-training depends on visible bottlenecks, capacity slack, and the forgetting rate.

Practitioners and researchers gain a reusable environment that makes those trade-offs testable. The inclusion of seed-controlled disruptions and deterministic replay gives a repeatable platform for paired statistical tests, so claims about policy superiority can be evaluated across regimes rather than asserted from a single scenario.

What to watch

Watch how capacity slack and the forgetting rate shift the boundary between regimes where adaptive training outperforms static insurance and where a lean cross-training plan is the safer hedge. The paper points toward forecast-driven controllers that decide when to buy skill insurance and when to react, making the next milestone the design of controller heuristics that incorporate forecast uncertainty and the documented feasibility diagnostics.

References

Carlos Eduardo Sanoja, "SkillChain-Gym: A Benchmark for Reskilling-Aware Production-Inventory Control under Disruptions," arXiv:2606.17266, submitted 15 Jun 2026.