jcurtiswolf

Execution-graded competency at the uv package manager (init/add/remove/venv/pin), verifiable

Execution-graded competency at the Django framework (project/app/migrations/check), verifiable

Latent (no-CoT) two-hop reasoning over real Wikidata facts (Balesni et al. 2024), model-graded

Adjudicate a medical claim to the plan-paid amount and denied lines (verifiable, RLVR)

Adjudicate insurance claims to a verifiable settlement amount (multimodal-ready, RLVR)

Compute reorder point, order decision, and EOQ for a SKU (verifiable, RLVR)

Three-way match an invoice against a PO and received goods, compute the approved payment (verifiable, RLVR)

Pick the best feasible carrier for a freight load (verifiable, RLVR)

Compute US federal income tax (ordinary brackets, preferential capital gains, child tax credit phaseout, FICA) owed or refunded for a synthetic tax...

Decide approve or deny on a consumer loan and compute the max approved amount (verifiable, RLVR)

Compute a verifiable financial metric (liquidity, leverage, profitability, cash flow) or recover a missing line item via multi-step identities from...