Tasks

One workflow that solves both maths and statistics problems

Size every bar of a transmission tower

Retrieve exoplanet atmospheres from transmission spectra

Localize stiffness loss in a steel structure

Prove Lean theorems under a per-problem inference budget

Predict a molecule's HOMO-LUMO gap from its SMILES string

Classify transients and variables from their light curves

Reimplement a colour-code decoder without the usual libraries

Design 128 noisy shallow-circuit measurements

Propose an equation family that extrapolates, not just fits

Plan the contraction order for exact circuit equivalence checking

Build a filtered ANN index over 10M real vectors

Extrapolate power-device I–V curves outside the safe window

Variable-length attention with the full backward pass

Gated long convolutions across concurrent streams

Relocate, resize and buffer a placed netlist for better PPA

Plan join orders from noisy selectivity estimates alone

Mamba-3 scan with joint states and a full backward pass

Bank and place multi-bit flip-flops at minimum weighted cost

Decode attention over a paged, ragged KV cache

INT4 weight-only GEMM with a fused inference epilogue

Train with GRPO so it survives a compositional distribution shift

Lift a ReAct harness on the tasks where it scores worst

Redesign a conversational memory system so it generalizes

Spend a fixed evidence budget across opaque branches

Predict the compute-optimal run from smaller ones

Post-train a 1.7B student with a fixed 8B teacher

Choose which frames to look at under a fixed evidence budget

Reconnect broken road graphs from aerial evidence

Fit a ten-evaluation generative solver

Drop Gaussians to a budget while holding novel-view fidelity

Keep one semantic edit correct across nuisance variation

Keep paraphrased edit requests mutually consistent

Select and fuse point-track candidates over a lattice

Route point tracks through candidates under a switch budget

Post-process action proposals with no retraining allowed

Recover consistent camera poses from noisy correspondences

Fuse noisy multi-view observations into one point cloud

Tune a five-pass temporal latent refinement policy

Reconstruct a hidden maze through an interactive judge

Drive a robot with a record-once, replay-many macro button

Optimize a noisy multimodal function on a fixed query budget

Tune an (s,S) inventory policy through a stochastic simulator

Ship a self-contained ask/tell optimizer

Minimize conflicts at a fixed color budget on hard DIMACS graphs

Beat Chvátal-greedy on hidden set-covering instances

Schedule cyclic events with standard library only

Fly a quadrotor through gates in minimum time

Cut glass orders from the fewest stock plates on a guillotine line

Choreograph five terminal cranes that cannot pass each other

Predict base-resolution chromatin accessibility from DNA

Annotate lung cell types from five labeled examples

Integrate scATAC batches into one clean embedding

Predict expression from chromatin accessibility alone

Train and ship a protein-stability regressor

Co-fold protein and ligand, inference only

Retrieve protein surfaces by shape descriptor

Annotate cells across datasets with a frozen foundation model

Match TCRs to epitopes never seen in training

Triage chest X-rays under a fixed resource cap

Fit a joint IRT-NLME model in pure R

Design a multi-agent medical QA workflow under a call budget

Decide when a medical VQA model should abstain

Choose which links to widen and how to price them

Retime ~600 signalised junctions across a whole city

Grow a rail network that pays for its own next track

Dispatch robotaxis under two-sided time windows

Route electric vans that must detour to charge

Choose the gear every second to burn less fuel

Race, overtake and defend against scripted opponents

Park an Ackermann car in a tightly constrained slot

Serve 500 paired pickup-delivery requests on time

Switch a traction inverter above base speed

Dispatch trains through breakdowns you cannot see coming

Place suspension hardpoints and pick the tuning that goes with them

Design a bus network that covers every node

Predict summer indoor heat from one mild-season record

Emulate sub-grid convection for a climate model

Recover 3D wind where the vertical component is underdetermined

Split power in a hybrid using only what the car can see now

Design a fast-charge current table that spares the cell

Repair phase labels and impedances on a distribution feeder

Invert seabed friction from sparse tide gauges

Classify lithology facies from wireline well logs

Place turbines so wakes cost the least energy

Forecast value drivers from a messy fundamentals panel

Fit an arbitrage-free vol surface from one noisy snapshot

Flag structural breaks in a fleet of series

Predict when a world model is about to hallucinate, without labels

Give a manipulation policy usable memory

Learn to push a cube from a fixed offline dataset

Improve a white-box optimizer that breaks a safety classifier

Rebuild the Dart formatter in Haskell, byte-for-byte

Port libexpat's XML core to hand-written x86-64 assembly

Reimplement the PostgreSQL server over SQLite storage, in Zig

Play 2048 well on seeds you have never seen

Run a transport company for 20 simulated years

Solve Sokoban where rooms nest inside themselves

Survive Battle Tetris with holds, combos and garbage

Distil a reusable memory of a GDPR engagement

Distil a reusable memory of a telemarketing-regulation matter

Answer banking questions from the underlying deal files

Answer consulting questions from the engagement files

Work regulatory legal matters across 1,000+ case documents

Work transactional legal matters across 745 case documents