RLTasksby ulam.ai
Sign inAll tasks
Operations & simulationsA3-OptiOps

A3-OptiOps

A 12-stage proof-carrying optimization-engineering benchmark built around a realistic GPU-cluster scheduling service. Agents extract policy semantics, build scalable anytime solvers, react to failures and uncertainty, diagnose infeasibility, preserve production state, implement repository-scale protocol changes, resist adversarial evaluation, transfer to unseen constraints, and demonstrate optimizer improvements.

Decision makingStateful environmentRLVR
Version2.0.0
Environments12
RewardScalar · 0–1
DeliveryPrivate ZIP
01 Task contract

A real, versioned RL environment.

A 12-stage proof-carrying optimization-engineering benchmark built around a realistic GPU-cluster scheduling service. Agents extract policy semantics, build scalable anytime solvers, react to failures and uncertainty, diagnose infeasibility, preserve production state, implement repository-scale protocol changes, resist adversarial evaluation, transfer to unseen constraints, and demonstrate optimizer improvements.

Agent objective

  • Interact with the supplied stateful environment.
  • Produce verifier-checkable actions or artifacts.
  • Maximise scalar reward under the package contract.

Evaluation

  • 12 verifier-backed environments.
  • Reported reward range 0–1.
  • Package-specific public and private checks.

Delivery boundary

  • Private object stored in Cloudflare R2.
  • Authenticated entitlement required.
  • Short-lived signed URL per download.
02 What you will work on

Distinct environments, one demanding research contract.

Enough detail to understand the intellectual terrain; generated instances, hidden mechanisms, and solution paths remain inside the private package.

EnvironmentMathematical or technical frontierAdaptive research problem
Stages 0–2Model extraction, scalable optimization, dispatchCompile business policy into a coherent model, solve at scale, and select algorithms according to structure rather than hard-coded size.
Stages 3–5Anytime, online, and robust schedulingImprove feasible schedules across budgets, replan after failures and arrivals, and balance expected cost with tail risk.
Stages 6–7Diagnosis and production stateProve minimally harmful feasibility repairs and support concurrent sessions, snapshots, restoration, migration, and deterministic replay.
Stages 8–10Repository change, adversarial robustness, transferExtend accelerator protocols without regressions, resist decoys and tampering, and handle unseen constraint families through plugins.
Stage 11Algorithm discoveryImprove an already strong portfolio solver and preserve the speed–quality gain under hidden families and scale shifts.
03 Why it is interesting

What the supplied evaluation reveals.

The evaluated patch produced valid submissions on all 12 stages and averaged 0.7303 privately, passing five. Public performance averaged 0.8425, but several stages fell sharply on hidden cases—model extraction dropped from 0.7243 to 0.1858. That public/private gap makes the suite useful for studying overfitting, operational shifts, persistent consequences, and real transfer. Validation reports 25 package tests and 84 workspace contract tests.

We publish aggregate behavior and task structure, while withholding generated instances, hidden labels, exact successful probes, private checks, and solution trajectories.

04 Supplied evaluation

Observed evaluation result.

Shown with its provenance and limitations; it is not a performance guarantee.

i
Methodology matters

One fresh hidden micro-tier evaluation per stage; 5/12 stages passed and all 12 submissions were valid.

Evaluated system / policyGPT-5.6 Pro-assisted agent

As identified by the supplied artifact.

Mean private score0.7303

12 reported runs.

Result artifactIncluded

score_report.json

Public result record

Machine-readable provenance and the exact displayed metric are available in results.json.

Open result JSON
05 Private delivery

The package stays off the public website.

The paid ZIP will live in a private R2 bucket. Vercel authorizes the buyer and issues a 2–5 minute object URL; R2 serves the bytes directly.

Included with purchase

  • Exact package version 2.0.0
  • Environment and task contracts
  • Verifier or scoring interface
  • Supplied reference/evaluation artifacts
  • Purchase record and licence v1.1
delivery flow
Authenticated buyer + entitlement check
+ private R2 object + 2–5 minute signed URL
= direct, auditable download

Package SHA-256
7f329daea1473507277e27ea34bf4e743f4d2dd1ffc02b43a380af156184456d
06 Licence v1.1

Commercial use, without exclusivity.

One purchase licenses this identified item to one legal organisation for worldwide, perpetual commercial model training, evaluation, research and development. Redistribution and resale of the package are not permitted.

Read the full licenceYotta Content LTD · business customers only
A3-OptiOps

Ready to add this environment?

Back to marketplace
Stripe checkout

Business purchase confirmation

Sign in or create an account, then complete secure Stripe Checkout. Access is granted only by the verified payment webhook.