Questions

What OHRQ is,and how it works.

Ten questions AI labs and manufacturers ask first: what each side gets, how answers are graded, what is in the benchmarks and the data, and how to start.

What is OHRQ?

OHRQ is one platform for physical intelligence in engineering and manufacturing. It serves two customers through one loop.

AI labs get benchmarks, evaluation as a service, reinforcement-learning environments and training data, all in physical engineering and manufacturing. Manufacturers get a factory system with an assistant that improves on their own work, plus engineering engines whose answers can be checked.

Each side supplies what the other lacks. Labs lack real, under-specified engineering work and real experts. Factories lack models that have been measured on that kind of work.

What does an AI lab get?

Four things, all in physical engineering:

  • Benchmarks. Versioned benchmark releases with measured floors and ceilings.
  • Evals. An evaluation server and batch grading on held-out tasks.
  • Gym. Reinforcement-learning environments and trainer adapters.
  • Data. Synthetic data products where every row is re-checked by a grader or solver.

The tasks are hard, the graders are objective, and the test sets are ones a model cannot have seen.

What does a manufacturer get?

A factory system, an assistant and engineering engines.

  • Plantline (factory system). Plant data, an assistant and an approvals inbox. It imports an ERP’s data and writes an accepted plan back.
  • Engineering engines. Physics, Process, Operations, Manufacturing, Knowledge, Experiments and Monitor, each giving answers you can check.
  • Learn. Traces, datasets, training and deployment.
  • Improve. A self-improvement loop where every change needs a signed-in person’s approval.

Nothing changes without a person’s approval.

How is an answer graded?

By a deterministic grader or a solver, never by a model. Every number in an OHRQ answer comes from a computation, never from a language model.

The same engines serve as tool, grader and data generator. The engine that answers an engineering question for a manufacturer is the one that grades a model’s answer to it and labels the data a lab trains on.

What is in OHRQ Bench?

Nine tracks, from CAD and G-code to plant operations:

  • Engineering (release 0.1). Design and manufacturing work with one checkable answer: CAD programs, G-code, PLC programs, wiring, tolerances, schedules, inspection plans.
  • Computer Use (release 0.1). Office and engineering-adjacent work on simulated computers; runs replay to the same state.
  • CAD Desktop. The same part-modelling tasks, solved by driving real CAD software from screenshots.
  • Factory Assistant. An assistant inside a running plant system, graded on outcomes.
  • Physics. Predict stress, temperature and surface pressure on shapes the model has not seen.
  • Process. Mass and energy balances, heat integration, pump and pipe sizing, equipment cost.
  • Operations. Scheduling and inventory decisions in a simulated plant.
  • Design. From a requirement to a part a solver and the manufacturing rules both accept.
  • Real Work. Tasks from real engineering work at companies that opted in.

Engineering and Computer Use are at release 0.1, with floors and ceilings measured.

How does the training data stay clear of the test sets?

Three rules apply to every data product:

  • Checked rows. Every row is re-checked by its source’s grader or solver in a fresh process before release.
  • No overlap with test sets. Anything matching a hidden benchmark task or a held-out engine task is dropped, so sold data cannot contaminate an evaluation.
  • Traceable deliveries. Deliveries use signed, expiring links, and each order can be traced if a copy leaks.

Each release has a card written from measured numbers, and a free sample pack. The catalogue has seven products: eng-sft, eng-pref, eng-traj, cu-transcripts, proc-qa, phys-records and mixed-curriculum.

Where does the real work come from?

From manufacturers who opt in. Real work, opted in and rights-cleared, becomes tasks, graders and data. Those train and evaluate models. Better models go back to work and produce more of it.

That is the loop between the two customers, and it is why OHRQ is one platform and not two.

Can the assistant change anything on its own?

No. Every change needs a signed-in person with a second factor, and it waits in an approvals inbox until that person accepts it.

Approval records cannot be forged or rolled back unnoticed.

What has been measured?

Every number comes from OHRQ’s own recorded runs:

  • Experiments passes 36 of 42 process-optimisation tasks; simple methods pass 4 to 12.
  • Monitor passes 76 of 102 diagnosis and quality tasks.
  • Machining passes 61 of 63 tasks on a seed it had not run before.
  • Knowledge finds a known item first for 91% of 120 queries; plain substring search finds 86%.

Frontier agents already pass 32 of 35 synthetic engineering tasks in pilot runs, which is why synthetic-only benchmarks saturate and why OHRQ builds on real work.

How do we start?

Request access and say which side you are on. An AI lab can ask for evaluation or data access. A manufacturer can ask for a pilot conversation.

One line on what you want to evaluate or build is enough. Request access.