Skip to content
Advancing Bits to Master Atoms

Rules and approvals.
Your rules come first, and the steps you choose wait for a person.

In governed environments, a step that breaks a rule is stopped before it runs, and the steps you select, such as an expensive simulation or a change to shared data, wait for a person to approve them.

Remember and govern · one of the fourteen capabilities of MS Duplex

A satin-silver barrier arm across a graphite track, a cobalt-blue key switch in its post and a small silver sphere waiting in front of the gate.A satin-silver barrier arm across a graphite track, a cobalt-blue key switch in its post and a small silver sphere waiting in front of the gate.

Your team gets

  • Costly steps checked before they start
  • Approvals on record, with who approved what
  • Rules that change today, with nothing to retrain

What rules and approvals mean for your team.

Rules and approvals mean the AI your scientists and engineers use works inside the limits your team sets. A rule is something your team requires, such as a budget limit or a protocol a step has to follow. A preference is how your team would rather work, such as a favored method or a default setting. MS Duplex keeps the two apart, and both stay visible in the reasoning, so anyone can see which one shaped a step.

In governed environments, a supported setup with hooks lets the steps you select pass through MS Duplex, so your rules apply before a costly simulation or run begins. There, a step that breaks a rule is stopped before it runs. The steps you mark for approval, such as an expensive simulation or a change to shared data, wait for a person, and each approval goes on record with who approved what.

Rules change as the research does. A new rule applies from the next request, with nothing to retrain and no new model version to validate, because your rules sit in MS Duplex beside the model, with its weights left as they are.

Why research needs AI that works within your rules.

Research agents now plan runs, call simulators and touch shared data. The more they can do, the more it matters that your team’s rules come first.

A research team runs on rules that are easy to state and costly to break: a compute budget, an approved protocol, reference data only certain people may change, a review before a result is marked final. A new colleague learns them over months. An AI agent needs them in view at every step, kept apart from the habits and preferences that are open to judgment.

When rules live only in a prompt or a note, an agent can miss them, and nobody notices until the run has already spent its budget. Approvals given in a chat are hard to find later. In governed environments, MS Duplex checks the steps you select before they start, and keeps every approval with the step it allowed, so the record shows who decided what.

  • Runs are expensive

    A simulation or experiment that should never have started costs compute and lab time that a check beforehand would have saved.

  • Shared data matters

    A change to data the whole team relies on deserves a person’s approval before it happens.

  • Rules change

    A new protocol or budget has to take effect from the next request, with no training cycle to wait for.

  • Decisions need an owner

    Your team needs to show later who approved a step and on what basis.

A straight graphite channel with three satin-silver locks along it, the middle lock sealed with a cobalt-blue band.A straight graphite channel with three satin-silver locks along it, the middle lock sealed with a cobalt-blue band.

How rules and approvals work in your team’s day.

Your team sets the rules and chooses the steps that need a person. MS Duplex keeps them in view and, in governed environments, applies them before a step begins.

  1. 1

    Your team sets the rules

    Rules and preferences are recorded apart, and your team selects the steps that wait for approval, such as an expensive simulation or a change to shared data.

  2. 2

    Your agent plans the work

    It proposes the next step with the model and tools your team already uses, and the rules and preferences that shaped it are visible in the reasoning.

  3. 3

    The step is checked

    In governed environments, a step that breaks a rule is stopped before it runs, and a selected step waits for a person to approve it.

  4. 4

    The approval goes on record

    Who approved what is kept with the step it allowed. A new rule your team adds applies from the next request.

An example, in any field where an agent plans costly runs.

A team’s agent plans a series of runs on a shared compute budget, in a governed environment. Over one week, the team’s rules shape what it does, and every approval goes on record.

  1. 1

    A rule and a preference

    On Monday the team records a rule that any run above a set cost waits for approval, and a preference for its usual analysis method. Both are visible in the agent’s reasoning from the next request.

  2. 2

    A step that waits

    On Tuesday the agent proposes an expensive simulation. In the governed environment, the step waits for approval. A senior researcher approves it with a note on why, and the approval goes on record.

  3. 3

    A step that is stopped

    On Wednesday the agent proposes editing a shared reference table, which a team rule reserves for the table’s owner. In the governed environment, the step is stopped before it runs, and the rule behind it is shown.

  4. 4

    A new rule

    On Thursday, after a review, the team adds a rule that a candidate design is marked final only after an expert check. It applies from the agent’s next request, with nothing to retrain.

Rules and approvals keep the other capabilities in your hands.

Every capability in MS Duplex works within the rules your team sets.

  • Simulation before action

    Simulation before action weighs a costly step against the goal, and in governed environments your rules decide whether it waits for a person before it runs.

  • Decisions that know their limits

    Hard calls go to a researcher with the reason, and in governed environments the steps you select wait for a person’s approval, so people decide where your team wants them to.

  • Personal, team and organization layers

    Your rules decide what moves between personal, team and organization layers, and an organization standard applies everywhere you want it.

  • Provenance, review and unlearning

    Every approval sits in the same history as the lessons, with who approved what, ready to review and export.

What your team gets.

AI that works within your team’s rules, with costly steps checked and every approval on record.

  • Costly steps checked before they start

    In governed environments, the steps you select are checked against your rules before compute or lab time is spent.

  • Rule-breaking steps stopped

    In governed environments, a step that breaks one of your rules is stopped before it runs.

  • Approvals where you want them

    The steps you choose, such as an expensive simulation or a change to shared data, wait for a person to approve them in governed environments.

  • Approvals on record

    Every approval is kept with who approved what, ready for review.

  • Rules and preferences apart

    What your team requires and what it would rather do are kept separate, and both are visible in the reasoning.

  • Rules that change today

    A new rule applies from the next request, with nothing to retrain.

Common questions.

Does MS Duplex stop a step before it runs?

In governed environments, yes. Where a supported setup with hooks lets the steps you select pass through MS Duplex, a step that breaks a rule is stopped before it runs, and the steps you mark for approval wait for a person. With a standard connection, your agent calls MS Duplex to bring back relevant experience, and your rules and preferences are visible in its reasoning.

What is the difference between a rule and a preference?

A rule is something your team requires, such as a budget limit or a protocol. A preference is how your team would rather work, such as a favored method. MS Duplex keeps the two apart and shows both in the reasoning, so anyone can see which one shaped a step.

Which steps wait for approval?

The ones your team selects, such as an expensive simulation or a change to shared data. In governed environments, those steps wait for a person before they run, and each approval goes on record with who approved what.

What happens when we change a rule?

A new rule applies from the next request. There is nothing to retrain, because your rules sit in MS Duplex beside the model, with its weights left as they are.

Make the next run benefit from the last.

Bring us the research your AI should get better at. We will plan a pilot on your own workflows, with your experts as the judges.