MS DuplexProvenance, review and unlearningShow allShow less
Provenance, review and unlearning.
Every lesson keeps its source and can be corrected or removed.
In MS Duplex, each piece of experience shows its source and what it rests on, in one history your team can review and export.
Remember and govern · one of the fourteen capabilities of MS Duplex


Your team gets
- Decisions another researcher can reproduce
- Mistakes fixed precisely, without side effects
- A complete, exportable history of the work
What provenance, review and unlearning mean for your team.
Provenance means every piece of experience in MS Duplex shows where it came from: the result, measurement or expert judgment behind it, and the conditions it was confirmed under. When an agent recommends a step, your team can see what that recommendation relies on.
Review means the whole record sits in one history your team can read and export. It covers what was learned, what was used and what was approved, and it is signed. Another researcher can rebuild a decision from it and check it for themselves.
Unlearning means a mistake can be fixed precisely. Correct a lesson and only that lesson changes. Remove it and the recommendations that depended on it go with it, while everything else stays as it was. A removed lesson can be restored if your team changes its mind.
Why research needs AI that shows its sources.
A conclusion is only as useful as the evidence behind it. AI that cannot show its sources asks your team to take its word.
In research, every result has to be traceable. A reviewer asks where a number came from, a colleague wants to check a decision before building on it, and a later project needs to know which assumptions held at the time. When the AI’s reasoning rests on lessons with no source, none of these questions has an answer.
Mistakes happen too. A measurement turns out to be wrong, or an expert’s correction held only under conditions nobody wrote down. When knowledge is baked into a model’s weights, fixing one thing means another training run, and other knowledge can wear down with it. In MS Duplex each lesson stands on its own, so your team corrects or removes the one that is wrong.
Traceable evidence
Every lesson points to the result, measurement or expert judgment it came from.
Reproducible decisions
Another researcher can rebuild a decision from the record and check it for themselves.
Precise corrections
One wrong lesson can be fixed without touching the rest of what the team has built.
Knowledge that moves on
A new measurement supersedes the old one, and the history of what changed is kept.


How provenance, review and unlearning work in your team’s day.
Your team keeps working the way it does today. MS Duplex keeps the record as the work goes on and gives your team the controls to act on it.
- 1
A lesson is kept with its source
When a result is confirmed or an expert corrects the agent, MS Duplex records the lesson with its evidence and the conditions it holds under.
- 2
Recommendations show their basis
When an agent proposes a step, your team can see which lessons it relies on and where each one came from.
- 3
The record is reviewed
People within their scope can review the history of what was learned, used and approved, and export it.
- 4
A lesson is corrected or removed
If a lesson is wrong, your team corrects it and only that lesson changes, or removes it and the recommendations that rested on it go too. It can be restored later.
An example, in any field where an agent ranks candidates.
A team’s agent ranks candidate designs for review. Over one week, a reference value behind several recommendations turns out to be wrong.
- 1
A question about a ranking
On Monday a reviewer asks why a candidate ranks first. The record shows the lessons behind the ranking, each with the result or expert judgment it came from.
- 2
A source found wrong
On Tuesday the team finds that one of those lessons rests on a measurement taken with a faulty setup. They remove it. The recommendations that depended on it go with it, and every other lesson stays.
- 3
A correction
On Wednesday an expert corrects a second lesson that held under narrower conditions than recorded. Only that lesson changes, and the agent’s next ranking uses the corrected version.
- 4
The decision rebuilt
On Friday a colleague rebuilds the new ranking from the record, sees every source and the history of both changes, and exports it for the project review.
Provenance makes the other capabilities trustworthy.
Everything MS Duplex learns, plans and decides rests on a record your team can check.
Continual learning
Every lesson learned continually arrives with its source, so the record grows as the AI learns.
Learning cause and effect
The link between an action, its conditions and its outcome stays on record, so your team can tell a finding from a hunch.
Rules and approvals
Approvals sit in the same history as the lessons, with who approved what.
Freedom to switch models
The history and its sources stay with your team when you change the model, the agent or the vendor.
What your team gets.
AI whose every lesson your team can trace, check and fix.
A source for every lesson
Each piece of experience shows where it came from and what it rests on.
Decisions another researcher can reproduce
A decision can be rebuilt from the record and checked by someone who was not there.
Precise corrections
Correct a lesson and only that lesson changes, without side effects on the rest.
Clean removal
Remove a lesson and the recommendations that depended on it go with it, while everything else stays.
Restore when needed
A removed lesson can be restored if your team decides it was right after all.
An exportable history
A complete, signed history of what was learned, used and approved, ready to review and export.
Common questions.
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.

