I do not want to have to remember everything.
How my experience becomes a requirement for reliability in your case.
When I work on my own case, I may remember that an event mattered, that something was said or that a document shed light on the situation. Finding the exact passage then means returning to the documents, their dates and the different accounts of what happened.
This difficulty goes beyond memory. You need to explain what you are putting forward, find its support and accept that another document may change your understanding of the case. This matters both to the person concerned and to the professional reviewing it.
This is where my experience informs OpenProof. It helps us set requirements that other users should be able to recognise and check for themselves.
01
A search should help me ask the right question
“Where is that document?” is not always a question I can start with. I may remember a scene, a decision or a particular day without knowing the name of the file that describes it.
The useful question then becomes: “Which documents could shed light on this event?” Followed by: “Do they really concern the same event? What different information do they add?”
I have also insisted on chronology: several things can happen on the same day, and a situation can unfold over months. A date should not hide the distinct events it contains. A document written later should be able to inform an earlier event without moving that event forward in time.
For you, the point is to be able to return to a question with the material you have, even if your case is not perfectly organised. This is the workflow we are working towards and need to test.
02
What research teaches us to examine
Agata Cybulska and Piek Vossen’s work on ECB+ examines a closely related problem: deciding whether passages in different documents refer to the same event. Their corpus adds events of the same type to make that distinction more demanding. ECB+ publication ↗.
This is called event coreference. In our context, it means examining whether two descriptions concern the same meeting or two different meetings. Matching people or similar wording cannot settle the question alone.
TimeML, described by James Pustejovsky and his co-authors, proposes representing events, time expressions and their relations separately, including when a time reference remains partly unspecified. TimeML publication ↗.
We take design and evaluation questions for OpenProof from this work. Applying them to French case documents needs testing in that setting. A published scientific result on news articles does not, by itself, demonstrate the reliability of our product.
What interests me for you is the practical consequence: finding a useful passage, understanding why it has been connected to an event and being able to challenge that connection.
03
Finding what contradicts me is part of the service too
In my own use, I want to examine what informs my interpretation and what weakens it. A contradictory document, an uncertain date or missing information can be decisive.
A clear response therefore needs to distinguish what the document says, what we infer from it and what remains a question. Connecting events in time does not establish that one caused the other. Finding the same people does not reveal their intentions.
For a business leader, someone preparing a case or a lawyer, these distinctions matter when examining possible explanations and deciding what to do. The reasoning needs to remain open to scrutiny.
04
Being able to correct, resume and hand over
We have set another principle: preserve an original and reference precise passages from it, together with the interpretations made of them. The same document may inform a chronology, a question about an actor and an observation in the analysis.
If I reject a connection or correct an interpretation, I want to find that decision when I resume my work. If a lawyer reviews the case, they need to return to the original passage and understand why it was selected.
This is the continuity we are working to build: an understanding that can evolve without losing the sources or corrections behind that evolution.
05
What we need to let you verify
My experience helps us formulate concrete tests. Using fictional documents, we need to check whether OpenProof finds a relevant passage filed elsewhere, keeps similar events distinct, surfaces a contradictory document and preserves a correction after reopening and export.
These questions require us to examine omissions and incorrect connections too. A well-written response is not enough to establish its fidelity to the case.
Our product record already documents targeted checks of preservation and reopening in the private pilot and in fictional tests. Cross-document retrieval in French still needs testing in the served workflow. I want us to report progress with its actual scope, so everyone knows what they can genuinely try.
06
Trust should be verifiable
This is the conclusion I draw from our work: my testimony is not sufficient proof of OpenProof’s reliability. It explains why we insist on sources, distinctions between events, contrary material and the ability to correct.
For you, trust should be something you can build through use: understanding what the product proposes, returning to its supporting material and retaining control of your interpretation of the case.
A situation deserves to be understood before it is judged. Our responsibility is to build a tool that helps you carry out that examination with verifiable reference points.
YOUR NEXT STEP
Would you like to discover OpenProof?
No card required, no payment to start and no automatic subscription.
Try OpenProof — 8 days free ↗The trial is currently offered in a test environment: use fictional documents only. It lets you explore the available version; the research capabilities described here are not all presented as accessible.
Already have an account? Sign in ↗
Would you prefer to tell us about your needs before starting? Contact the OpenProof team ↗
