Every result you have, in one graph
Backbond reads your papers, notebooks, spec sheets and run data and turns them into one connected causal graph. Then you ask that graph questions in plain language and watch it reason, instead of taking its word for it.
Built for the way a lab actually records things
No migration, no re-entry, and no asking scientists to write for a machine. Backbond meets the record where it already is, and everything below runs on the same project.
Reads what you already have
Papers, lab notebooks, spec sheets, slide decks and run data, including scanned and table heavy documents. No template, no re-entry, no schema to agree on first.
One graph per project
Components, the properties they move, the direction and size of each effect, how confident the evidence is, and the study behind every claim.
Answers you can audit
Ask in plain language and watch the whole derivation: the recipe it rebalanced, the pathways it traced, the findings it weighed and the conclusion it drew.
Documents in, structure out
Point Backbond at a folder and it reads every file end to end: the rules stated in a discussion section, the formulations in a spec table, the results in a run sheet. What comes out is structured, linked to the file it came from, and ready to be asked about.
- Parses scanned and table heavy PDFs, not just clean text
- Extracts rules, formulations and results as separate objects
- Keeps every claim pointed at the document it was read from
One picture of everything you know
Ingredients, the properties they move, and which way each one pushes. This is a real project graph after an analyze run, and it is live: drag a node to pull the graph apart, click one to open every relation it takes part in, and read the mix itself in the composition panel.
- Green raises the property, red lowers it, amber means the drivers disagree
- Every relation carries a size, a confidence band and its source study
- The composition panel answers what the graph cannot: how much of each
Change one thing, and everything else moves
A formulation is a closed system: take a point out of one ingredient and the rest have to absorb it. Backbond rebalances the whole recipe before it predicts anything, so what you are shown is the effect of the change plus the effect of the rebalance, not the change alone.
The cone of what it touches
From the thing you changed, Backbond walks the graph outward to every property it can reach, and shows which way each one moves. It is the step that catches the second-order effect nobody was looking for, which is usually the one that costs a batch.
Every finding it weighed
The findings behind the answer, strongest first, each with the direction it supports and how confident the source is. Nothing is asserted without the evidence for it being on the screen, and where the corpus disagrees with itself, that shows too.
A question, and the whole derivation behind it
- Ask in plain language. No query syntax, no naming the nodes you mean.
- Read the reasoning as it is built: the question it formulated, the recipe it rebalanced, the pathways it traced, the evidence it weighed.
- Every conclusion carries its confidence and the findings that produced it, so a reviewer can disagree with the argument rather than with the output.
- Investigations are saved and replayable, so the reasoning behind a decision is still there six months later.
Frequently asked questions
What file types can Backbond read?
PDFs, Word documents, PowerPoint decks, Excel workbooks, notebooks and raw run data. Scanned and table heavy PDFs go through a dedicated parser; everything else is read locally. You can also link a Google Drive or SharePoint folder and let it stay in step.
Does it need my data in a particular format?
No. The whole point of the analyze step is that it reads what a lab already writes. There is no template to fill in and no schema to agree on before you start.
Where do the relations in the graph come from?
From your own documents. Every edge carries its direction, a magnitude, a confidence band, the numeric range behind it and a pointer to the source it was read from, so any claim on the graph can be traced back to the page it came from.
What happens when two sources disagree?
The graph says so rather than picking a winner. A property whose drivers disagree is marked as contested, and the relations that conflict with another claim in the corpus are flagged on the card, which is usually the most useful thing on the screen.
Is my data used to train anything?
No. Your data is never training data for anybody else. Backbond runs in our managed cloud, in your own VPC, or entirely on-premise and air-gapped, and the knowledge graph it builds leaves with you.
See what your team already knows
Bring one project and the documents behind it. We will run the analyze step with you and show you the graph it builds from your own data.