How an investigation is made
Four stages, what is checked along the way, and what you can inspect at the end. The counts are from the three published investigations, read from the investigations themselves.
The four stages
- 01
Supply the research
Upload 25–100 papers and tell Between Papers what you're trying to understand.
- 02
Build the evidence
Technical findings, experimental conditions, limitations, agreements, and contradictions are extracted and connected across the corpus. Each claim is checked against the passage it came from.
- 03
Investigate what they enable
Findings are connected into emerging technical capabilities, including potentially useful connections from adjacent fields and, where it helps, sources from outside your corpus.
- 04
Explore commercial implications
Capabilities become potential applications and business opportunities, while evidence, assumptions, and uncertainty remain visible.
What is checked
A claim is kept only after it is checked against the passage it cites. The checks and the corrections are counted on the investigation, not summarised away.
| Published investigation | BatteriesCheaper batteries for homes and the grid | Oil & gasCutting methane leaks in oil and gas | FinanceAI against payment fraud and financial crime |
|---|---|---|---|
| Seed papers | 25 | 25 | 25 |
| Sources looked at | 396 | 426 | 266 |
| Screened from title and abstract | 396 | 367 | 266 |
| Shortlisted | 58 | 93 | 50 |
| Could not be fetched | 36 | 63 | 30 |
| Read in full, and added to the report | 6 | 13 | 5 |
| Checks run against passages | 314 | 371 | 240 |
| Claims caught and corrected | 86 | 94 | 46 |
| Claims kept | 158 | 209 | 172 |
| Findings | 6 | 6 | 6 |
| of which contradictions | 1 | 2 | 1 |
| Opportunities | 3 | 4 | 4 |
| Combined ideas | 3 | 2 | 3 |
Each figure is read from the investigation it describes, at build time. Open any of the three to see the same numbers on the report.
What is looked at, and what is read
Every candidate source is screened from its title and abstract before any is read in full. A source that could not be fetched is counted and listed, not dropped quietly.
36 shortlisted sources could not be fetched (no open copy, or unreadable) and are listed as such. Every source read in full is cited in the report, with why it was added.
See every source it looked at63 shortlisted sources could not be fetched (no open copy, or unreadable) and are listed as such. Every source read in full is cited in the report, with why it was added.
See every source it looked at30 shortlisted sources could not be fetched (no open copy, or unreadable) and are listed as such. Every source read in full is cited in the report, with why it was added.
See every source it looked atWhat you can inspect
Every important conclusion walks back to the original papers through the same chain. On an investigation page, each step is one click.
- Conclusion
- Reasoning
- Supporting evidence
- Conflicting evidence
- Assumptions
- Exact passages
- Original papers
- Conclusion
- The answer, a finding, an opportunity or a combined idea.
- Reasoning
- The finding's detail, an opportunity's why-now and rating reasons, a combination's why-new.
- Supporting evidence
- The claims a conclusion cites, each numbered the same wherever it appears.
- Conflicting evidence
- The contradictions section, and the qualifying claims a finding cites against itself.
- Assumptions
- Listed on every opportunity, with its main risk; where it breaks, on every combination.
- Exact passages
- The passage each claim was checked against, with its page and section.
- Original papers
- The seed papers and every source the investigation added, with why it was added.
What it does not claim
- A commercial application is a possibility with the evidence behind it, not a prediction that the technology will work.
- An investigation reads what is openly available. Where a source could not be fetched or a paper could not be parsed, the investigation says so.
- The investigation reads papers you have legitimate access to. It never bypasses a paywall.
Here is the complete work. Challenge it.
Open an investigation, follow any conclusion to its passage, and decide for yourself.
The flagship: cheaper batteries for homes and the grid, 25 seed papers, 396 sources looked at.