Nivuno's research combines statistical testing, recorded evidence and separate review. The testing process is designed to distinguish a promising observation from a finding that withstands scrutiny.
Information available at the time
Historical analysis must respect when information became available, not simply when an event occurred.
Our software checks availability records and keeps later information separate from earlier inputs. Observations without sufficient timing evidence are excluded from point-in-time analysis.
Discovery and confirmation
Data used to explore an idea is distinguished from data reserved to evaluate it.
For machine learning, training, validation and test periods are arranged chronologically. Checks exclude samples that cross the prescribed boundaries, and learned preprocessing uses training data only. Access to protected confirmation data is restricted and recorded.
Decisions recorded in advance
Before a decisive evaluation, the research question, selected measurements, comparison, exclusions and criteria for interpreting the result must be fixed.
Changes require a recorded revision. This prevents the original test from being quietly rewritten around its outcome.
Statistical scrutiny
The analysis must consider sample size, uncertainty and dependence between observations. Testing many ideas also requires controls for findings that could arise by chance.
The statistical software is itself tested using examples with known answers and deliberately introduced errors, checking that its safeguards detect the faults they are intended to catch.
Review and traceability
Behavioural analyses require a separate audit that reproduces the calculations from the source observations. An additional adversarial AI review challenges the method, evidence and interpretation before acceptance.
Records identify the data, software, settings and exclusions behind each result. Positive, negative and inconclusive outcomes are preserved, with conclusions limited to what the test supports.
Receive future research updates and programme overviews. Register for updates