Mathematical discovery and verification.
The research behind Panbiota: finding relationships that hold, computing results exactly, and checking every one.
Four lines of research.
Exact calculation and checking
An exact core that takes a set of statements, checks that they are consistent, works out what follows with a full calculation trace, and refuses rather than guesses when an answer cannot be determined. Quantities, units and rates are computed exactly, never estimated.
- Where it stands
- Demonstrated on generated problems inside our research. Not yet a public service.
- Next
- Hosted verification and calculation services, each released only after a test on structured inputs.
A mathematical language
Turning plain language into exact mathematical statements, so that a question asked in words can be calculated rather than approximated.
- Where it stands
- In development, measured on text not seen during development at every stage.
Mathematical discovery
Methods that search data for relationships and keep only those that hold, written as reusable mathematical structures that can be checked, combined and used again.
- Where it stands
- In development. First domain: Solana, where every relationship found can be checked against what followed.
Evidence and provenance
Every figure traced to its source and its time; observed facts, inferred relationships and hypotheses recorded separately, so that each claim can be followed back to what supports it.
- Where it stands
- Live on every instrument and in every answer of the AI access point.
How a result earns its place.
Fixed before the test
What counts as success is written down before a test runs, and is not changed afterwards.
Checked beyond its own data
A relationship counts only if it holds on data it was not found on.
Compared with a baseline
A method must do better than straightforward alternatives to be worth keeping.
Correlation is not cause
A relationship that predicts is not assumed to explain, and an accurate prediction is not assumed to be valuable.
Every attempt recorded
Approaches that do not work are written down too, with what was learned, so they are not repeated.
One question, against frontier models with tools.
“A new Solana token has just launched. Who is behind it?” Panbiota proposes this question as a benchmark against frontier AI models equipped with equivalent blockchain data access and analytical tools, asked of both on the same launches. It will measure:
- Accuracy of the relationships identified
- Completeness of the answer
- Response time
- Cost per verified answer
- Evidence quality and provenance
- False-positive rate
- Recognition of networks seen before
Answers in under a second are a performance target, not an established capability. Panbiota does not claim to outperform frontier models until the comparison is run and its results published, with their scope and their limits.
Results with their scope.
Panbiota publishes no benchmark figures yet. Results will be published with their scope, their baselines and their limits when they are ready. A paper on the approach is in preparation.
Asked and answered.
What is Panbiota researching?
Four lines: exact calculation and checking; a mathematical language that turns plain words into exact statements; mathematical discovery, which finds relationships in data and keeps only those that hold; and evidence and provenance. Only provenance is live today; exact checking is demonstrated in research and the rest is in development.
Has Panbiota published benchmark results?
Not yet. Results will be published with their scope, their baselines and their limits when they are ready, and a paper on the approach is in preparation. Panbiota makes no claim of superiority before it is measured.
How will Panbiota compare itself with frontier AI models?
With a proposed benchmark: who is behind a new Solana token, asked of Panbiota and of frontier models equipped with equivalent blockchain data access and analytical tools, on the same launches. It will measure accuracy, completeness, response time, cost and evidence quality, with false positives and recognition of networks seen before. It has not been run yet, and answers in under a second are a target, not a capability.
How does Panbiota decide that a result is real?
What counts as success is fixed before the test runs; a relationship counts only if it holds on data it was not found on; a method must do better than straightforward baselines; and correlation is never taken for cause. Approaches that fail are recorded too.
Is Panbiota's exact checking available as a service?
Not yet. It is demonstrated inside Panbiota's research; hosted verification and calculation services are in development, and each will be released only after a test on structured inputs.
Why start discovery with Solana?
Because its ledger is public and time-stamped, so every relationship a method finds can be checked against what followed. It is a demanding place to prove the methods before they are extended to other domains.
Panbiota provides the intelligence behind the answers.
Intelligence, not advice. The decisions are yours.