Thread by @balajis

Someone should set up a site like Scott Aaronson's Complexity Zoo, but with all the proof-of-X papers and code.

If you combine reproducible research, cryptographic proof-of-X, and federated analytics, you can envision a world where every paper isn't just reproducible, but where every row of data is logged on-chain with all relevant proof-of-Xs to demonstrate data integrity.

Premises:
- Many papers are based on datasets
- Those datasets are rollups of individual data points, like measurements on an individual
- The datasets (and sometimes the raw data points) are currently in silos like NCBI or Bloomberg
- But you could put them on-chain

- If every data point was on-chain, you could accompany it with proof-of-X
- Proof-of-who: who made the measurement
- Proof-of-when: it was made at this time or earlier
- Proof-of-where: where it was made
- And so on

A chain of custody for individual data points, on-chain.

For example, covidtracking.com goes to admirable lengths to show data provenance. You can track back each case to individual state agencies. But you can't track it back to the very origin, which is a hospital reporting the data to the state.
covidtracking.com/data/state/ala…

Covidtracking is one of the best projects I've ever seen in terms of data integrity. If you go just one step further though, you could double click on (say) 1000 cases reported by a given state to see 37 hospitals in that state each posting individual data points on-chain.

That last step (posting raw data points directly on-chain, with digital signatures, rather than merely putting aggregates online via a state agency site) might seem like a small thing, but it would allow for total independent reproducibility by pulling all data from a chain.

Where you'd probably start this though is with financial data. Every archival data point currently accessible via the Bloomberg Terminal, but on-chain. A number of NBER-type papers could then be turned into Jupyter notebooks where everything could be reproduced by hitting enter.