Thread by @sinahab

1) Reputation jam with good people yesterday – @danfinlay, @decentralion, @nayafia, @sunnya97, @rsepassi, @sidrmsh, @graeme_tweets, Ryne, Thor, @jwmares.

Our discussion was inspired by PageRank, EigenTrust, TrustDavis, Social Collateral, PGP, Circles, ...

Some takeaways:

2) Reputation leads to a unique form of liquidity within social communities

3) When things go wrong, do you need recourse (which relies on dispute resolution) or can you simply ostracize the defecting individual?

4) Historically – e.g. in the case of credit card networks:

Recourse has been important – i.e. challenging a bad charge.

Ostracizing has also been used – i.e. kicking someone off the network after multiple misbehaviors.

cc Ryne

5) reputation lets you take a chance on people. Have a first interaction with a stranger based on their reputation, and then develop a direct relationship (i.e. an edge)

6) Pattern matching on the graph is a useful technique for Sybil resistance.

An example heuristic: “these 5 accounts have 20k exchanges with each other, but never with the outside. They obviously trust each other, but I don’t trust them.”

cc Thor

7) reputation for underwriters is an important area of research for Dharma. Ideas they're exploring: 1- make underwriters co-sign loans (i.e. costly in capital), 2) make underwriting require past successful behavior (i.e. costly in time), 3) chains of trust, like CAs

cc @sidrmsh

8) a useful mechanism as inspired by PageRank: the random surfer breaks out of Sybil sub-graphs because there’s a chance the algorithm jumps to a random website.

Same mechanism exists in the EigenTrust algorithm, which jumps to one of pre-trusted peers.

cc @rsepassi

9) Dense tightly-couple graphs are useful for high-value interactions and trust (e.g. debt); large loosely-tied graphs are useful for low-value interactions and access (e.g. flow of information)

10) There should be transitive decay as you look further away in the trust graph. “I trust friend of a friend more, than the friend of a friend of a friend”.

11) Behavior vs. judgement are potentially different. One can behave well, but a poor judge of others.

Revelant: Bayesian Truth Serum

12) Self-sovereign identity & reputation are important. If Amazon goes out of business, reputation of all merchants goes away along with it, even-though they are no less reputable. Something crypto networks are suited for.

13) SourceCred uses existing behavior of Git PRs, issues, and code dependencies to build a reputation graph. Leverages existing offchain behavior; code commits as "Proof of Work”; small communities; and the project maintainers judgement on managing the system.

cc @decentralion

14) Interoperability could be more important than scale; create lots of small local systems that can interoperate, rather than one global system that scales.

cc @nayafia

15) Local trust graph are powerful.

Social collateral for small amounts of money, no questions asked, composes into graphs which enable you to interact with people you don’t directly trust.

cc @danfinlay

16) Thanks to our good friend and undercover crypto geek @jwmares for hosting us, appreciate it buddy.

Missed @robbiebent1 at this one, but took inspiration from his community magic.

Looking forward to more jams in the future!