Plebbit is a theoretical design for a Reddit alternative built partly using IPFS' experimental pubsub feature. View the whitepaper.
One of the design that needs to be tested is how big can the pubsub network scale, and how much spam resistance does the captcha service over pubsub provides.
Multiple scenarios should be tested:
How much time does it take on average to complete a full captcha challenge request-answer-validation exchange.
What is the maximum amount of posts per minute that can be published without breaking.
Can the captcha service successfully prevent attackers from spamming captcha challenge requests.
Can the captcha service successfully prevent attackers from spamming captcha challenge answers that are incorrect.
The "captcha puzzle" doesn't need to be implemented, the captcha challenges and answers should be simulated for the experiment. The captcha answers validation and blocking of peers who submit too many incorrect captcha answers (or too many captcha challenge requests without answers) should be implemented. This might require forking the IPFS go client to add the functionality of blocking a pubsub peer. It might be possible without forking using IPFS swarm filter.
The experiments should be done using at least 3 nodes on 3 servers: A subplebbit owner's node running the simulated captcha service, one (or more) well behaving nodes publishing posts, and one spammer node trying to break the pubsub.