PoW + reputation: honest limits

p2pbuilders is permissionless: keys are free, posting needs nobody's
approval. Everything standing between the feed and a spam firehose is four
layers of cost-raising — and the most useful thing we did was write down
exactly where they fail.

What we run

1. Proof-of-work on post/comment/board ops (~80ms / ~20ms / ~1.3s).

Un-worked ops are refused at indexing time — PoW is a policy each
peer enforces for itself, not a consensus rule
(the foundations chapter has the framing).

1. Per-key rate limits at the indexer; violators' cores stay replicated

(cheap) but stop being indexed (the part that matters) for a cooldown.

1. Reputation-weighted votes — weight grows with key age × received

upvotes, floor 0.02, cap 1.0. Age is bounded by *earliest op observed by
the network*, so it cannot be backdated.

1. Subscribable blocklists — chosen curators, revocable by

unsubscribing.

The attack math we published

From the spec's sybil analysis:

Reference: one legit 90-day user ≈ 0.93. So a patient cross-voting ring
equals ~35 real users. v0.1 does not detect voting rings. It says so in
the spec, in bold, with the mitigation roadmap (cluster detection; until
then, human curation via blocklists).

Why we design this way

• Layers, not walls. Each layer is defeatable; stacked, they push spam

from "free" to "expensive and slow", which is where communities survive.

• Raise costs at the index, not the transport. Replication stays cheap

and dumb; every peer applies its own admission policy. Censorship
resistance and spam resistance stop being in tension.

• Publish your attack math. Writing the sybil table changed our design

(the 0.02 floor, network-observed age) and tells users exactly what trust
to place in rankings. Security honesty is a feature.