kiosk-pow-argon2id is the legacy proof-of-work backend for the Kiosk framework: an Argon2id search-form proof of work. The shipped default is Equihash (kiosk-pow-equihash). Argon2id stays a legacy backend because its 64 MiB verify is too expensive for a gateway. Ships the operator-side Ruby verify (one Argon2id evaluation, no loop) and a Python solver (solve.py, argon2-cffi). The Ruby verify and the Python solver produce byte-identical Argon2id digests. The gem's `:parity` spec examples check this, and so does the `rake parity` task. Neither the Rakefile nor spec/ is in the package. Built on kiosk-pow. kiosk-reputation reaches it when a host registers it under "argon2id"; it does not register itself. Does not depend on kiosk-core or Rails: pure Ruby and libargon2 through FFI.

kiosk-pow-cuckoo is an optional Cuckatoo-Cycle (Cuckoo Cycle variant) proof-of-work backend for the Kiosk framework. Ships the provider-side Ruby VERIFIER only. The verifier is cheap (a handful of hash evaluations; no wall-clock figure is benchmarked), security-critical, and validated against Grin's Cuckatoo29 CI test vector. A reference Python solver (solve_cuckoo.py) ships alongside for parity testing; production solving is the client's concern. The implementation is clean-room from Tromp's public-domain algorithm spec (doc/spec, doc/mathspec) — no GPL/FAIR-MINING code is included. BLAKE2b-256 comes from kiosk-pow, the base of every Kiosk PoW backend. SipHash-2-4 is pure Ruby from the public-domain SipHash spec with Cuckatoo's non-standard initialization (keys feed directly into v0..v3 without XOR-ing the 0x736f6d65... magic constants). Does not depend on kiosk-core or Rails — pure Ruby, no native extensions.

kiosk-pow-equihash is the shipped default proof-of-work backend for the Kiosk framework: a pure-Ruby Equihash (Biryukov & Khovratovich birthday- collision PoW) VERIFIER built on kiosk-pow, which holds the pure-Ruby BLAKE2b-256. The gem depends on neither kiosk-core nor Rails. `Kiosk::Pow::Equihash.verify(salt:, params:, nonce:)` recomputes the 2^k BLAKE2b-256 hashes named by the proof's indices, checks that they XOR to zero over all n bits, and walks the Wagner collision tree (per-level XOR cancellation plus the Zcash-canonical subtree ordering that pins one canonical form per solution). Difficulty is set by (n, k) alone — there is no post-hoc target check. Default parameters are n=168, k=7, chosen by the benchmark in bench/: the largest params whose reference solve stays inside a consumer-laptop budget (p95 ~10 s on one M-series laptop core, the only hardware the seconds have ever been measured on; the ~1.3 GiB peak is THIS solver's table, not the host's -- and not a floor (n, k) imposes on every implementation, since a memory-optimised solver trades it for time) while verify costs 128 BLAKE2b evaluations — ~22 ms and a few KB in pure Ruby without YJIT. That asymmetry is the point: the gem prices a request, it does not equalise hardware. Equihash is neither ASIC- nor GPU-proof (it was ASIC'd on Zcash); abuse resistance comes from the reputation policy's N-proofs knob and caps, not from this algorithm. Ships the provider-side verifier only. A reference Python + numpy solver (solve.py) rides along for parity testing and for assistants that need one; production solving is the client's concern.