consumes the network.bsky.jetstream archive API (planBackfill / getBlock / getSegment): plan → fetch sealed jss v1 blocks → zstd decompress (std lib, plain frames, no dictionary) → columnar decode → deliver each matching row as a jetstream Event through the same onEvent handler the live client uses, records decoded from DAG-CBOR into the same json.Value shape. max_blocks bounds exploratory runs (a filtered plan can span tens of GB). verified live against stream.waow.tech via 'zig build archive-smoke' (pollz collections, 25 blocks, 78 events: 13 polls, 65 votes). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>