Each row summarizes how balancers handled large one-sided flow over the trailing 7 days. Reading the columns:
big imbalances (7d) - trades (or bursts within 60 s) that pushed the market's long/short imbalance further out by at least the dollar bar shown next to the market name.
answered by program - share of those events where a Balancer Program desk pushed back within 60 minutes with a trade at least a quarter of the event's size.
program median - typical wait for that program push-back; >60 m means most events never got one within the hour.
market median - same clock, counting a push-back from anyone (program member or not): how fast the market as a whole reacts.
recovery median - typical time until the imbalance actually came back down by half the event's size, whoever traded it back.
unrecovered at 60 m - share of events where even that halfway recovery had not happened an hour later.

Range: to 3d 7d all .jsonl.gz.csv Download range (one file)

One JSON object per line, one line per ~120 s poll. Key groups: markets.* per-market skew, price and impact-pool reads; state.* the risk ledger (X, D, C_mark, incidents, machine); exit.* the exit-cost test (U dial, tail, B_net). Ranges exceeding the archive clamp to it (ask for everything, get everything). The range download is standard multi-member gzip - any tool decompresses it as one stream. CSV: converted in your browser - nested fields become dotted columns (markets.BTC.s_usd), one header across the whole range even where the schema grew, arrays as JSON cells; needs a modern browser. Column key: columns.md - every column defined, with units, sign conventions and the joining formulas (give it to any analyst or LLM alongside the data). Scripts: manifest.json lists every day; per-day URLs are stable. Today's file is partial through the newest tick.