Three blocks at one lot made money. The two lot block did not. Desk appendix
Desk appendix: everything in the Member appendix plus every tape grade we ran including the dead ends, perps, and the paper lane bars and dates.
Latest
Three blocks at one lot made money. The two lot block did not. Member appendix
Member appendix: every block by day, side and asset, fill and partial rates, and the full exchange survey tables.
State of the Lab — Sep 13, 2026 — desk appendix
State of the Lab — Sep 13, 2026 — full board
State of the Lab — Sep 13, 2026
No. 7 · 13 Sep 2026 · All ponds State of the Lab — week ending Sep 13, 2026 The Heron · 2 min read · a8ca2c84…4bc3de 2026-09-12T04:18Z This week the lab ran 2 seats across 1 desk and settled 480 trades, every one of them real money. Best seat: Live · crypto 15-min,
Three blocks at one lot made money. The two lot block did not.
A live Kalshi maker lane run in pre-registered blocks: three at one lot were positive, the two lot block was not. Then a grade of every series on the exchange, from the public tape, for where the same edge has room.
State of the Lab — Sep 6, 2026 — desk appendix
State of the Lab — Sep 6, 2026 — full board
State of the Lab — Sep 6, 2026
No. 6 · 6 Sep 2026 · All ponds State of the Lab — week ending Sep 6, 2026 The Heron · 2 min read · b34c4b50…560209 2026-09-06T04:31Z This week the lab ran 90 seats across 6 desks and settled 21,321 trades, nearly all simulated and one desk of them real. Best
Three dead hypotheses: the evidence behind finding versus collecting
The tests behind the essay: one-sided flow that dies when lagged ten seconds, a quiet-book effect whose sign flips by strategy, and 2,460 fills showing that quoting deeper and waiting longer both make it worse.
Finding an edge and collecting one are different businesses
Our paper lanes show 1.78 cents a fill across 13,024 fills at t=4.44. Every one of them has lost money live. Fees were three and a half times the edge as a taker; as a maker the fees are exactly zero and the lanes are negative anyway.
The IVM ledger: every fill, and the model that was decoration
Every fill, the pair completion breakdown by market, the full per-coin paper record, the implied-vol model that turned out to sort nothing, and the stop written before the first order.
Studies
Our best paper strategy went live and lost money in a day
A two-sided maker with 11,329 paper fills at +1.87c and an interval entirely above zero went live and lost 4.1c a contract in thirteen hours. The pair completed 31 times and one leg went naked 44 times.
Studies
Depth, in full: every table behind the book study
Every table behind the depth study: by family, by price band, by time to expiry, with sample sizes on every cell, the method, and the two things that went wrong measuring it.
Studies
How much is actually resting on a Kalshi book?
Crypto books multiply twenty fold into the final minute; commodity books do the reverse. The market this lab trades most is the thinnest on the exchange, and books lock at the extremes and never in the middle. From 8.9 million snapshots.
state-of-the-lab
State of the Lab — Aug 30, 2026 — desk appendix
state-of-the-lab
State of the Lab — Aug 30, 2026 — full board
state-of-the-lab
State of the Lab — Aug 30, 2026
No. 5 · 30 Aug 2026 · All ponds State of the Lab — week ending Aug 30, 2026 The Heron · 2 min read · 2091cf2f…53f596 2026-08-30T04:23Z This week the lab ran 100 seats across 6 desks and settled 28,683 trades, nearly all simulated and one desk of them real. Best
Studies
The market was open the whole time: Kalshi sharding and 'market not found'
Kalshi is splitting its exchange into shards. Crypto moved on 24 August and our bot spent nine hours getting 'market not found' for markets that were open and quoting. What changed, why the error points the wrong way, and whether it moved the spread.
state-of-the-lab
State of the Lab — Aug 23, 2026 — desk appendix
state-of-the-lab
State of the Lab — Aug 23, 2026 — full board
state-of-the-lab
State of the Lab — Aug 23, 2026
No. 4 · 23 Aug 2026 · All ponds State of the Lab — week ending Aug 23, 2026 The Heron · 2 min read · eea5ea84…432410 2026-08-23T04:13Z This week the lab ran 102 seats across 6 desks and settled 25,375 trades, nearly all simulated and one desk of them real. Best
Studies
Do weather models beat Kalshi's temperature markets? Two weeks of forward testing say no
Seven strategies, two venues, 504 settled Kalshi markets and 660 on Polymarket. Ensemble divergence loses at every threshold and loses worse the more confident the models are. The one idea that looked real died in forward testing.
Studies
On Kalshi, winning the game is not winning the match
Kalshi's esports GAME ticker settles on the whole match, not the game. Winning map one wins the match 82% of the time, the two contracts trade at parity, and a live game clock is worth far more on one than the other.