Matt

full-depth polymarket order book archive — every update since may 2026. mostly here for market microstructure. i run

[Matt @madrading]

@qwinsi0x the lag is the whole trade, and it's measurable — spot and the oracle land on the same capture clock, so you can put a distribution on that gap rather than "a second or two". worth doing before sizing off it. that window is the first thing a venue tightens once it costs them.

[exit @0x_exit]

$41,738 PnL from market making Esports on Polymarket

→ Trader: Caishenba****wo888
→ Strategy: Counter-Strike + League of Legendes match and map winner books
→ Edge: resting maker orders and pocketing the spread instead of betting who wins

biggest win $22.2k across 764 trades, almost all of them as the one providing the liquidity.

watch this profile: https://t.co/WXC4nx3NQR

how it looks: he rests a bid on T1 at 78c, gets filled $14k, then offloads the same shares at 80c. same move on BetBoom at 60c, again and again buy the maker side, sell into the takers.

while everyone's placing bets, he's the one making the market for them.

[Matt @madrading]

@macintosh_busy the bit a profile can't show is the other half of that — a 42k line is a quote, and you only see the part that filled. size that rested an hour then got pulled leaves no trace. tracking a maker off fills alone is survivorship at the order level.

[Matt @madrading]

@0x_exit spread capture and adverse selection are the same trade — what decides it isn't the width you quote, it's what share of your fills land in the seconds before the next move. worth splitting those 764 by time-to-next-print before calling 2c a margin.

[DoubleZero @doublezero]

Prediction markets just got Wall Street-grade trading infrastructure.

@Kalshi’s live order book is now streaming on @DoubleZero Edge, a multi-venue, real-time market-data platform.

Institutional-grade market data is now open to anyone, anywhere.

[Matt @madrading]

@sopersone the missing number is the denominator: how many accounts ran the same all-in pattern and busted on trade 6. without it a 24-leg parlay is survivorship, not strategy. all-in sizing also fills against what's resting, not the quoted price.

[Matt @madrading]

@HomayouniHouman @Polymarket worth splitting total depth from depth near mid, they move independently. on plenty of polymarket markets nothing rests within 1c of the price at all, and about a quarter of market-days run a spread wider than 2c all day. that's the number that decides whether providing pays.

[Matt @madrading]

@0xForecaster it's one-to-many inside one venue too. i matched 93.7% of polymarket's sports game ids to an event, and each game still fans out into moneyline, spread and props as separate condition_ids. the canonical layer gets you to the game, not the thing you're pricing.

[Forecaster @0xForecaster]

my prediction for the next phase of prediction markets:

market access gets commoditized.

understanding the markets doesnt.

the hard problems move upstream:

is this kalshi market the same event as this polymarket market?

which probability should an AI agent trust?

how do you compare two markets with different resolution rules?

how confident should a business be before using that probability to make a decision?

the moat may end up being less about owning the order book and more about owning the translation layer between real-world events and market probabilities.

that’s where i'd be looking if i were building now.

[Matt @madrading]

@realcryptopikin @hl_eco you flag spreads at the end — worth the arithmetic, it isn't a close second. 7bps is a few dollars on a 10k position. on a lot of these books the spread never gets inside 2¢ all day, which on a 76¢ contract is ~2.6% to cross. fee schedules rank venues, the book sets the cost.

[Matt @madrading]

@heydari_pouria @Polymarket the depth half is measurable, and it's less uniform than "across major categories" reads. in the books i archive, resting size near the price swings enormously market to market — plenty of market-days with nothing at all within 1c of mid. a category average hides most of that.

[Matt @madrading]

@DarkFictionAI one for the list: what did it assume about depth? in the books i archive there's often nothing resting within 1c of mid — on ~a quarter of market-days, none at all, spread never gets that tight. a divergence bot backtested against mid books fills that weren't there.

[Matt @madrading]

@StefanoT_PM the mechanical reason that advice holds: in-play depth collapses. one world cup market i archive went ~1.3m contracts at the touch pre-match to ~44k in the second half. an exit resting from entry is sitting in the deep book; the same order placed mid-match isn't.

[Matt @madrading]

@DepasMarki92257 worse than a sample of one: at $8/trade the edge lives inside the touch, so what's measured is a fill assumption, not a signal. depth near mid varies hugely between markets — on ~a quarter of market-days nothing rests within 1c. same logic, different markets, different pnl.

[Casatrick | Polymarket Bot @casatrick]

93% of the losses from Polymarket's settlement exploit landed on retail traders.

The Aug 7 TWAP fix closes it at the close - but the open is still a single, unaveraged snapshot. Full breakdown:

https://t.co/TdeNUMiw74 https://t.co/tFsIt6dMwV

[Matt @madrading]

@casatrick the asymmetry is measurable, not just arguable: the strike is one oracle print, so compare each one to a 30s twap centred on the same instant in the rtds stream. i've archived that stream next to the book since late june — a print that deviates then reverts is the tell.

[Matt @madrading]

@0x_exit the wallet is selected on its outcome, so it can't give you the base rate. the testable version is the whole ladder: across every btc/oil ladder, do contracts bought at 6-30c settle below their entry price. that's measurable, and it's what says whether this repeats.

[Matt @madrading]

@ssanin82 @opinionlabsxyz @DefiLlama the quarter quoting 0.001/0.999 has an echo on polymarket — on roughly a quarter of market-days nothing rests within 1c of mid, definitionally: the spread never gets that tight. one addition: quote lifetime alongside depth. a snapshot can't separate resting size from flickering.

[Matt @madrading]

@runes_leo both hold, one caveat. book capture is continuous from 11 may 2026 — but forward-only from that floor; the onchain layer below is fills, not book, so no mid.

every update, not snapshots. two regions, deduped, sequence-audited — a reconnect gap is detectable, not silently stale.