The chain's pre-market on io

Wall Street closes at 16:00 New York time; the US stock perps on Hyperliquid's EntropyIO (io) deployer keep trading. This board shows how far each one moved on-chain between the previous close and two moments: 04:00 ET, before any US venue opens, and 09:00 ET, after five hours of US pre-market. Once the bell has rung it shows whether the real opening gap went the same way.

as of 2026-10-08 09:00 ET window overnight · Wed 7 Oct 16:00 → Thu 8 Oct 09:30 ET open opened at 09:30 ET 2026-10-08 · gaps filled permalink →

markets with data

2 up · 3 down by 09:00 ET

median |move| · 09:00 call

1.03%

close → 09:00 ET

09:00 call agreed

—

n=5 with both |moves| ≥ 0.10% (<15)

scorecard · all nights

84% / 78%

agreed · 09:00 / 04:00 call · of 100 / 95 · ±0.10% · details

5 marketsdefault order: largest |09:00 move| first · click a column to sort

market	class	04:00 call	09:00 call	high / low	real gap	right before · ±0.10%	volume
EWYus etf	us etf	-2.22%✓	-2.28%✓	+0.74% / -2.69%	-2.15%	09:00 n=12 (<15)04:00 n=11 (<15)	$314k
NBISus equity	us equity	-1.55%✓	-1.78%✓	+0.88% / -2.76%	-2.49%	09:00 79% of 2404:00 77% of 22	$4.6M
SNDKus equity	us equity	-1.02%✓	-1.03%✓	+1.28% / -2.06%	-1.46%	09:00 84% of 2504:00 83% of 24	$22.8M
GPROus equity	us equity	+1.35%✗	+0.73%✗	+3.28% / -0.39%	-0.83%	09:00 78% of 1804:00 60% of 20	$66k
IONQus equity	us equity	+0.83%✓ · 23m old	+0.68%✓	+2.25% / -0.68%	+0.63%	09:00 88% of 1604:00 n=13 (<15)	$100k

04:00 call = last 1-minute candle close before 04:00 ET vs the candle close in the last minute of the previous session (15:59 ET, 12:59 on half days) 09:00 call = the same at 09:00 ET "12m old" = the candle used closed that long before the target minute (shown above 5 min; older than 30 min is excluded from every rate) ✓ same direction as the real gap, ✗ opposite, and · inside the 0.10% dead band (not scored) high / low = extremes of 1-minute highs and lows while the market was closed, same base path = candle close every 15 minutes from the previous close to 09:30 ET, scale per row, centre line = previous close, ticks = 04:00 and 09:00, dashed = no candle in between real gap = oracle price at 09:30 ET vs at the previous close, on io's own oracle; on the 65 symbol-nights xyz also lists, it was within 0.05 pp of xyz's (median) and went the same way 61 of 62 times right before = this symbol's past nights with the same 0.10% band, shown from n=15 volume = sum of 1-minute base volume × close in the window, approximate definitions in methodology

How often the chain called the open30 nights · 2026-08-20 → 2026-10-08 · us_equity + us_etf only

Every completed night since 2026-08-20, two calls against the real opening gap. The 09:00 call has had five hours of US pre-market trading (from 04:00 ET) to follow, so much of it repeats prices already being set on US venues. The 04:00 call is taken before any US venue opens; that part is what only a 24/7 venue can provide. A call counts only when both moves are at least 0.10%. This is a description of what already happened, not a forecast of the next open.

Method notes. Each call is the 1-minute candle close just before 04:00 or 09:00 ET against the candle close in the last minute of the previous session; the real gap is the deployer's oracle price at 09:30 ET against the oracle at the previous close. 1-minute candles can be sparse overnight, so a quiet market's path has holes. The oracle is sampled about once a minute, so the reference close and real open approximate the official exchange prints. Pre-IPO and non-US names are excluded because they have no US open. A call whose candle closed more than 30 minutes before its target minute is stale and excluded from every rate and chart here: 2 09:00 calls and 6 04:00 calls so far; boards still show them with their age. Full definitions: methodology. Data: latest.json · scorecard.json · one file per date at /api/io/premarket/<date>.json.

04:00 call agreed · ±0.10%

78%

of 95 calls · before any US venue opens

09:00 call agreed · ±0.10%

84%

of 100 calls · after 5 h of US pre-market

median |gap| · real

1.42%

oracle at 09:30 vs previous close

nights

2026-08-20 → 2026-10-08 · 3 of 5 symbols with n ≥ 15

04:00 vs 09:00 call · all nights

every rate with its N · nothing shown below n=15

	04:00 call	09:00 call
calls scored · ±0.10%	95	100
agreed · ±0.10% band	78% of 95	84% of 100
agreed · no band	76% of 106	81% of 110
mean abs error	1.41 pp	0.79 pp
beta · gap on call	0.65	0.96

mean abs error = mean |call − real gap| in percentage points and beta = OLS slope of the real gap on the call (no intercept; 1 means the gap matched the call on average), both over every completed pair, the no-band N only the agree rate uses the 0.10% band calls older than 30 min excluded

calibration · call vs open

04:00 call · each dot = one symbol on one night (105 shown) · shaded quadrants = same direction · grey cross = ±0.10% dead band, not scored

45° · gap = callslope 0.65 (OLS)symbol-night

1 point beyond the axis range not drawn; they are in scorecard.json.

by window

agreed with the ±0.10% band, by how long Wall Street was closed

window	n · 04:00	04:00	mae · 04:00	n · 09:00	09:00	mae · 09:00
overnight	74	80%	1.26 pp	81	83%	0.79 pp
weekend	18	78%	1.19 pp	16	88%	0.61 pp
holiday	3	n < 15	—	3	n < 15	—

by size of the call

bigger moves, better calls? buckets of |call move| (each call bucketed on its own move), ±0.10% band

\|call move\|	n · 04:00	04:00	n · 09:00	09:00
0.10–0.25%	9	n < 15	5	n < 15
0.25–0.50%	10	n < 15	9	n < 15
0.50–1%	19	84%	20	50%
>1%	57	89%	66	98%

a coin flip is 50%

by symbol · 04:00 call agreed most often

ranked by the 04:00 rate · symbols with n ≥ 15 · ±0.10% band

market	n · 04	04:00	β · 04	n · 09	09:00	β · 09
SNDK	25	84%	1.05	26	85%	1.00

by symbol · 04:00 call agreed least often