ochinimus/xstocks-peg-ticks

xStocks Peg Deviation Ticks

31,228 paired observations of tokenized US equities trading on Solana against their underlying reference price. 11 symbols, 19–29 July 2026, ten days at a nominal 300-second cadence with 99.12% coverage.

Each row is one moment in time where both prices were known: what the token was worth on-chain, and what the equity it tracks was worth off-chain. The gap between them is dev_bps.

As far as I can establish, no comparable dataset is public. Tokenized equity prices are available, and equity prices are available, but the paired series — sampled on a single clock, including the hours when the equity market is shut — is not something you can currently download anywhere.

Why the closed hours are the point

US equities trade roughly 6.5 hours a day, five days a week. These tokens trade continuously. 78% of the rows here (24,485 of 31,228) fall in windows where the reference price is frozen and the token is still moving.

Median absolute deviation, split on whether the underlying was trading:

symbol	n open	median open (bps)	n closed	median closed (bps)	ratio
MSTRx	613	13.6	2220	103.4	7.61
CRCLx	613	14.4	2220	107.4	7.46
TSLAx	613	7.5	2222	48.2	6.43
GOOGLx	613	23.4	2229	52.0	2.22
METAx	613	24.2	2229	40.3	1.67
AMZNx	613	15.6	2230	22.4	1.44
QQQx	613	27.1	2227	38.5	1.42
COINx	613	45.5	2221	64.3	1.41
MSFTx	613	44.7	2232	61.6	1.38
SPYx	613	56.9	2228	64.4	1.13
AAPLx	613	27.8	2227	22.2	0.80

Ten of eleven widen when the reference stops updating. Three carry most of it, and they are not a random three: MicroStrategy, Circle and Tesla are the most crypto-correlated names in the basket. Bitcoin moves over the weekend, those equities would move if anyone could trade them, and the last printed close cannot reflect it. The token reprices and the reference goes stale.

TSLAx makes the case cleanly. It is the tightest token on the board while the market is open — 7.5 bps median, better arbitraged than SPYx — and widens six-fold the moment the reference freezes. That is not a poorly maintained peg.

AAPLx inverts at 0.80, which is what you would expect from the name least likely to move on overnight crypto news.

For MSTRx and CRCLx the effect is present on all ten days, with a per-night floor around 35–42 bps, so it is not an earnings-week artifact. Both drift upward across the final three nights (MSTRx 186 → 297 → 330, CRCLx 176 → 196 → 357). Ten days is not enough to say whether that is a regime change.

Read this before drawing conclusions

The denominator is stale by construction. During closed sessions the reference is the last trade before the bell. Elevated dev_bps in those rows partly measures information the reference structurally cannot contain. Treating it as a mispricing, or as something capturable, misreads what was recorded.

Single venue. Every row is Raydium. Deviations are Raydium-specific and say nothing about the same token elsewhere.

Single window. 19–29 July 2026, which overlaps Q2 earnings. Ten days.

Reference sources are mixed. 31,163 rows from Alpaca, 65 from a Yahoo last-price fallback (ref_source). The fallback rows are 0.2% of the set and carry a different staleness profile.

One symbol was collected and removed. NVDAx was dropped before publication after its reference series failed validation. It is absent rather than silently wrong, and no NVDAx rows appear here.

Deviation is signed. dev_bps keeps its sign; the tables above use absolute values. Directional analysis is left to you.

Schema

column	type	meaning
ts	int	Unix epoch seconds, UTC. Authoritative timestamp.
symbol	str	Token symbol, e.g. TSLAx.
onchain	float	Token price in USD from the pool.
ref	float	Reference equity price in USD.
ref_source	str	alpaca or yahoo-last.
session	str	open, premarket, afterhours, overnight, weekend.
dev_bps	float	Signed deviation of onchain from ref, basis points.
liq_usd	float	Pool liquidity in USD at observation time.
dex	str	Venue. raydium throughout.
ts_iso	str	ts rendered as UTC datetime. Convenience column.

Session counts: overnight 7,931 · weekend 7,050 · open 6,743 · premarket 5,808 · afterhours 3,696.

Per-symbol counts run 2,833–2,845, so coverage is even and no symbol needs a gap explanation.

Load

import pandas as pd
df = pd.read_csv("hf://datasets/ochinimus/xstocks-peg-ticks/peg_ticks.csv")
df["closed"] = df["session"].ne("open")
df.groupby(["symbol", "closed"])["dev_bps"].apply(lambda s: s.abs().median())

Never pool dev_bps across symbols. The per-symbol spread runs from 7.5 to 57 bps during open hours alone, so a pooled median describes nothing that exists.

Collection

Collector on a nominal 300-second cadence, on-chain price read from the Raydium pool and paired with the reference quote at read time.

Coverage against the 300-second grid is 99.12%: 31,228 observations against 31,505 slots, so 277 are missing. These are per-symbol reference-fetch failures rather than collector downtime — MSTRx and CRCLx have no gap larger than 300s, while TSLAx has a single 8,400-second hole where the reference feed stopped serving that symbol. Per-symbol coverage runs 98.68% (TSLAx) to 100% (MSTRx, CRCLx). Five intervals fall below 300s, from process restarts sampling on boot. There are no duplicate (symbol, ts) keys. Session labels are assigned from the US equity calendar at observation, not backfilled. Rows are written only when both sides resolved, so there are no imputed prices.

License and citation

CC BY 4.0. Use it commercially, build on it, retrain on it — attribution required.

@misc{xstocks_peg_ticks_2026,
  title  = {xStocks Peg Deviation Ticks (Solana / Raydium)},
  author = {ochinimus},
  year   = {2026},
  url    = {https://huggingface.co/datasets/ochinimus/xstocks-peg-ticks}
}

Collection is ongoing. Corrections and pull requests welcome via the Community tab.