Skip to content

About

basis-collector

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

AfterHours

What is a tokenized stock worth when the market behind it is closed?

Live dashboard: https://lasina67.github.io/basis-collector/ Findings: RESULTS.md · Pre-registered forecast: PREDICTION.md


The gap nobody measures

Tokenized equities trade 24/7. The shares behind them do not.

From Friday's close to Monday's open, AAPLx has a live, moving price and AAPL has none. For about 62 hours every week — plus 17.5 hours every weeknight — the largest tokenized asset class on Solana is priced against a number that no longer exists.

Nobody was publishing what happens in that gap. AfterHours has been sampling it every few minutes since 11 September and has not stopped.

Thirteen days. 1,475 snapshots. 14,750 rows. 252 hours observed with no underlying market open. Ten xStocks, plus eleven pre-IPO tokens across two issuers. Every sample committed to git, so the whole history is timestamped and independently checkable.

What the data says

1. The gap is 3.5× more volatile when the market is shut than when it trades, averaged across ten tokens.

2. Tokens are followers, not leaders. We scored what tokens implied on Sunday night — before CME futures reopen at 22:00 UTC, when nothing anywhere is trading — against Monday's open:

Direction right Rank correlation
Before any venue opened (both weekends) 10 / 20 0.59, 0.60
Once US pre-market was trading 18 / 20 0.85, 0.98

With nothing to copy, direction is a coin flip. Once there is a quote to follow, tokens track it closely.

3. Something does survive. That rank correlation of about 0.6, measured before any venue opened, held on both weekends. Tokens appear to carry some information about which names will move more than others, even when they cannot say which way. Weak and borderline at ten names — but it replicated.

4. A recurring Sunday-morning drop in the same clock window on both weekends, which we cannot explain.

Pre-IPO: the same question without any market at all

A tokenized share of SpaceX has no open market behind it. Not on weekends — ever. The only reference is the issuer's own valuation mark, and the dislocations are an order of magnitude larger than anything in public equities.

At the time of writing, SpaceX trades 21% below its mark at one issuer and 33% above it at the other. Same company, same day, opposite directions.

Several companies are tokenized by both Tessera and PreStocks. Token prices are not directly comparable — they are scaled differently — so each is converted to an implied company valuation:

implied_valuation = mark_valuation × (token_price / mark_price)

Two markets pricing the same private company, with no public market anywhere to tie them together:

Company Apart by
Kalshi 107%
SpaceX 43%
OpenAI 39%

Why this matters

Tokenized equities are increasingly used as DeFi collateral.

A lending protocol liquidating a position at 3am on a Sunday is working with two numbers it cannot trust: a reference price two days stale, and a live token price that — as the data above shows — is not a dependable substitute for it. Meanwhile the gap between them is several times more volatile than on any trading day.

That gap was unmeasured. It is now published continuously.

The feed

Every run publishes machine-readable state:

  • data/latest.json — xStocks: current price, reference, how many hours old that reference is, and how unusual the current gap is for that specific token
  • data/private_latest.json — pre-IPO, including cross-issuer implied valuations
  • data/basis.csv — the full append-only history

reference_age_hours is the field most worth having and the one nobody else publishes.

Repository

collect.py xStocks sampler. Stdlib only, no dependencies.
collect_private.py Pre-IPO sampler, Tessera + PreStocks.
.github/workflows/ Scheduled jobs that run them and commit results.
data/ Every sample ever taken, plus the live feeds.
index.html The dashboard. Single file, no build step.
RESULTS.md Findings, and every claim we have retracted.
PREDICTION.md A forecast committed before the market opened. Unedited.

Method

Scheduled GitHub Actions jobs sample prices and commit them back to this repository. No server, no database — the dataset is the git history, so every sample carries an independent, publicly verifiable timestamp.

On-chain prices come from Jupiter's Price API v3, batched into a single request so every token in a snapshot is priced at effectively the same instant.

Reference prices for xStocks come from Finnhub. While the market is open this is the live last trade; while it is shut, the field holds the closing print — exactly the stale reference a protocol would be marking against. Pre-IPO marks come from the Tessera and PreStocks APIs.

Mint addresses are resolved, not hardcoded. The collector searches Jupiter's token list for each symbol and accepts only an exact match. A hardcoded address that is wrong does not throw an error — it quietly prices a different token, and you find out days later.

Every sample is labelled with the market session and with hours elapsed since the last real closing print. That second field is what makes the central question answerable.

Baselines are session-aware. Each token's normal range is computed from its own history, split by whether the market was trading, because the same token behaves completely differently in the two states.

Data dictionary

Column Meaning
ts_utc Sample time, UTC
symbol Token symbol on Solana
underlying US ticker it references
mint Solana mint address
onchain_usd Token price in USD from Jupiter
ref_price Underlying price, or the last close when shut
ref_prev_close Previous session's close
session OPEN / PREMARKET / AFTERHOURS / WEEKEND
hours_since_close Hours since the last real closing print
basis_bps (onchain − ref) / ref × 10000, signed

Limitations

Stated plainly, because a measurement project that hides its caveats is not a measurement project.

  • Two weekends is not a pattern. Everything above should be read as a first measurement, not an established result.
  • Sampling is uneven. GitHub throttles the schedule to roughly 13 minutes. Real timestamps are recorded rather than assumed, so gaps are visible rather than interpolated, but the series is not a fixed grid. Worst observed gap: 70 minutes.
  • One issuer per asset class for public equities — all xStocks are Backed Finance. Findings may not generalise to other tokenization models.
  • Pre-IPO valuations may not be defined identically between issuers — fully diluted versus post-money, different share classes, different SPV structures. Some of the cross-issuer gap is likely definitional rather than disagreement, and we cannot separate the two.
  • Pre-IPO markets are thin. One token moved 27% in four hours on its own.
  • No liquidity weighting. Every token counts equally in aggregate figures.
  • Sigma is a screening marker, not a probability. The basis distribution has fat tails.

We correct ourselves in public

After the first weekend we published a forecast of Monday's open before the bell, so we could not quietly revise it afterwards. The second weekend forced us to retract three of our own claims, including our headline.

The forecast, the scoring, and every correction are in this repository, unedited, with commit timestamps. See RESULTS.md.

Reproducing it

Fork the repo, add a free Finnhub API key as a repository secret named FINNHUB_KEY, and enable Actions. It will start collecting into your own fork. Nothing else is required — no server, no paid tier, and no key at all for the Jupiter or pre-IPO sources.

Built for

Stocklana, September 2026.

About

basis-collector

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages