Product

The landing page, and why it is ordered this way

Rebuilt 2026-09-03. It measured 7,123px — 7.9 screens at 1440x900, 2,878 words, about a twelve minute read — and the most persuasive fact on it, the dollar cost of landing in the wrong pool, sat 3,600px down as section four of seven.

Now 5,922px, 6.58 screens, with four themes added that were not there before. The order:

Detail opens, it does not scroll. Every section is scannable at a glance and every piece of depth sits behind a [ … ] toggle on the same page — the full pool ladder, all twelve tokens, all nine columns, the multiplier arithmetic, each engine's published rules. Nothing was moved elsewhere and nothing was deleted. The uncapped PriceLadder in the execution check alone was 920px of the old page; it is a PoolStrip now, with the ladder one click away.

The hero names both numbers. It shipped with a 546 bps spread in 200px type and a −4 bps basis in 12px type directly beneath, neither labelled, which read as one measurement contradicting itself. They answer different questions and the page now says which: spread asks whether the pools agree with each other, basis asks whether the token is priced correctly against the real world. The hero also carries the dollar consequence, computed through the same executionQuote() the section below uses, so the two can never disagree.

Every claim about an unbuilt feature carries its gate. See docs/roadmap.md — two claims on this page were written in the present tense for features that do not exist.

What basis measures

A Stock Token on Robinhood Chain is an ERC-20 giving economic exposure to a listed share. It has two prices at all times, set by completely unrelated processes.

The reference price is what the underlying share is worth, published by the issuer, times the ERC-8056 multiplier. The pool price is whatever the last person to trade that token in a Uniswap pool happened to pay. Nothing forces them to agree.

basis   = ( deepest pool / reference − 1 ) × 10 000     // bps
spread  = ( dearest pool / cheapest pool − 1 ) × 10 000 // bps

Two numbers, deliberately. basis asks "is the token priced correctly against the real world". spread asks "do the pools even agree with each other". On 2026-09-03 NVDA was perfectly priced on aggregate — 2 bps — while carrying a 571 bps disagreement internally. Publishing only the first would hide the whole story.

What it is not

• Not a price feed. basis does not tell you what a token is worth. It tells you how far two existing prices have drifted apart and how much confidence each deserves right now.
• Not a router. It never signs a transaction, never holds funds, never takes custody. It shows prices and links to pools.
• Not a signal service. A wide gap is an observation, not an instruction. Whether it is an opportunity or a trap depends on depth, which is the hardest part — see below.

Sessions, and why a reading is meaningless without one

The pools trade all 168 hours of the week. The underlying trades about 32.

Chainlink's documentation on these feeds is explicit: off-hours they "may hold the last published price" and "do not have heartbeats during off-hours". The contract still answers. The answer just stops changing.

A 4% gap during a regular session and a 4% gap at 03:00 on a Sunday are not the same event, and basis will never label them the same way.

Why this matters commercially. On 30 August 2026 a token launched against AMC pushed the on-chain AMC price 35× past the real one, with the equity market shut and no market maker on the other side. The launchpad's own announcement the next day described adding "risk labels so traders see a premium before they buy rather than after". That is the gap this instrument fills, described by an incumbent.

Flags

Published so they can be argued with. A flag is a function of the numbers and the session, with no discretion in it.

DARK outranks everything. A token cannot be TIGHT at 2am on a Sunday because there is nothing to be tight against. Getting this precedence wrong is the most misleading thing a readout can do, and it is the rule a naive implementation breaks first.

Pool disagreement gets its own scale, because it is a different problem with a different cause: < 100 normal, 100–400 notable, ≥ 400 flagged.

Depth, and the net problem

This is the difference between a screener and an instrument, and the part most likely to be got wrong — including by us, in a first attempt that had to be thrown out.

A published spread is gross: the difference between two quoted prices. What you can actually execute is smaller, sometimes much smaller, sometimes negative. Swap fees on every leg, price impact against real depth, and routing hops when two pools do not share a quote asset all eat it.

An early model applied constant-product maths to pool reserves and produced +83% on a $2,000 trade. The cause: these are concentrated-liquidity pools. The reserve ratio does not give the price and TVL does not give the depth. It follows that nobody can compute an executable number from a screener API — including every dashboard currently quoting these spreads.

So basis:

1. Publishes gross, labelled gross. Never dressed up as achievable.
2. Publishes depth alongside every price. A 35× print on a $4k pool and a 2% gap on a $3.8M pool are different facts and never share a column.
3. Applies a size guard. A pool is only offered for a given notional if it holds at least four times it. Below that the tool says so rather than quoting a price it cannot stand behind, and above a tenth of the best pool's depth it warns that the quote is the top of book, not your fill.

Which pool counts as "the" pool

basisBps compares the reference to one pool, so which one is chosen is a methodology decision, not a detail.

Ranking on TVL alone is wrong. Parked liquidity nobody trades against holds a stale price, and depth-ranking picks it. Observed live on 2026-09-03: TSLA's largest pool held $418,744 but had traded $8,835 in 24 hours — 2% turnover — and quoted 356.24 while the reference was 383.29 and every actively traded pool sat at 382–384. Ranking on depth would have published a −706 bps dislocation that did not exist.

So a pool has to be both deep and awake: among pools above the dust threshold, deepestPool() takes the deepest that turned over at least 10% of its own depth in 24 hours. Where no pool reports volume — the stored capture does not carry it — it falls back to depth alone. With the rule applied, that same TSLA reading is −25 bps, TIGHT.

This is the same class of error as comparing a pool to a raw share price: a number that looks like a finding and is actually an artifact of how it was computed.

Who this is for

People about to overpay — the larger audience by far. Someone buying $100,000 of TSLA who lands in the wrong pool pays $3,669 more for the identical tokens (measured live, 2026-09-03). Far more people lose money this way than will ever make money on the gap. basis is the thing you check before you press buy.

Arbitrageurs — real but contested. The gaps do close: QQQ went from +217 bps to +172 bps in 46 minutes while we watched. Capturing that means racing bots on a sequencer with first-come-first-served ordering, so it is a latency race rather than a capital one.

Protocols — a launchpad's peg guard, a lending market's collateral haircut, a router avoiding a dislocated pool. This is what the on-chain oracle in docs/token.md would serve, and it ships only when a named consumer wants it.

The structural finding

Once pools are ranked properly, the dislocation is not where people assume. The deep stablecoin pools cluster tightly — SPY's $3.8M USDG pool and its next four largest sat inside a 0.5% band. The outliers are almost all stock-token-against-stock-token pairs, where both legs are tokens and neither side is anchored to anything.