Roundtrip measures whether tokenised stocks on Robinhood Chain can actually be sold, and makes every published round recomputable.
Executability Report — whether tokenised stocks can actually be sold, measured and published.
A collateral parameter depends on what comes back when you sell, not on the quoted price. So this project pins a block and quotes the sell side — the whole position back to USDG, through the best route we can enumerate — for the tokenised stocks on Robinhood Chain, at four sizes from $100 to $100,000. Venues are found by following counterparty flow rather than a factory pattern; we state the method and the windows, and do not claim that set is exhaustive. Each round is canonically serialised, hashed and published with a checksum, and the round hashes are anchored in an append-only EvidenceLedger — testnet 0xc4f7…6fb9 (chainId 46630) and mainnet 0x7f54…fab6 (chainId 4663) — so a published number cannot be quietly rewritten. Two series run: a 30-minute round trip since 2026-09-04, 22 days and 1,042 rounds published through 2026-09-25, and an hourly one-sided series, registered 2026-09-25, whose forward window opened 2026-09-26.
The question turned out to matter more than the answer, and our own first answer was wrong twice over. At block 54,088,399, QQQ at $100,000 came back at 0.03% — through the one pool a v4-only table had picked. The same held-book construction, in the pool that can serve that size, returns 99.74%. The sell leg alone returns 99.87%, because a held-book round trip is charged in both directions: 0.133% on the buy plus 0.131% on the sell. A forced seller walks one side.
Anyone can check it: python code/verify.py 61129566 recomputes the round hash from the published files and compares it with the record — recomputation needs an archive endpoint; the on-chain anchor does not. Fourteen of this project's own errors are published too, numbered, with which were caught before publication and which were not.
<p>Dated, because the measurement line predates the buildathon and it matters which is which.</p><p>2026-09-04 — the measurement line started, ten days before the buildathon. It is the first thing that is true of this project and it is not a buildathon artefact.</p><p>2026-09-14 — the experiment was pre-registered before the forward window opened. Four commits, all on the 14th (git log 68b2fad^..feb560a): an absolute floor trigger, five stable token names and the forward window; canon pinned to a serialisation version; the sampling rate and the drop policy; and block selection — recorded in addendum 2 as deliberately unpinned, then pinned in addendum 3, whose own message says addendum 2's section 6 "claimed more than it argued". The point was to write down what would prove the result wrong before any forward data existed, and to correct the registration itself while correcting it still cost nothing.</p><p>2026-09-15 — the forward window opened, and EvidenceLedger was deployed to Robinhood Chain testnet at 0xc4f7c2ed489d9f521d65b43cc4929d3c642c6fb9 (chainId 46630), with its commit script.</p><p>2026-09-19 — the backfill was broadcast: 610 rounds committed on chain in 10 batches; the ledger watermark moved from 0 to 64,907,249.</p><p>2026-09-23 — a second EvidenceLedger was deployed to Robinhood Chain mainnet at 0x7f5446b920e09531f443ce951076cbaed09dfab6 (chainId 4663), and 850 published rounds were backfilled through block 69,196,861 in fourteen transactions.</p><p>2026-09-25 — a second series was pre-registered with its own canon (rhdepth-oneside-v1), measuring the sell side alone — the side a forced seller walks. Its forward window opened the next day at 00:00 UTC and runs hourly, alongside the 30-minute series, which continues unchanged.</p><p>2026-09-26 — the published material was brought in line with the corrections: the demo film was re-cut, the four submission images replaced, and the artifacts the cards cite published in the repository under measurements/oneside-depth/, so a reader following a citation lands on the data rather than on a filename. The finding we had published earlier — that failure came in three shapes — was withdrawn: it was read off a series that quoted one venue and priced a round trip against a book that was not allowed to move.</p><p>Also during the window: the verifier was rewritten so its user-facing output is English and so that it refuses explicitly rather than hanging when pointed at a non-archive RPC endpoint (the refusal is measured, not asserted); a canonical serialisation specification (rhdepth-v2) was published so a second implementation need not read the reference code; the error archive grew to fourteen numbered entries — seven caught before anything was published, seven after — each with what it was and how it was caught; and the field we proposed to the SEC was corrected to one-sided depth, because a round trip blends the buy and sell directions while a forced seller walks one. The fourteenth was found on 2026-09-27: a block's state root had been reported as a revert selector, and it had already been published in a pre-registration before anyone understood it.</p><p>The measurement line itself has run continuously since 2026-09-04: 1,042 rounds over 22 days through 2026-09-25 (34 on the first partial day, 48 a day after), plus three quarantined directories kept rather than deleted. The hourly series that opened on 2026-09-26 exposed a venue class the collector can enumerate but not quote — reserve-based pools, for which it has no swap path — and every round now reports those as present_not_quoted rather than dropping them silently.</p><p></p>
<p>No external funding. Self-funded.</p>
