# Executability — data and verification Independent measurement of executable depth across on-chain aggregators, tokenized stocks on Robinhood Chain, and prediction markets. **Report:** https://dirkdiggler1026.github.io/executability-report/ ## Contents ``` index.html the report (English, canonical) index.zh.html Chinese snapshot, may lag data/ raw measurements / quotes.jsonl.gz one row per measurement rounds.jsonl canonical hash per round MANIFEST.sha256 checksums snapshot-v18-12r/ frozen 12-round snapshot referenced by the report .rhdepth-v1-defective/ quarantined, see below code/ collect.py the collector build_pool_table.py reproduces the pool selection verify.py independent verification rhchain.py evm.py chain access, zero dependencies stock_pools.json the pool table in use ``` ## What is measured For each tokenized stock, at four notional sizes, at a single pinned block height: **buy in one pool with USDG, then sell the entire position back through that same pool.** The figure reported is how much of the notional comes back. The round trip is deliberately closed inside one pool. An earlier version took the best buy pool and the best sell pool separately and derived a rate between them; that rate belongs to no actual pool and produced recovery figures above 100%, which cannot persist. A same-pool round trip needs no rate assumption and carries its own sanity check: it must be ≤ 100%, and the shortfall is fees plus price impact. ## Which pools Pool selection is part of the conclusion, so it is reproducible: ```bash python3 code/build_pool_table.py --top 8 --probe 10000 > stock_pools.json ``` 1. Full-chain scan of Uniswap V4 `Initialize` events, keeping pools that pair a stock token with USDG and were actually initialised. **Any failed block range aborts the run** — a partial pool table silently produces "this token cannot be sold" for tokens whose real pools were never scanned. 2. Probe every candidate with a $10,000 buy; drop what cannot fill it. Keep the N that return the most. Ranking is by realised output, not by fee tier, pool count or TVL. Pool count is actively misleading here: AMC has the most pools of any token and among the fewest usable ones. ## Verify it yourself Requires an archive node for Robinhood Chain (historical `eth_call`; public endpoints usually retain only ~128 blocks of state): ```bash export RHCHAIN_RPC="https://" python3 code/verify.py python3 code/verify.py --latest ``` The script replays every round-trip at that block height, recomputes the hash, and compares it against ounds.jsonl. Run it from the repository root: it looks for data/ next to the script (or honours RHDEPTH_DATA). Quarantined v1 rounds are excluded by their canon field and reported as defunct (exit code 2), never as mismatches. ## Canonical preimage `rhdepth-v2` Only fields a third party can reproduce exactly at the same block: ``` per line: block|sell_sym|side|size_usd|route |amount_in_raw|mid_amount_raw|amount_out_raw|status integers as decimal strings (uint256 exceeds JS safe-integer range) lines sorted lexicographically, joined with \n, prefixed with "rhdepth-v2" on its own line, then keccak256 ``` `side` is always `roundtrip`. `mid_amount_raw` is the token amount received at the midpoint; without it a third party could check inputs and outputs but not which pool the middle leg went through. Timestamps, network latency and all derived floats are excluded. They cannot be reproduced across runs or languages, and including them would silently break the claim that anyone can recompute and check. Integrity check: ```bash cd data/ && sha256sum -c MANIFEST.sha256 ``` ## Quarantined data Three directories are kept but excluded from every conclusion. They are not deleted: removing a superseded version would make every other claim harder to check, not easier. | directory | rounds | size | defect | |---|---|---|---| | `2026-09-03.pre-pinned-block` | 2 | 44K | not block-pinned, and v1 (broken pool table) | | `2026-09-03.rhdepth-v1-defective` | 14 | 64K | v1: incomplete pool table + cross-pool rate | | `2026-09-04.rhdepth-v1-defective` | 15 | 68K | v1: incomplete pool table + cross-pool rate | The `.pre-pinned-block` set has both defects: calls in a round were not pinned to one block height, and it was also collected on the broken pool table (canon `rhdepth-v1`), so neither the recorded block nor the pools hold. The `rhdepth-v1-defective` sets were collected against an incomplete pool table (55 pools where the chain has 25,122) with the cross-pool rate derivation described above. **They produced published conclusions that were wrong**; those are corrected in the current report. These three directories are shipped once, by hand, and never touched by the daily sync (which only pushes strict `YYYY-MM-DD` directories). `verify.py` excludes them by the `canon` field in each record, not by directory name. ## Limits - Quotes come from simulated Quoter calls, not executed trades. Real execution is affected by MEV, slippage protection and gas, and is generally worse. - USDG-quoted pools only. ETH/WETH pools are not included; TSLA is known to have additional depth in an ETH pool, and other tokens may too. - The pool table is internally consistent (fail-loud scan, zero failed ranges) but has **no external cross-check**. DexScreener counts uninitialised pools in its liquidity figures and is not used as a reference; the block explorer's API is behind anti-bot protection. - The baseline spans hours, not weeks. It does not extrapolate across weekends or earnings windows. - No dependencies: `code/evm.py` implements keccak-f[1600], static ABI encoding and eth_call using only Python 3 and curl.