We are scored too

A check that rates other people’s records should keep one of its own. Each call below was written down with a date, and with the result that would show we were wrong, before any outcome existed. Every reading is published here, whatever it shows. A reading date that passes without a result shows as overdue.

Calls with a test

1 · Papertrade’s leaderboard does not predict what comes next Pending

Written 2026-10-16, before the freeze (commit 2cd7deb, then the frozen file). Frozen 2026-10-16 03:01 UTC. Readings 2026-10-23 (interim) and 2026-10-30 (primary).

What would count against us: at the 2026-10-30 reading, the 24-hour PnL top 100 beats a seeded random control by more than 15 percentage points in the share of wallets that made money after the freeze, with a median above zero. Then the board carries information, and we will say so. Fewer than 30 wallets trading after the freeze reads as “not enough data”, not as a result. Full text: PREREG.md · audit 010.

2 · Hyperliquid’s top boards hold little forward skill Pending

Frozen 2026-10-14 09:48 UTC: every address on that day’s all-time PnL and 30-day ROI top 100s, 200 wallets, scored only on what they do after the freeze. Reading 2026-12-09, the first day a wallet can have eight weeks of record after the freeze.

Written 2026-10-18. The list and its rule were frozen on 2026-10-14, but the result that would count against us was not written down then; it is written here, four days later and before any wallet can be scored.

What would count against us: of the cohort wallets with at least eight weeks of record after the freeze, more than 10% reach t ≥ 2 on that record at the 2026-12-09 reading. Luck alone gives about 2.3%; if most of them had a true Sharpe of 3, roughly one in five would (normal approximation; the small-sample correction makes it somewhat lower). More than 10% would mean these boards pick up more skill than we say, and we will say so. Reported alongside, with no pass mark: for wallets that were promising before the freeze (t ≥ 2) against the rest, the share with t above zero after it. Fewer than 30 wallets with eight weeks reads as “not enough data”, and the reading moves to 2027-01-06.

3 · Our score orders the same wallets Pending

Written 2026-10-20, six days after the freeze, before any wallet has eight weeks of record after it. Same cohort and reading day as test 2. Test 2 asks whether the boards hold skill; this asks whether our own score can tell better wallets from worse ones, which any use of it as a gate depends on.

What would count against us: the rank correlation between each wallet’s t before the freeze and its t after it is zero or below. We would then stop describing the score as usable for gating money. It counts for us only if the correlation is above zero with a 95% interval (resampling wallets) that excludes zero; anything between is inconclusive. Reported alongside: the same correlation for profit rank before the freeze, and the gap between the two, claimed only if its interval excludes zero (the retrospective test did not clear that). Full text: COHORT_PREREG_ORDERING.md.

Lists we froze

Each file is in the repository with its full history, so a change after the freeze would show in git and break the hash.

Readings we cannot edit

Written to the ERC-8004 reputation registry from 0xF02cF39d6b91Ac62B1D21d0F1C39335118c1F9A1, the project’s only feedback address. These are readings, not predictions: t on the agent’s record since it registered. How they are built.

Errors

Every number on the site is recomputed by scripts in the repository, and every change to a published page is in its git history. If you think a number is wrong, open an issue with the page and the number; corrections are made in public and listed here with the date and the commit. Corrections so far: none. This list is kept by hand; every issue ever filed is public, so you can check it against them.