We stopped copying their fills. We studied how they decide.

The raw material is our own copy sim: 43 closed mirrors and 6 still open, every one recorded with the whale's fill, our fill, the exit and the P/L — a record that cross-sums exactly to the sim's realized +$416.64, so the analysis stands on the same numbers published on the Copy Sim page. On top of that we pulled the live activity of the seven wallets this research has covered — 200 activity items each, 1,400 in total, in one clean pull with no rate-limit gaps. The wallets haven't been equally busy lately, so the windows differ (about 1 hour of activity for the busiest wallet, about 14 days for the quietest) — the study says so wherever it matters. What follows is behaviour, not signals: price bands, categories, exits, position-building and hedging, turned into rules we can test.

Favourites win often and pay almost nothing. Longshots pay fortunes and almost never win — our whole longshot profit is a single trade (K27, entered at 0.14). The mid-band is where a follower's money was actually made. And the wallets behave as if they know it: googleplay and monkeymashingkeyboard — two of the week's top earners — keep 100% of their live buys inside 0.30–0.65. googleplay placed 193 live buys, every single one in-band; 62 of them were $500+ conviction buys and 58 of those 62 were NBA.

CS2 carried the esports record (+$216.40 over 9 mirrors, +$24.04 per mirror) while League of Legends filled the most slots and earned almost nothing (+$6.92 over 12, +$0.58 per mirror). Politics and futures were the only losing category — and tellingly, all four of those mirrors ended because the whale sold, not because the market settled. The category table may really be a wallet table (one wallet supplies nearly all the CS2), and the study flags that honestly: it is a hypothesis about skill, not proof about categories.

Nobody stop-losses sports. 39 of our 43 mirrors (90.7%) ended by market resolution, netting +$423.75. The only 4 exits where the whale sold first netted -$7.11 combined — all four were one wallet (BreakTheBank) trimming long-dated positions: -$5.26, -$1.47, -$1.09 and +$0.71. Median hold was about 5 hours overall, about 4 hours for resolved sports. Selling, for these wallets, is portfolio management on slow positions — not a reaction to a losing bet.

They ladder in. A whale's first buy is rarely the whole position: repeat buys on a single market ran to 74 fills (jakepaulthegoat2008), 75 (googleplay) and 168 (Abkhauae). Three of the wallets have never sold at all in the record — they size in and let settlement decide. Median single "conviction" buy ($500+): googleplay $8,547 · Abkhauae $7,559 · BreakTheBank $1,900 · jakepaulthegoat2008 $1,350 · monkeymashingkeyboard $1,309 · UpTheBlues $807.

Some buy both sides. UpTheBlues held both outcomes in 11 of his 52 live events. In our mirror record, three both-sides pairs netted +$27.26 taking both sides — but taking only the cheaper side of those same pairs would have netted +$60.59. Hedging is position management with their bankroll; for a $25 follower it mostly buys the expensive side of a coin flip.

And the certainty style fails its own test. The near-certainty method we run on the Certainty page — the style of the week's most famous grinder — is underwater in our own sim (-$61.25) despite a 96.63% win rate. What the whales taught us points the other way: mid prices, real uncertainty, held to settlement.

In-sample vs out-of-sample, in plain words. In-sample is testing the rules on the same trades they were learned from — like practising on the actual exam questions. It flatters every rule, ours included. Out-of-sample is a fresh exam: trades that happened after the rules were written, that the rule-writing never saw. It is the only honest test — and ours is five trades small, which the results below make very clear.

The honest catch — read this before the tables impress you. In-sample, the unfiltered copy sim ($1,396.80) beat every learned filter except the cheaper-side pair rule ($1,430.13). The reason is one trade: the filters skip the K27 longshot (+$153.57), which alone is 37% of the sim's total profit; without it the raw sim's in-sample profit falls to about +$243 and the pair rule leads by roughly $187. So what do the learned rules actually buy? Look at the COMBINED row: 68.4% win rate against the base's 63.2%, and a worst drawdown of -$25 instead of -$50. The rules buy smoothness and drawdown control — not proven extra profit. Anyone selling you a filter that "beats the market" on the data it was built from is showing you the practised exam.

Five trades is not a test — the combined rules took only 2 of the 5 and both happened to win, which validates nothing yet. That is exactly why the forward sim below exists. For reference, the full week (all 43 closed mirrors): BASE $1,416.64 (+$416.64) · R1 $1,274.45 (+$274.45) · R2 $1,281.60 (+$281.60) · R3 $1,216.40 (+$216.40) · R4 $1,449.97 (+$449.97) · COMBINED $1,282.78 (+$282.78 on 21 trades, 71.4% win rate).

The cross-check. As a second sample we replayed the raw whale activity feed the study pulled — 33 qualifying $500+ BUY signals, of which only 11 had clear exit evidence by pull time. On those 11: BASE $1,135.49 (+$135.49) · R1 $1,061.25 · R2 $1,060.83 · R3 $1,053.19 · R4 $1,135.91 · COMBINED $1,061.25 (+$61.25). The sample leans on a single trade (a whale selling a position at about 0.99) and shares its underlying activity with the mirror record, so it is a consistency check, not independent evidence.

The limits, all of them. About one week of data. The rules were learned partly from the same record they were backtested on (in-sample bias). Winner selection itself is biased — the watch list comes from a weekly profit leaderboard, so we only ever study wallets after a hot week. Fills in both the sim and the backtest are approximations; a real follower fills later and worse — the lagged Reality twin of the raw sim is at -$114.72, not +$416.64. And past performance, especially one week of it, does not predict future results.

No trades yet — the sim started Nov 2, 2026. Every position the Learned Whale sim opens and closes will be listed here in full — whale, market, outcome, the whale's price, our lagged fill, the lag cost, the exit and the P/L — the same standard as the Copy Sim and Certainty pages. Until the first qualifying signal passes the filters, this section stays empty on purpose: an honest blank beats a backfilled record.