Crypto · Shadow books

Self-learning research book

Ten research engines, Wave's own signals and outside AI and quant models, record dated views on coins. A $10K paper book weights the engines by their measured record and buys the top-ranked coins. A frozen twin keeps the starting weights forever, so the gap between the two measures the learning and nothing else.

Collecting

Both books recorded on 26 days; 3 weight changes so far.

Pre-registered as LEARN-1. From the test's own records with a 5-minute delay, as of 2026-11-11 15:34 UTC. Paper only, not investment advice.

Does weighting research engines by their measured track record pick better coins than fixed weights?

Learned book: $9,116.65
Frozen twin: $9,116.65
Weight version: v3
Days recorded: 26

Where it stands

The test, stage by stage

RegisteredDoneOct 17, 2026Engines, learning limits and reads fixed before the first run.
Historical testNot usedNone: engine views exist only from the registration on, so it is judged forward.
Forward recordNowOct 16, 202626 days; weights v3.
First readAheadJan 15, 2027Interim read: a report, not a verdict.
DecisionAheadApr 16, 2027

Results so far

The running record

Learned bookFrozen twin

How it works

The rules, as registered

Every engine view is stored with its time before its outcome is known.

Each day both books rank coins by the weighted views and hold up to 5, at most 20% each, for 7 days unless a coin falls 15%.

An engine's weight changes only after 20 matured views and an interval that excludes zero: by at most 25% a step, at most once a week, between 0.25 and 3.

A 25% drawdown stops a book. Both books pay the same modelled costs.

Limits

What this test cannot tell you

The two books are identical until the first weight change, which needs weeks of matured views.
Shadow books: nothing here changes Wave's trades.
Before the interim read the record is a running total, not a verdict.
Paper only: no orders, no real money, and not investment advice.

Related tests