The most expensive habit in trading is conviction on a single signal. The insider bought, so the stock is going up. The chart broke out, so this is the trend. The model said buy, so we buy. Each of those sentences is a story, and each story is built on one witness. One witness, in any other domain, would not be enough to convict. We don't see why markets should be different.
A single signal is, by construction, a coin flip with edge. The very best layers we have studied — Form 4 cluster buys, post-earnings drift, activist filings — clear a 50-55% hit rate on a meaningful holding period. That is real, and it is not nothing, but it is not a portfolio. A 53% hit rate with a 1:1 reward profile produces a Sharpe of roughly 0.4 before costs. After commissions, slippage, and the cost of being wrong eight times in a row, you are running at break-even on a strategy that someone is going to call you brilliant for.
Why twelve, not four, and not forty
The instinct is to add layers until the signal is silent. That is the wrong question. Each new layer adds information only if it is independent of the layers already counted. Two technical-momentum layers do not give you twice the information — they give you the same information twice, and they correlate the noise. The art is finding witnesses that see different things.
Our twelve layers were chosen because their underlying signals come from genuinely different sources. Insider buys are reported by individuals on Form 4. Activist filings come from registered 13D / 13G entities. Catalyst flow comes from government databases (USAspending, openFDA). Politician trades come from the STOCK Act feed. Quality and valuation come from accounting fundamentals. PEAD comes from market microstructure. No two of these layers are downstream of the same primary source — when they agree, they are agreeing despite seeing the world through different eyes.
How we count to twelve
Each layer produces a sub-score on a normalised 0-100 scale. The sub-scores are weighted by their historical information ratio — layers that have been informative get more weight, layers that have been noisy get less. The weights are re-estimated quarterly on a rolling 36-month window. We do not let the weights drift on monthly noise; we do not let them ossify across regime changes.
Each layer also carries a decay function. Insider buys are most informative in the first 30 days after the filing. PEAD decays over 60 days. Activist filings can stay informative for two quarters. The composite score is the time-decayed, IR-weighted sum. A name that scores 38 today is a name where the chorus is loud right now — not a name where the chorus was loud six months ago.
The boring half of the job
On most market days the chorus is a soloist. One layer flashes, the other eleven shrug, the composite score sits at twelve, and there is no position to take. The boring half of the job is sitting through that and not making something up. The fund's edge is measured not by the trades we put on but by the trades we did not put on when we were tempted.
The chorus is loud or it isn't. We do not pre-front-run the chorus. We do not lean in when only the soloist is singing. We wait.
That is the trade. Everything else is execution.