Lab Notes

A small pool of algorithms gives a clean answer, but it can be clean for the wrong reason. If every algorithm sees the same narrow pattern, the output may look confident while saying very little.

A huge pool has the opposite problem. It may contain useful disagreement, stale behavior, duplicate logic, and beautiful-looking backtests that do not survive live timing. More voices only help when the selection layer knows which voices still matter.

The recent Signal Key work moved us toward that discipline. The Lab can generate large contour families, single towers, group towers, and observer candidates. The public layer should only accept the pieces that survive train, holdout, live-safe timing, settlement, and fresh observation.

Scale is a discovery tool. It is not the product. The product is the sealed key cell that made it through the policy before the game started.

The easiest way to make a sports signal product look busy is to publish something every time the board opens. That is also the easiest way to turn the product into noise.

The Signal Key is built around a different idea: a cell should be filled only when the active policy has enough reason to accept it. Some days will produce several cells. Some days may produce one. Some days may produce none.

That silence matters. If the research layer sees weak agreement, suspicious odds bands, one-sided behavior, or a shape that only worked in a narrow historical pocket, the correct public decision may be to wait.

A quiet key is not exciting in the moment, but it protects the record. The archive should show both action and restraint.

A public signal has two jobs that pull in opposite directions. It must be committed before the result is known, and it must avoid exposing the actual pick before the intended reveal window.

That is why the key uses a sealed file and an OpenTimestamps proof. Before the event begins, the sealed signal file is hashed and timestamped. The site can show that a commitment exists without showing the match, side, or units.

After settlement, the readable signal file is published next to the timestamp proof. The user can check that the revealed file matches the earlier commitment, and the result audit can be read in the same key context.

This is not about proving that every signal will win. It is about making the record hard to rewrite after the game has already told us the answer.