What is a play?
Everything c8alpha publishes traces back to a play — one specific, testable idea about market behavior, made mechanical, validated against real statistics, and then measured in public for as long as it lives.
One idea, written down before it trades
A play starts as a hypothesis in plain language — for example, "when institutional buyers quietly accumulate a stock's options over several days, the stock tends to drift upward over the following weeks." Before it can produce a single pick, that sentence has to become a machine: exact entry conditions, an exact exit rule, and a defined universe of stocks it applies to.
This is the difference between a play and a "trade idea." A trade idea can quietly mean whatever its author needs it to mean after the fact. A play is pinned down in advance — so when we measure it, there is no wiggle room about what was actually tested.
Every play has the same anatomy
- Hypothesis — the one-sentence claim about why the edge should exist.
- Entry conditions — the mechanical trigger. No judgment calls at fire time.
- Exit rule — decided before entry, tested as thoroughly as the entry. A strategy that only works with one lucky exit is fragile, so exits are swept across many configurations during validation.
- Evidence — the validation record: permutation test, out-of-sample split, sample size. Covered in depth on the methodology page.
- Tier — where the play currently stands in the lifecycle below.
The lifecycle: earn your way up, get retired in public
Plays are promoted on evidence and demoted on evidence. The tiers:
Cleared validation on historical data and newest on the roster. Forward signals are recorded and measured, and any published pick carries the Emerging label so the depth of evidence behind it is always visible.
Forward performance is accumulating and being compared against what the validation predicted. Still earning trust.
The deepest live track record — continuously measured against what its validation predicted. If live results diverge from the benchmark for long enough, the play is demoted automatically.
Killed on evidence. The reason is written down and the play's history stays in the record. Kills are permanent unless a genuinely new validation earns a fresh trial.
There is also a watchlist — plays parked because they need data we don't have yet. Parked, not hidden.
What survives
6 plays are live today, each with backtested, walk-forward out-of-sample alpha — and every pick they publish is cryptographically timestamped at publication. Beneath them is the search: most ideas die in research before ever becoming a play, and the 24 full plays that made it and later stopped working were killed in public, with the statistics that killed them — several are written up in full in our research teardowns. A service that never kills anything either isn't testing or isn't telling.
Where the LLM fits (and where it doesn't)
A language model investigates every signal a play produces — recent news, catalysts, sector context — and writes the thesis paragraph published with each pick, along with a conviction score. What it never does is override the mechanics: it cannot invent a pick no play fired on, and it cannot trade. The signal decides; the LLM explains.
Go deeper
The plays page describes each live play's mechanism. The methodology page covers the validation gates in detail. One live play sells options rather than buying stock — cash-secured puts, explained. The difference shows how to judge our claims — or anyone else's.