The realized scoreboard. Not yet.

What we'll publish

Five numbers, with their definitions stated up front. No invented metrics, no in-sample contamination, no caps that hide drift.

Two cohorts will be shown side by side, labelled separately:

• All published signals — every signal the engine generated and surfaced, regardless of grade.
• Grade A only — signals that passed the ten-check Grade A gate (the cohort we publish as our headline.)

No mixing. The two numbers will diverge — Grade A is meant to be the sharper subset — and the divergence is itself informative.

What we won't publish

Some things are deliberately out of scope.

• Forward-looking projections. No "expect 40% returns", no "target Sharpe 2.5". The engine has internal targets; they are not marketing claims. Forward-looking copy is the kind of thing a regulator builds a case on.
• In-sample backtest numbers as if they were realized. Backtests are useful internally. They are not realized performance. If we ever show a backtest, it's labelled and lives under the hypothetical-performance disclosure.
• Per-asset-class breakdown. Member dashboards show this; the public scoreboard won't. The breakdown reveals which markets we have edge in, which is competitive information.
• Per-equation attribution. The seventeen equations and the brain's weights on them are proprietary. The aggregate is public; the components stay internal.

Where the numbers come from

The data path is short, and on purpose. There's nothing in it that can hide.

1. Engine generates a signal during a scheduled scan. The signal is written to tracked_signals.json with timestamp, symbol, entry, stop, target, grade.
2. The trade closes — at TP, at SL, or at timeout. The outcome is recorded against the same record.
3. Closed trades feed back to the brain via brain.ingest_client_outcomes(). The brain updates equation weights from realized outcomes, not from backtests.
4. The aggregate stats exposed on this page are computed off the same closed-trade ledger. Member dashboards and the public scoreboard pull from the same source.

The single ledger is the discipline. There is no second set of books. If the public number ever stops matching the internal one, the bug is in the page, not in the data.

When the live numbers go up

Three trigger conditions. All three must be met. No partial publication.

1. At least 200 closed trades on the realized ledger. Below that, the win-rate confidence interval is too wide to be useful. A 55% win rate over 50 trades and a 55% win rate over 500 trades are different claims. We want the second one.
2. At least six months of realized window. One regime is not a sample. Six months covers at least one volatility cycle, typically one earnings cycle, and an FOMC turn. Less than that, and we're publishing a momentum-regime number that won't survive the next mean-reversion stretch.
3. Volume-confirmation gate directional for at least thirty days. The tenth Grade A check (volume confirmation) is currently in canary — advisory only. When it flips directional and runs for thirty days clean, the gate has settled and the realized cohort under the gate is representative. Numbers measured during a gate-flip transient aren't the system you'll actually run.

The three together are conservative on purpose. We'd rather publish six months late and hold the number for a decade than publish in week two and explain a regression six weeks in.

Why we don't just publish whatever's there now

A scoreboard you publish too early becomes a number you have to defend against statistical luck. A scoreboard you publish too late becomes a marketing claim with no evidence. We picked the side that errs on credibility.

The honest version of "we are great" is "here is the realized number; here is how we computed it; here is why we trust it; here is the date the cohort started." That sentence needs the cohort to be old enough. It isn't yet.

If you want to see realized numbers right now, sign up for the trial. The dashboard widget shows the same aggregates the public scoreboard will show, with the per-asset-class breakdown the public page won't. Members are committed enough that we trust them with the cohort detail.

What changes when we publish

• The status badge above flips from Tracking — pre-publication to Live, with a "first published" date stamped.
• The five metrics render with their realized numbers, both cohorts side by side, and a "last updated UTC" timestamp.
• An archive grows. We don't take old numbers down when new ones land. The history of the scoreboard is itself part of the scoreboard.
• If a number ever goes down, the page shows it. We don't redact losses. The whole point of publishing is that we can't.