An evidence-gated signal system for short-term equity research.
Pulsenal scores movement and direction as two separately validated axes. A market is only shown a published number after it clears a purged, chronological walk-forward gate — where the evidence falls short, the system reports that instead of a number.
Overview
Pulsenal follows short-term equity signals from the moment they are generated through a fixed forward window, and separates two claims that are commonly merged into one score: whether a stock is likely to move, and whether that move is likely to be favorable. Each claim is validated on its own evidence and published only where that evidence supports it.
- System type
- Signal research and evidence tracking. No order execution path exists in the system.
- Primary markets
- United States and Korea equities, tracked from 30 minutes to five trading sessions after each signal.
- Secondary pools
- A small isolated ETF pool and a small isolated crypto pool, evaluated separately from equities.
Validation method
Every published figure passes through the same four-step discipline before a user sees it.
| Step | What happens |
|---|---|
| Two axes, never one | Movement and direction are scored and validated independently. A market can be shown as likely to move without ever being shown as likely to move up — those are separately measured claims. |
| Purged walk-forward | Validation runs in five expanding, chronological folds with an embargo window around each holdout boundary, so no training row shares a label window with the data it is tested against. |
| Fixed statistical gate | A head publishes only when its mean walk-forward AUC clears threshold in at least four of five folds, and its calibrated Brier score beats a constant baseline. |
| No self-promotion | Challenger models are evaluated continuously against the same gate but cannot replace the active model automatically. Promotion is a manual, reviewed decision every time. |
Current coverage
Reported market by market and axis by axis, rather than blended into one number.
| Market | Move | Direction | Note |
|---|---|---|---|
| United States | Validated | Validated | Both axes cleared the gate |
| Korea | Validated | Withheld | Direction did not clear the gate on measured evidence |
| ETF pool | Isolated | Separate evidence, not blended with equities | |
| Crypto pool | Isolated | Separate evidence, not blended with equities | |
Operating constraints
Structural limits, enforced in code rather than stated as policy.
| Constraint | Detail |
|---|---|
| No automatic orders | No broker connection and no order-execution path exists. Every position decision is entered by the account holder. |
| No manufactured confidence | A score is shown only where validation supports it. An unvalidated market is ranked, never given a number implying more certainty than exists. |
| No silent promotion | Every challenger model runs in shadow only; nothing reaches production scoring without a logged manual review. |
| No blended evidence | Stocks, ETFs and crypto keep separate outcome records and separate calibration. |
Provenance
Pulsenal is developed and maintained under a single continuous internal research process. Every score currently in production has a recorded training date, feature set and validation report behind it, and every change to what is shown to a user is logged against that record.
Access is deliberately limited while the system is under active development. This is not a public product. There is no public sign-up, no email verification flow and no self-service account creation — every session requires a credentialed access key issued by the system operator.
Access
If you were issued a key, continue below. Unrecognized keys are rejected without further detail, and repeated failed attempts pause new access from that device.