Until today a NextScalp alert could arrive with a full trade plan attached: an entry, a stop loss, two take-profit levels, a risk-to-reward ratio and a position size calculated from your capital settings. As of this release it does not. No alert prints an entry, a stop, a target, an R:R, a size or a LONG/SHORT instruction any more.
This is not a feature being trimmed for cost. It is the product being brought back in line with the standard we set for ourselves.
The standard we could not meet
NextScalp keeps an internal paper-trade book. Every signal the bot fires opens a virtual position at the live market price, resolves it by cron, and accumulates statistics: win rate, expectancy, directional accuracy, maximum favourable and adverse excursion. The whole point of that book is to answer one question honestly - does this actually have an edge?
The answer, for the automatic trade plans, is: we cannot show that it does.
- On a one-to-five-minute horizon, the formations the bot detects behave close to a coin flip once fills are modelled realistically. Earlier results that looked positive were measured against a patient limit entry at a retest the market frequently never gave. When we replaced that with an entry at the live price you would actually have paid, the sign flipped on several cohorts.
- The samples are small and the measurement window is short. At the kind of Sharpe ratio these cohorts show, a few weeks of data carries almost no information either way - including when the numbers look good. Favourable noise is still noise.
- There is a timing problem on top of the edge problem. A scalping-horizon plan that arrives after the move has already extended is not a plan; it is a description of something you missed.
Our own rule has always been that a strategy has to prove itself on paper and then survive a small live phase before the machine is allowed to act on it. Shipping an entry and a stop to you is the machine asking you to act on it. Under that rule, an unproven plan should never have gone out, and now it does not.
Note the exact claim: we cannot defend the quality of those plans. That is different from "we proved they lose money". We are not making the stronger claim, because the data does not support that either.
What an alert looks like now
Every alert is an informational market-structure event. It tells you what happened, where, and how fresh it still is:
- the event and the observed direction - a break above a level, a rejection below one, a sweep, an approach, a volume anomaly;
- the level itself, its timeframe and how many times it has been tested;
- the context you would otherwise assemble by hand: higher-timeframe structure, volume multiple, open interest, funding, order-book walls, volume profile;
- a freshness line: the live price, how far it has travelled from the level, that distance in units of volatility (NATR), and how long ago the event fired.
And for continuation breaks - the ones where being late genuinely destroys the read - the bot now stays quiet instead of pinging you. If price has already extended roughly a full volatility unit beyond the level by the time the alert would go out, it is not sent at all. Events where moving away from the level is the content of the event, like a sweep or a fakeout, still arrive; there the distance is information, not staleness.
The chart, the watchlist, the density map, the funding scanner, /focus and the live monitor are
unchanged. What is gone is the instruction.
The AI co-pilot changed too
It would be incoherent for the deterministic engine to stop giving directions while the AI button underneath the same message kept printing "LONG, entry here, stop there". So the AI attached to alerts no longer emits a direction or price levels either. It explains what it sees in the structure. Over the next release it moves fully to a structural read: what is confirmed, what is broken, which levels are active, what contradicts what, and what changed since the last look.
What happens to plans under the hood
They keep being computed - and never shown. The plan builders, the gates that kill weak plans, and the paper-trade book all continue to run as an internal research system: hypothesis, simulated execution, outcome. That is the only way we will ever be able to say a particular setup has an edge rather than hoping it does.
If something in there ever does clear the bar, it graduates into the autonomous strategy book, where the machine trades it with its own money and its own risk. It does not come back to you as a suggestion.
If you are paying for this
Two things. First, the plan block was explicitly part of what the paid tier advertised, so if it was
the reason you subscribed, that is a change to what you bought: message us and we will refund or
grandfather you, no argument. Second, the paid tier is now about depth rather than instructions -
the full context on every event, the chart, the whole set of alert types, your own watchlist,
/focus and the live monitor.
What comes next
- A risk calculator you drive: you give it your intended entry and your own invalidation level, it gives you a size from your capital and risk settings. Your levels, not ours.
- The AI in full structural mode, quoting only levels the engine actually computed.
- Descriptive statistics about how these events behave - how often price returns through the level, the median excursion each way, how long an event takes to resolve - published as research, with the sample and its limitations stated, never as a win rate. The first set is up: How Our Events Actually Behave.
We would rather ship a bot that tells you the truth about the market and lets you decide, than one that tells you what to do and cannot defend why.