Notebook 005 - The Wrong Guardrail

The next lesson came faster than expected. On Thursday, Onyx learned that pullback-only was safe but incomplete. Strong names can run without giving the discount we wanted, and if the daily plan authorizes only a pullback, the trader has to sit still. That was the right miss. It was disciplined. It also exposed a gap in the plan language.

So on Friday, we gave the plan a second path. Not for everything. Not for every green candle. Not as a new default strategy. Just a guarded opening-range momentum backup for selected names where the homework already supported the idea that strength might continue without a clean reset. That was the right direction. But the first version had the wrong guardrail.

This was a paper-trading day, and this is not financial advice. The point is what the system learned about designing constraints.

A good guardrail should block the behavior you are trying to prevent. It should not block the behavior you are trying to authorize. On Friday, one did.

The backup existed

This is what makes the lesson different from Thursday. Thursday's clean miss was simple: the plan liked a strong leader, but only authorized a pullback. The pullback never came, so Onyx did nothing. That was not an execution bug. It was a narrow plan. Friday was more interesting because the plan had already changed.

For a small number of stronger candidates, Onyx had permission to look for an opening-range momentum setup. The idea was controlled: wait for the first range to form, require strength above that range, require price above VWAP, require confirmation from the tape, use smaller size, and keep no-loss exits intact. That was the right shape. It kept the daily plan as the source of truth. The trader still did not get to improvise. The plan had to name the symbol, the setup, the window, and the constraints before the trade could happen.

The system was not being told, "Buy strength." It was being told, "If this specific stock, on this specific day, with this specific market backdrop, proves strength in this specific way, then a smaller backup entry is allowed." That is the difference between flexibility and chasing.

The runner it blocked

RKLB became the clean test case. The setup was not invisible. The morning work had already identified it as an event name worth watching. The move had a catalyst behind it. The stock held its opening structure, cleared the opening range, stayed above VWAP, and had positive momentum confirmation. That is almost exactly what an opening-range momentum backup is supposed to notice.

But Onyx did not enter. The reason was not that the trader missed the breakout. The reason was not that the stock was outside the approved list. The reason was not that the plan lacked a backup path. The reason was that one guard said the move was too hot. The opening-range RSI was above the plan's ceiling, so the backup rejected the trade.

On paper, that sounds sensible. RSI is supposed to warn us when something is stretched. We do not want Onyx buying exhaustion candles just because volume is loud and the chart is vertical. A momentum backup without any anti-chase protection would be reckless, especially inside a system that refuses automated loss-taking sells.

But in practice, this particular guardrail was too blunt. By the time a real catalyst runner confirms an opening-range breakout, RSI may already look hot. That does not automatically mean the trade is bad. Sometimes it means the stock is doing the exact thing we asked it to prove: strong demand, clean continuation, no meaningful reset. If the guard rejects every hot reading, it may reject the strongest leaders by design. That is what happened.

Momentum needs confirmation and anti-chase

The mistake was mixing up two different jobs. A momentum backup needs confirmation. It needs evidence that the move is real enough to consider. That is where things like opening-range breakout, VWAP, volume, and trend confirmation matter. They answer the question: is there actual demand here, or are we just staring at a random spike?

A momentum backup also needs anti-chase protection. It needs evidence that the entry is not absurdly far from the structure it is trying to trade. That is a different question: even if demand is real, are we paying too much for it right now? The first job is about strength. The second job is about distance. The RSI ceiling tried to do both. That is why it failed.

High RSI can be useful as a warning, but it is not the same as distance from structure. A stock can have a hot RSI and still be breaking only slightly above its opening-range high. Another stock can have a less alarming RSI but be wildly extended above VWAP. Those are different risk profiles.

For this specific setup, the better guard is not "RSI must stay below a fixed number." The better guard is, "Price must not be too far above the opening-range high or VWAP." That is a more direct anti-chase rule. It asks whether the entry is still close enough to the structure that justified the trade. It lets RSI act more like a strength floor than a hard ceiling. We still do not want weak momentum. But if a true leader is hot, the question should become: is it controlled-hot or runaway-hot? That distinction matters.

The AMD contrast

AMD taught the other half of the lesson. AMD also ran. The tape gave signs of strength after the open: the gap held, the opening low held, and price later cleared the opening range with trend confirmation. But AMD did not have the same authorization.

Earlier, the open position had already been resolved profitably before the main session. After that, the plan treated fresh AMD as a smaller, more conditional candidate. It was pullback-only. No momentum backup. So when AMD ran, Onyx did not chase it. That was disciplined.

It is tempting to look backward and say the system should have bought it. But that is not how a plan-driven desk is supposed to work. The real question is not whether AMD went up. The question is whether the premarket process should have upgraded AMD before the open and explicitly allowed a smaller continuation backup. Maybe yes. Maybe no. But the authorization had to happen in the plan, not in the heat of the move.

That contrast is useful because it separates two misses that can look similar from the outside. RKLB had the right trade family authorized, but one guard was wrong. AMD did not have that trade family authorized, so the miss was a planning question, not a trigger question. Those are different fixes.

The danger of fixing the wrong thing

This is where trading systems can quietly drift. After a miss, the first instinct is often to make the system more aggressive. Loosen the rules. Add more symbols. Let more grades use the backup. Increase the window. Reduce the confirmations. Make the machine catch the next one. That would be the wrong lesson.

The journal did not say, "Buy all momentum." It said the opposite. The strongest lesson from the week is that the daily plan is working precisely because it forces us to name the allowed behavior before the market starts moving. The system avoided random bot trades. It kept the no-loss rule intact. It respected position management. It stayed inside the written plan. We do not want to lose that.

The fix should be narrower: keep opening-range momentum limited to A and strong-B names, keep it smaller than the primary setup, keep the market-regime requirement, keep the VWAP and opening-range structure, and replace the blunt RSI ceiling with extension limits. That is not making Onyx more impulsive. It is making the guardrail point at the actual risk.

The weekly lesson

By the end of the week, the pattern was clear. Onyx's new operating model is working. The system behaves less like a strategy bot and more like a desk process: research first, write the plan, execute only the plan, review the results, and let the next plan get smarter.

The no-loss rule held. That matters. It also means every entry has to be treated with respect, because bad entries can become holds. The system cannot rely on a stop-loss to clean up sloppy authorization. The pullback-first posture also worked. It protected the system from buying every hot candle in a risk-on market. But the same posture missed leaders that never came back to the preferred level.

The opening-range momentum backup was the right response to that problem. The RSI ceiling was the wrong first guardrail. That is a good kind of mistake. It is specific. It is visible. It is fixable. It did not come from the trader going rogue. It came from a constraint that made sense in theory and failed in a predictable way once the tape got hot enough.

That is exactly why the journal exists. Not every miss becomes a new rule. Not every strong move deserves a trade. Not every constraint is bad because it blocks something profitable in hindsight. But when a guardrail blocks the exact behavior it was designed to safely permit, the guardrail needs to change.

Friday's lesson is simple: do not remove the brakes. Aim them better.