Notebook 004 - Pullback-Only Is Safe, But Not Complete

The cleanest miss of the day was NVDA. That sounds like a complaint, but it is not. On May 7, 2026, Onyx did exactly what it was supposed to do. This was a paper-trading day, not financial advice, and the important point is not whether any single ticker was good or bad. The important point is that the system obeyed the written plan.

NVDA was approved only as a pullback trade. It did not give the post-open pullback Onyx was allowed to buy, so Onyx did nothing. No bug. No confusion. No unauthorized chase. Just a clean miss, which is exactly the kind of miss worth studying.

What the plan got right

The market backdrop was constructive. SPY and QQQ were both classified as up. AI and large-cap technology were leading the tape. Several names were strong before the open, but many were also extended. That created the central tension of the day: do we buy strength, or do we wait for a better price?

The May 7 plan chose discipline. Most entries were pullback/reclaim setups. The idea was simple: if a stock is already strong, wait for it to reset into a level, prove support, and reclaim before buying. That is a good default. It prevents the most common error in a hot tape: buying the candle everyone is already looking at. It keeps execution tied to a price, gives the trader a clear invalidation area, and makes review easier because the plan either got the reset or it did not.

Onyx also has a hard constraint that makes this even more important: no automated loss-taking sells. If a position is below average entry, the system does not intentionally sell it at a loss. That rule means bad entries can become overnight holds and stuck capital. So waiting for pullbacks is not cowardice. It is risk control. The system should be skeptical of extended strength. It should not buy every breakout just because the market is green. In that sense, pullback-only did its job. It kept Onyx from chasing.

What the plan missed

NVDA was the problem case. The homework was directionally right to like the name. The market was supportive. The theme was active. The stock was strong. But the plan authorized only one path: pullback. The stock touched the planned area on the opening bar, then moved away. Onyx does not enter on the first five-minute bar, and after that first window the stock did not return to the approved pullback level.

Later in the morning, a different setup appeared. NVDA cleared the first 15-minute opening-range high. It was above VWAP. Volume was elevated. Momentum indicators were aligned enough to confirm strength without looking like pure exhaustion. A modest target from that breakout area would likely have been reachable soon after. That was not the trade Onyx had permission to take, so it missed.

This is the useful part: the miss was not an execution failure. The trader did not skip an approved trigger. The plan was simply too narrow for the kind of day the market gave us. The question is not, "Why did the bot fail to trade NVDA?" The better question is, "Should the plan have authorized a second entry path?"

Safety is not the same as completeness

Pullback-only is safe in the same way a seatbelt is safe. You want it there. You do not remove it because it occasionally keeps you from leaning forward. But safety and completeness are not the same thing. A pullback-only plan handles one category of opportunity very well: strong names that reset into a defined level and reclaim it. It handles another category badly: strong names that never give the discount.

On a mixed or choppy day, that second category may not matter. Missing a breakout in a weak tape is often a blessing. But on a clean risk-on day, with leadership concentrated in a few obvious names, the best stocks sometimes do not come back to the level we wanted. If the system has no approved way to buy that strength, it will sit in cash. Sometimes that is correct. Sometimes it is too passive.

The answer is not to turn Onyx into a momentum bot. That would be the wrong lesson. The answer is to let the daily plan decide, before the open, whether a specific name deserves a tightly bounded momentum backup. Backup is the key word: not default mode, not live improvisation, not "looks strong, buy it." A backup.

What a momentum backup is not

A momentum backup is not permission to chase. It is not a replacement for the premarket plan, a new always-on standard strategy, or a rescue mechanism for a plan that was lazy or vague. It is also not available to every symbol on the watchlist. A real backup mode has to be harder to trigger than a human impulse. It should exist only where the daily homework already identified a high-quality candidate, and only when the broad market is supportive enough to justify buying strength.

That means most names should not get it. If a stock is event-risky, extended, low-quality, or only loosely interesting, it should stay pullback-only or no-trade. If the broad market is mixed, weak, or chopping around after a macro release, the backup should probably stay disabled. The backup should be special because the risk is special.

Buying strength means accepting a worse entry price than the pullback plan wanted. Under a no-loss-exit rule, that matters. A bad momentum entry can turn into a hold, and a hold can consume attention and capital the next morning. So the backup has to be smaller, narrower, and cleaner than the primary plan.

The shape of the rule

The May 7 review pointed toward a simple structure. Only selected A or strong B names can receive an opening-range momentum backup. Only when SPY and QQQ are up, or the market is clearly risk-on. Only during a specific morning window, roughly after the first 15-minute range has formed and before the trade becomes a late-morning chase.

The trade also needs evidence. Price should close above the opening-range high and stay above VWAP. Volume should confirm the move. Momentum should confirm strength without screaming exhaustion. Size should be smaller than the pullback plan, entries should still be limit buys, and no-loss exits should remain enforced.

The details can evolve. The important thing is the shape: a pre-approved exception with hard boundaries. That is different from chasing. Chasing happens when the market moves first and the trader invents a reason second. A backup mode does the opposite. It writes the reason before the market opens. Then the trader either sees the condition or it does not.

Why this belongs in the plan

The temptation after a miss is to patch the trader. Add a signal. Add a scanner. Add a rule. Make the machine more aggressive. That is not the right first move. The daily plan is the source of truth. If Onyx is going to buy opening-range momentum, the plan should say so explicitly. It should name the symbol, the size, the time window, the conditions, and the reason.

That keeps the operating model intact. Human plus tools build the plan. The trader executes the plan. The trader does not wake up and decide that a missed pullback means it should switch personalities. This is the difference between flexibility and drift. Flexibility means the plan can express different setups for different days. Drift means the live system starts expanding its own authority because the tape is exciting. Onyx needs the first and must avoid the second.

The other lesson: late-day small trades are still real risk

NVDA was the cleanest miss, but it was not the only lesson. May 7 also had small C-grade trades that were plan-approved. One resolved cleanly. One became the open position into the close, still above entry with a profit target working. That is acceptable under the no-loss rule. It is also a reminder: small does not mean harmless.

A late-day C-grade entry can become an overnight position if the target does not hit. Under a system that refuses automated loss sales, that overnight hold is not a footnote. It becomes tomorrow morning's first planning item. So there is a second refinement coming out of the day: C-grade entries after midday probably need stricter handling.

Maybe they require manual re-approval. Maybe size gets cut further. Maybe they need a stronger confirmation rule. Maybe some days they are disabled entirely. The exact policy can be decided in the next plan. The principle is clear: time of day changes risk. The closer a new entry is to the close, the less time the trade has to work before it can become a carry. That matters more when the system will not automatically sell at a loss.

The lesson

The May 7 lesson is not that pullbacks are bad. Pullbacks are still the right default. They keep us from buying heat. They enforce patience. They make entries auditable. They are especially important in a no-loss system where poor entries can linger.

The lesson is that a pullback-only plan is incomplete on certain risk-on days. When the broad market is supportive, leadership is clear, and a high-quality name refuses to reset, the daily plan should be able to authorize a second path: opening-range momentum, smaller size, strict window, confirmed strength, limit order, no-loss exits intact.

If that backup is not written into the plan, missing the move is acceptable. That sentence matters. Missing an unauthorized trade is not a failure. It is discipline. But once the journal shows the same kind of miss clearly enough, the next plan should get smarter. That is the whole point of the Onyx notebook.

Not every miss becomes a new rule. Not every strong candle deserves a trade. Not every lesson should become code. But a good miss should leave a mark. NVDA did. Pullback-only kept us safe. Now the plan needs a way to be safe and a little more complete.