Marking Up a Chart From Scratch

Here is a blank Apple chart, a date with a future, and every rule the levels module measured, applied one step at a time.

It is the last video of the levels module, and it introduces nothing new. And it ends on the one thing the measurements never asked — whether price gets to a level at all — which turns out to be the most useful thing in the module. By the end you'll be able to mark up a blank chart by rule and check it against what price did.

The setup first, because the honesty of the whole exercise depends on it.

The markup is done on a date that has a future. Apple, the sixteenth of March 2026. Everything before that date is allowed. Everything after is hidden until the end.

Price closed that day at two hundred and fifty two dollars eighty two. The usual daily range was four dollars and seventy three cents, which is one point eight seven per cent, measured over the previous sixty days.

Step one: find the candidates. Every turn and every wide-bar edge in the last two hundred and fifty days that price has not since broken — closed a full daily range beyond, which the ranking video counts as finished.

There are thirty five of them. Twenty five turns and ten wide-bar edges.

Marking all of them is the same as marking none, so the rest of this is subtraction.

Step two, and it is the cheapest cut available. Distance.

A level eight daily ranges away is not a decision about tomorrow. Price will take weeks to get there, and by then the level will have aged.

That removes fourteen of them and leaves twenty one. Still too many.

Step three: rank what is left, using what the turns, ranking and visible-levels videos measured.

Each level gets a score, and every number in that score is lifted directly from a measurement, not chosen here.

And only one number survived. A turn barely beat its shifted twin and a wide-bar edge barely beat its own, so neither gets points — the turns video could not tell them apart from chance.

Nothing for a turn made on a wide bar, nothing for a sixty day extreme, nothing for age — the ranking video and the visible-levels video measured nothing there.

Subtract six points if price has already tested the level since it formed. That is the one difference the ranking video could clearly tell apart from chance.

And when two levels score the same, the nearer one comes first. Two lines closer than half a daily range count as one area, because the turns video found that half a range makes no difference.

No weights were fitted here. Weights tuned on this chart would only show that weights can be tuned.

That leaves five lines on the chart. Here they are, with what each one is.

Two forty four sixty eight: a turn from late January that price has not come back to since it formed, under two daily ranges below price.

Two sixty two forty eight: a turn from the tenth of March, two ranges above. Then two seventy two eighty one, the edge of a wide bar, and two seventy six eleven, another turn — four and five ranges above.

And two twenty five ninety five, a turn made on a wide bar last September, almost six ranges below. All five are untested, so all five share the top score.

That is the markup. Five lines, each one justified by a measurement, no opinions anywhere in the process.

Now the part that was not planned. Here is what price actually did in the next hundred and twenty days.

Of those five lines, one was never reached at all — the farthest one. Price did not come within a quarter of a daily range of it in six months.

The four it did reach all produced a bounce.

So compare it with the naive markup — the five lines almost anybody would draw. The levels that have been tested again and again.

Four of those five were reached, and all four bounced. Two of them sat above the price, and Apple then climbed by more than a third.

On Apple, then, a draw: four of five reached on both, and every one of them bounced. But there is something in how the two were chosen that is worth noticing.

Every number in the levels module answered one question: given that price arrives at this level, what happens next?

Not one of them answered the other question: will price arrive at all?

And the two questions pull in opposite directions, mechanically.

The one criterion the ranking video could clearly measure was that an untested level outperforms a tested one — by about six points.

But a level is untested precisely because price has not been going there. Selecting for untested is selecting for unvisited: across six instruments, price reached seventy seven per cent of the untested candidates and eighty nine of the tested ones — close to, but not beyond, what this many lines can tell apart. Untested here means since the line was born — the price itself may have been visited before.

That is not a flaw in the measurement. The measurement was right. It is a flaw in using one measurement as if it were the whole job.

So the fix is a different order of operations, not different weights.

Filter first for what price can reach, then rank the survivors by quality.

Run that on all six instruments: keep only levels within three daily ranges, then score them the same way.

The share reached goes from ninety per cent to ninety six. Bounce rate among the ones reached rises from seventy eight to eighty five.

On this sample it is better on both counts: more events, and no worse ones.

Rare in this business — but read it as a direction, not a proof.

Now the limits, and this time the biggest one is about this video itself.

Five lines on six instruments is thirty lines and a handful of events. Nothing here proves that one markup beats another — on this sample the naive version reached as often and bounced a little less, and that proves nothing either.

The measurements were in the turns, ranking and visible-levels videos, on thousands of touches — and most of them came out as nothing. This one shows how what survived behaves on a blank chart, which is a different and smaller claim.

Second limit: one markup date. A different date gives a different chart and different lines, and I have not tested a hundred dates.

Third: the score now has one term, so nothing in it has to add up. But a score with one term ranks almost nothing — distance does most of the work.

So what does the finished procedure look like, after all of it?

Find every turn and wide-bar edge that price has not broken since. On Apple over a year that was thirty five.

Now divide them by distance. Within three daily ranges: these are live, and you rank them. Beyond that: mark them and forget them until price approaches.

Rank the live ones by the one thing that measured: untested first, nearest first, and lines within half a daily range of each other counted as one.

And keep the count small. Five lines you can defend beat twenty you cannot.

One more thing the markup is not, and it is the mistake that makes all of this useless a week later. It is not a one-off job.

Every new bar adds candidates. On Apple that is zero point four new turns and wide-bar edges per day — about two a week.

Tesla, about two a week. The index fund, two or three. On the coins, two to three.

And price moves, which changes which of them are live. On these six instruments, between seven and fifteen candidates currently sit within three daily ranges of price.

So the working habit is not marking up a chart. It is keeping a short list current: check it when price moves a daily range, and rebuild it when the week ends.

What to go and do, and it takes about half an hour.

Pick a chart and a date six months in the past. Mark it up without scrolling forward — that part is the whole exercise, and it is harder than it sounds.

Then reveal the six months and count two things separately: how many of your lines price actually reached, and what happened at the ones it did.

If your misses are lines price never visited, they are not wrong reactions — those lines were never tested at all.

Quality moves the odds at a level a little. Distance decides whether anything happens at all. Educational content only. Nothing here is financial advice.