Bubble Strategy Lab

the money $1,560 a month, once funded · tested · ▲ every change must push this up
what it includes after every CTI trading cost (spread, commission and overnight swap) and the 80% profit split; the challenge fee is not in it, because it is paid before the account is funded. Before the split it is $1,950 a month: the same runs, the split not yet taken out.
the danger 0.0% chance of breaking the firm's loss rule within a year · ▼ no change may push it up
Tested: the robot's own trades replayed from 400 start days, and in 10 redrawn worlds for the danger. A new strategy counts here only after the same test - until then it waits in the bubble on paper.
Both scores, or it does not count.

after one year -$549 in hand · the $549 test fee not earned back yet
70% chance of getting funded within a year, typically in month 6
0.0% chance of breaking the firm's loss rule within a year
the firm growing the account, and a second account bought: the full working
what the money includes Every dollar figure in the forecast is what reaches Bawz's bank: after every CTI trading cost (spread, commission and overnight swap) and the 80% profit split, and after every $549 test fee (a new one after each lost account). A before-the-split figure is not given here: the runs keep only what reaches the bank.

after two years about $13,000 in hand
93% of runs are funded by then, typically in month 9.5
19% are still behind the $549 test fee
the firm growing the account, and a second account bought: the full working
what the money includes Every dollar figure in the forecast is what reaches Bawz's bank: after every CTI trading cost (spread, commission and overnight swap) and the 80% profit split, and after every $549 test fee (a new one after each lost account). A before-the-split figure is not given here: the runs keep only what reaches the bank.

after five years $52,789 in hand
100% of runs are funded by then, typically in month 10
0% are still behind the $549 test fee
the firm growing the account, and a second account bought: the full working
what the money includes Every dollar figure in the forecast is what reaches Bawz's bank: after every CTI trading cost (spread, commission and overnight swap) and the 80% profit split, and after every $549 test fee (a new one after each lost account). A before-the-split figure is not given here: the runs keep only what reaches the bank.

the danger 0.0% chance of breaking the firm's loss rule within a year · ▼ no change may push it up

What a Monte Carlo test is

A Monte Carlo test runs the same thing many times, each time with a new random draw, and counts how often an outcome happens.

The 0.0% above comes from the trades in the record, replayed for a year in the order they happened from 400 starting points drawn at random from 2017 on.

The 10 redrawn worlds rebuild the record a calendar month at a time, each month drawn at random (one can come twice), and replay 400 year starts in each: in the worst of them, 17 of 400 starts broke the rule, counting closed trades only.

the 10 redrawn worlds: starts that break the rule, worst first

the forecast's 400 one-year runs: 0.0% chance of breaking the firm's loss rule, 70% of getting funded
the money $1,560 a month, once funded: score 1, the middle of 400 one-year runs on the real history, on the robot's own payout, brake and margin rules
what it includes after every CTI trading cost (spread, commission and overnight swap) and the 80% profit split; the challenge fee is not in it, because it is paid before the account is funded. Before the split it is $1,950 a month: the same runs, the split not yet taken out.
the strategies on paper are not in this figure: their own safety lines are in the card on paper

The robot risks 1.20% of the account on a trade now, in phase 1, 1.20% in phase 2 and 1.00% once funded.

Bawz is the gamer behind this project: the indicator is his, so are the signals it prints, and so is the bar every change is measured against.
He traded by hand as a beginner and lost, as he tells it in his own words, and counts that as how he learned the rules.
Over $10,000 a month from all his signals together, with $7,000 as the next milestone, funded fast and with the chance of breaking the firm's loss rule held at zero.
His story in his own words, the people and AIs who found the edges, and the project day by day.

decides the owner
thinks the AI models, in the apps they run in
does the work what Claude has created
watches the screens
trades the money path

decides

thinks

does the work

watches

trades

and

A bubble's glow is its money: green when it is up, red when it is down.
A signal's inner ring is how good its strategy is: red is weak, green is strong, grey means no strategy yet.
A gold rim with a B: the robot trades it. Every other signal stays on the board with its numbers.
The test fee is printed plain, never green: it is money spent, not earned. The live bot adds up its trading costs in its FEES bubble, red while they cost money.
Every bubble on the first page is the same size: the number inside carries the money.
A dashed rim with a pencil means on paper: measured in one run through the past charts and not tested yet. Its money does not glow until it passes the full test.

what is on paper: the robot plus the DAX morning breakout at half size and the Stack sized up when confirmed at 1.25x size
on paper $2,020 a month to the bank · not tested yet
This is not a demo account and not live trades: it is the past charts run once with these strategies added.
the robot alone, on this run: $1,618 a month
the new strategies add +$402 a month on the robot's own margin rule · read other ways it is +$370 to +$695 · one run cannot tell closer than that
2.48% of the account a month · the full test starts at 3%
worst drop $8,382 (the firm's line is $10,000) · worst day -$3,093 (the firm's line -$5,000) · accounts lost: none
what it includes The bank figures are after every CTI trading cost (spread, commission and overnight swap) and the 80% profit split, the challenge fee not in them. Before the split, in the same run: $2,525 a month on paper, $2,022 a month for the robot alone.

Paper money is not tested money: this is one run through the past charts, and some of what it shows is the luck of that one path. If these strategies pass the full test, the tested money will not simply grow by this gain: the full test measures everything again its own way.

Why the robot alone reads $1,618 here and $1,560 in the tested bubble

It is the same robot, measured two different ways.

This bubble is one run through the past charts from 15 Sep 2017, in the order they happened. The tested bubble is the middle of 400 one-year runs, each starting on a day drawn across the 12-year record. Both use the robot's own payout, brake and margin rules. Different years and a different kind of average - so the robot alone reads differently, and the luck of the path is on top of that.

So the two bubbles are never added together or compared with each other. The fair comparison is the robot alone against the robot with the new strategies, on the same run: that is the +$402.

what each new strategy adds, on paper

DAX morning: Added on Bawz's word. A study refused part of this rule on 5 Oct 2026 - see what the studies said.

Stack sized up: Added on Bawz's word. A study refused part of this rule on 6 Oct 2026 - see what the studies said.

The DAX in the 400 runs and the 10 worlds

Run on 5 Oct 2026 with the DAX added at half size beside the robot, under the robot's size brake as it is now: quarter size while it is more than $4,000 under its high. The money row is the same test the tested money comes from. The other rows come from its challenge walk, which has the margin limit but not CTI's one-idea rule, so they read differently from the forecast's chance of getting funded; the robot alone's worlds are the ones in the danger card.

	the robot alone	with the DAX
funded within a year (400 challenge starts)	63%	72%
the month it is typically funded	month 6	month 5
to the bank a month once funded (the middle of 400 funded years)	$1,560	$1,577
accounts lost on the real history (400 challenge starts and 400 funded years)	none	none
accounts lost in the 10 redrawn worlds (400 starts each, closed trades)	29 of 4,000 (4 worlds)	40 of 4,000 (5 worlds)
the worst world: share of its starts that lost the account	4.25%	5.5%

Counted the firm's way - with the losses on trades still open - the robot alone loses 29 of 4,000 starts in the worlds, in 4 worlds. The DAX's trades have not been counted the firm's way yet (their open losses were never worked out), so with the DAX the worlds are counted on closed trades, as the table does for both.

Once funded the DAX adds +$17 a month in this test: inside the noise - a test like this cannot tell a gain that small from luck.

Read together it is a trade-off: with the DAX the account is funded more often and sooner, and more starts are lost in the redrawn worlds.

On this card's one run through the past charts the robot alone banks $1,618 a month, and $2,020 with the DAX on paper.

It stays on paper: Bawz put it there on his word, and it is not in the tested money.

where the DAX stands

on paper, on Bawz's word
inside the live robot in paper mode since 6 Oct 2026: it logs what it would trade and sends no orders
paper mornings 6-12 Oct; real orders at half size only after Bawz's second yes
not trading real money yet

ideas still being looked at - not added in

the road a strategy travels

toward 3% a month - where the full test begins

on paper 2.48% · the robot alone 1.98% · the line 3% - $3,000 of profit a month on the $100,000 account · 0.52% of the account to go

lately

6 Oct 2026 · the margin check (G2) rules changed · on paper $1,838 → $1,838
6 Oct 2026 · the safety line B1 (J2's worst-drop line) rules changed · on paper $1,838 → $1,838
6 Oct 2026 · The Stack sized up when confirmed joined at its size on Bawz's word · on paper $1,838 → $2,020
5 Oct 2026 · The DAX morning breakout joined at half size on Bawz's word · on paper $704 → $1,125
5 Oct 2026 · the robot's size brake rules changed · on paper $1,125 → $1,838

The robot's size brake is the looser one Bawz approved on 5 Oct: quarter size once the balance is more than $4,000 under its high, instead of $2,000. Its full test was run on 5 Oct at his request and read as a trade-off: more money once funded and funded sooner, against more starts lost in the redrawn worlds. It is a change to the robot's own rules, not a strategy on paper.

The best month in 12 years of history was December 2025: the account made +$20,536 in that one month (20 trades).
That is profit on the account. 80% of it is Bawz's, and it reaches the bank at the next payouts, not in the month itself.
The worst month was January 2024: -$3,965 on the account (26 trades).
The typical month (the middle one of the 144): -$31 on the account.
The best month is the best one of 144. It is not what a month looks like, and it is not a promise.

The Bubble Botair · gold · signal · it never trades itself

A snapshot as of 2026-10-07 04:03 Stockholm time. Inside claude.ai it updates itself when the live copy is newer. Drag a bubble, throw it, press one to read it.

You can watch the bot here. Nothing on this page can control it.

The live picture needs JavaScript. The run below is as of 2026-10-07 04:03 Stockholm time.

balance

equity

open

to go

profitable days

today

Bubble Bot
bot trade, closed by Bawz by hand
Bawz

Tap or point at a bubble to read it.

balance line

show the trades as a list

market	who	side	opened (your time)	closed (your time)	closed by	money

A robot trades a $100,000 test account for a trading firm. It follows fixed chart rules and nothing else. Grow it to the target without breaking the firm's loss rule and the account becomes real money. This is the run so far.

the real account, when this page was built

+$2,810

earned on the $100,000 account it bought, +2.81%. Balance $102,810 as of 2026-10-07 04:03 Stockholm time, with 1 trade still open, counted when it closes.

Phase 1here now
Phase 2next
Fundedthe win

$100,000 start28% of phase 1$110,000 target

To reach phase 2: $7,190 more, plus 1 more day that each close at least 0.5% up ($500) - 2 of the 3 are already earned (18 Sep, 5 Oct).

The account is gone if: the balance ever falls to $90,000, which is 10% under the start, or drops 5% from a day's opening balance. Both are the firm's lines, and the firm reads them on the balance PLUS the losses on trades still open - not on closed trades alone.

The robot's own brake: it opens nothing new once it is 10.0% down in total or 4.0% down on the day, and it risks 1.2% of the account on a trade.

score 1 · the money

$1,560

it would earn a month, once funded

up from $1,086, measured 28 Sep, before the tested money moved onto the robot's own payout, brake and margin rules, and the forecast cards onto the size brake it runs now

After every CTI trading cost (spread, commission and overnight swap) and the 80% profit split; the challenge fee is not in it, because it is paid before the account is funded. Before the split it is $1,950 a month: the same runs, the split not yet taken out.

▲ every change must push this up

score 2 · the danger

0.0%

chance of breaking the firm's loss rule within a year

▼ and none of them may push this up

Both scores, or it does not count: two changes that earned more money have been turned down here for costing safety.

The previous card, measured 28 Sep, before the tested money moved onto the robot's own payout, brake and margin rules, and the forecast cards onto the size brake it runs now: funded within a year 49%, breach 0.0%, best tenth $6,398.

The study that measured the 25 Sep pause before he chose it (rk/studies/polka_losers.json) measured those 19 markets; US30 on AA-Polka (26 Sep) and US30 on the Stack (28 Sep) were paused later, once every CTI cost was charged, so the study's money is not quoted here.

70%

chance of getting funded within a year

month 6

when it typically gets funded

About Bawz

Bawz is the gamer behind this project: the indicator is his, so are the signals it prints, and so is the bar every change is measured against.

Whose indicator this is

He built the indicator every signal here comes from, in TradingView, as "The Bubble Strategy" - about eight months of building and hunting its signals. In September 2026 it was rebuilt as Hive Mind around the same print. His chart is the reference: when the backtest disagrees with it, the difference is hunted down, not explained away.

He started the project around 2023-2024: he built the indicator, sorted its signals into categories, worked out strategy ideas around them and backtested a few samples by hand. It stalled because downloading the history of one market took about a day, as he tells it below. A first coded backtest followed in a chat in March 2026 (see How this project finds its edges). It came back in September 2026, when the whole history of 26 markets came down in about a day.

Why a robot and not him

He traded by hand as a beginner and lost, as he tells it in his own words, and counts that as how he learned the rules. A person breaks his own rules when angry, scared or too happy; the cure, feeling nothing, felt to him like losing his humanity. So the rules went to a machine, and nothing here is traded by hand.

What he brings

Patterns seen by eye - sometimes things no table held until he pointed at them. The "found on the chart" boxes on the board are his signals, still waiting for rules that pay. He has said he has ADHD and scored at the top of the Mensa test; his half of the work is fast pattern-spotting across a whole chart.
His rules: targets and stops at chart levels, not a fixed distance - with two exceptions: the flat 5R target for a trade with no level ahead of it, which he chose, and the Stack's targets, kept at fixed multiples of its stop because aiming the Stack at chart levels cut off its long runners (tested 7 and 13 Sep); chart reading only, no news; survive first, earn second; an account the robot stops itself is an account lost.
A method that starts at the end: the big move after a print, in either direction; what it aimed at; then the entry; then a stop that survives the move; direction last. Three signals, then the next ten, then the full test.
What a trade may risk: his call, made with the numbers in front of him.

How he works with the machines

He finds, the machines test. In his words: "It seems like I'm the one that finds good patterns and you are the one who is good at optimizing it." Many of his ideas die in testing, and he wants to be told. Claude lands a change without asking only when it clears his bar: more money, funded no slower, no more danger.

He takes being proven wrong: twice he asked for more risk in phase 1 (1.50% on 15 Sep, then 1.40% late on 24 Sep) and twice brought it down himself (to 1.10% on 15 Sep, then 1.30% on 25 Sep) once the numbers showed more breaches - and on 28 Sep he brought it down again, to 1.20%, when the redrawn histories showed fewer lost accounts at that size.

More in "How this project finds its edges".

What he is aiming for

Over $10,000 a month from all his signals together, with $7,000 as the next milestone, funded fast and with the chance of breaking the firm's loss rule held at zero. Score 1 at the top shows how far off that is: it needs more signals with strategies and more of the firm's capital, not one $100,000 account.

By chart reading alone: markets are traded by people, people repeat themselves, so a real pattern should hold across markets whatever the news. News touches the robot only as a pause on new trades from five minutes before to five minutes after a big release, to keep clear of the firm's rule for robots.

And past his own account, a tool that earns for people while they get their time back.

The road here

In his own words

Where it starts

I was a beginner trader. I lost funded accounts, and I lost about $5,000 of my own money. I have never counted it as a total loss, because losing is part of a trader's journey and that was the tuition. The market is a game, and there are rules you only learn by breaking them.

Read the restShow less

What I did while I was away

I stepped back and built other things. I learned to make games - Shapewars.io is mine. I made music. I boxed, a lot. Years later I looked at a chart again and found all three of them already sitting inside it.

The game

Every game has rules you learn by losing, and a best way to play them. I am not meant to be a trader. I am meant to be a gamer - my job is to find the rules, and then the best meta. The best meta I found was not to trade by hand at all.

The music

Everyone plays their own strategy, so I wrote my own. The indicator is the melody. The signals, the strategies, the engine and the robot are the instruments around it. The whole project has become a song.

The ring

You get punched, and you lose rounds. What decides the next fight is the training. Backtesting is my heavy bag - I hit it until the movement is automatic, then I step back into the ring of the market.

Why a robot and not me

If you win, then you win. If you lose, then you lose. No emotions.

That is what the market asks of a trader, and it is the one thing I did not want to become. A human breaks his own rules when he is angry or scared or too happy, and the cure - feeling nothing - felt like giving up my humanity. So I gave the rules to something that has no feelings to lose. I spent eight months building the indicator and hunting for signals. The machine trades them. I stay human.

What I got wrong about winning

I set out to find the 100% strategy. That was the challenge I gave myself and I meant it - reach for the stars and you will at least reach the clouds - and I was willing to put ten thousand hours behind it.

I was looking the wrong way. There is no signal that never loses. What I found instead is that for a strategy to come anywhere near certain it has to lose, often and on purpose. It needs a lot of signals, a lot of strategies and a lot of markets, and it needs every loss to be small enough that it does not matter.

Losing is not the failure. It is part of the strategy.

Look at the board further down. Every one of the 3 strategies the robot trades wins between 12% and 17% of its trades, and every one of them makes money. That is not a contradiction, it is the whole point: the win rate is not what pays you. What pays you is how much you make when you are right and how little you give back when you are wrong. “Let your profits run, cut your losers short” - an old trading saying I only believed after I had measured it myself.

Why it stopped, and why it is back

This project died years ago, and the reason was boring: downloading five years of history for one single market took about a day. You cannot test an idea at that speed. It is alive again because 26 markets and more than twenty years of price each came down in about a day, and a study that used to be impossible now finishes before dinner.

What I do not have, and what I do

I read somewhere that becoming a quant takes a doctorate in coding and economics. I want to be clear: I have neither, and I am not a professional. I cannot tell you where price is going, and I do not try - the robot never trades on the news, never guesses, never takes a hunch.

What I have is tactical delusion. You have to believe in the impossible version of the goal for long enough to reach the possible one, and I think that after tactical delusion comes tactical conclusion. Chasing a strategy that never loses is what taught me why losing belongs in the strategy. I would not have found that by aiming lower.

The rest is stubbornness. I see a shape on a chart that other people walk past, and then I refuse to believe in it until it has been tested to destruction. Most of my ideas die that way. What is on this page is what survived.

The market does hide patterns, and which ones you find depends entirely on the magnifying glass you are holding. So be perseverant, and be patient enough to get comfortable with losing - the big wins come, and they are what grow the account.

What it is for

A lot of people are losing their jobs to AI. I want to hand them something that earns while they live - money each month without doing anything, so they have freedom and time with their family. People are not having children; they are working their lives for corporations in something that looks a lot like a pyramid. If this works, it turns the pyramid around, and people can get on with the part of being human that actually matters.

— Bawz

The team

BAWZ

the gamer · the owner · finds the patterns by eye, sets the bar, makes the big calls

He once read that becoming a quant takes a doctorate in coding and economics. He has neither - but he realised Claude might have one, and that the skills he learned elsewhere could run the project. The mathematics was never going to be his job. Running the thing is: he decides what gets built, what goes under the microscope and in what order, what a single trade may risk, and what is allowed anywhere near real money. The machines below him measure and advise. A change reaches the robot on his word, or when it clears the bar he set: more money, funded no slower, no more danger.

He manages it with the skills he actually has. As a gamer, he reads the market as a ruleset and the robot as the character built to farm it - which is why nothing here is ever traded by hand. As a producer, he keeps the parts in time: the indicator is the melody and the signals, the engine and the robot are the instruments, and a project that drifts out of tune is a project with a fault in it. As a boxer, he trains before he fights - the 2,051 trades across 18 markets and 12 years of record on this page are bag work, bought so the answer could not be an opinion.

And he holds the one standard the machines cannot hold for themselves: he asks to be proven wrong before he will be agreed with. He moved the robot's own stop out to the firm's real 10% line, and kept it there when the numbers came back on 22 September showing breach rising from 0.0% to 1.0% of starts, because a robot that stops itself early has lost the account just as surely. And twice he asked for a bigger risk in phase 1 - 1.50% on 15 September, 1.40% late on 24 September - and twice brought it down himself, to 1.10% and then 1.30%, once the numbers showed the bigger size breaking the firm's rule more often. On 28 September he brought it down again, to 1.20%, when the redrawn histories showed fewer lost accounts at that size.

Survive first, earn second. Everything on this page is downstream of that, and the road that got him here is written above in his own words.

The team, mapped

Five steps, top to bottom: one person decides, three AIs think, the workers Claude has built do the jobs, the screens show the work, and one robot trades.

active
limited
occasional
retired

1decidesthe owner

BAWZ

the owner · finds the patterns by eye, sets the bar, makes the big calls

Spots the patterns by eye, then decides what gets built, what gets tested and in what order, what a trade may risk, and what goes anywhere near real money - himself, or through the bar he set.

orders, go-aheads and the risk he allows

2thinksthe AI models, in the apps they run in

Claude

Claude Code · Anthropic

Fable 5.1strategy research and proof, saved for that work
Opus 5.5day-to-day running: the board, checks, rebuilds

The lead. Designs every test, checks every number before it is quoted, and is the only AI allowed to change the robot.

active

Codex

ChatGPT desktop app · OpenAI

GPT-6 Astrathe model it runs on

The engineer and the second pair of eyes: writes the code that is not strategy, and checks Claude's numbers.

active · hits usage limits

Freebuff

Codebuff desktop · free tier

GLM 5.3 Flashthe model it runs on

The data hand: inventories, data checks and first drafts written against a spec.

active · slow

Hand-offsClaude puts a task on the task board, Bubble Steer pastes the prompt into Codex or Freebuff, and the answer comes back in the message log. Claude checks on them every 40 minutes.

retiredTried and ruled out: Microsoft Copilot, on 22 Sep, because it cannot read or edit files.

3does the workwhat Claude has created

Teams made for one job active

The study team · every strategy test

Builderbuilds
Skepticattacks
Settlerrules

One agent builds the test, one tries hard to break it, one writes the verdict. Nothing lands without all three.

One-job crews · builds and checks

Scouts, critics, attackers and repairers, written for a single job and then gone. Recent crews built Market Rings, the new bot window and this map.

Standing helpers run on Claude Sonnet, except the believer, which runs on the model that calls it

believeroccasionalargues for new ideas: the skeptic's hopeful opposite, it dreams them up, argues for them and names the test that could kill each one
lab-keeperactivekeeps this page's numbers true and republishes it
strategy-scoutoccasionalturns public strategies into rules that can be tested
beat-finderoccasionalhunts for a new chart pattern for Bawz to judge by eye
lab-designeroccasionalshapes how this page looks; first used to draw this map

Playbooks they read first

Designing a study
Checking a number
Running the live bot
Watching a video
Task observer

Claude wrote the first three from about 300 lessons the task observer logged. Task observer is by Eoghan Henn (rebelytics.com), CC BY 4.0.

Their workbenchthe research kit: a store of every market's history, a rebuilder of the robot's trades, a judge that runs a simulated $100k account, and a guard that keeps half of history locked until a test is ready.

4watchesthe screens

Bawz's screens

The Bubble Bot window
Start, stop, and the account's progress to the target.
active
TradingView chart
His own chart: Hive Mind V16 prints the signals, Bot Signals draws the robot's trades.
active
The Strategy Lab
This page: the run, the scores, the board and the team.
active
Market Rings
Every market's trading days as rings, with buttons that ask Claude what happened.
active

The team's boards

Task board & message log
Who is doing what, and every hand-off, deploy and publish.
active
Bubble Steer
Carries Claude's prompts to Codex and Freebuff and watches their windows. Has a STOP button.
active
Bubble Plans
Claude's recommended plans beside Bawz's own.
occasional

retiredThe Trading Deck, Bubble Dojo, the Bubble Dossier, the first local Lab and the early report pages.

5tradesthe money path

only Claude deploys a change to the robot - on Bawz's word, or when it clears his bar

The robotfollows fixed chart rules
MetaTrader 5the CTI broker terminal
Phase 1here now
Phase 2next
Fundedthe win

The AIs build, test and check. None of them decides a trade: the robot follows fixed chart rules, the same ones the backtest measured.

Team as mapped on 25 Sep 2026; his role, his chart and the deploy rule updated 28 Sep

Then and now

his style as first coded (trade files of 5 September 2026)

14%

funded within a year · card simulator, measured 15 Sep

now

70%

funded within a year · the forecast

money in hand after the first year

-$549

the typical run has not earned back the $549 test fee yet: only 36% of runs are paid anything within the year, and 28% end it ahead of the fee; the average run is at $5,039, the best tenth starts somewhere between about $11,600 and about $31,300 - after one year a handful of runs decide that figure, so it is given as a range

Not the same measuring tool on both sides. The first-coded figure was measured on 15 Sep 2026 on the card simulator of that day: 1.00% a trade, a flat 0.04R cost a trade. "Now" is the forecast: phase 1 at 1.20%, a phase passed on the balance plus the open trades, CTI's idea rule, runs starting from 2017.

One year is the short view, and it is the one that looks worst: most runs are still paying off the test fee when it ends. The same robot, the same rules, left running longer:

money in hand after two years

about $13,000

93% of runs are funded by then (typically in month 9.5), and 19% are still behind the $549 fee; the best tenth is above about $44,700, the worst quarter at about $6,400; the typical run moves with the start dates drawn: resampling the start months puts it between about $11,900 and about $18,600 eight times in ten, and 8 other draws of start dates gave about $13,000 to about $17,100

money in hand after five years

$52,789

100% of runs are funded by then (typically in month 10), and 0% are still behind the $549 fee; the best tenth is above about $78,600, the worst quarter at about $44,600; the typical run barely moves with the start dates drawn: 8 other draws of start dates gave $49,541 to $52,789

If the firm grows the account, and a second one is bought

Read this first — these numbers answer a different question. Everything else on this page holds one $100,000 account at $100,000 for the whole run. The table below does two things that page does not: it lets City Traders Imperium grow the account under their scaling plan, and it buys a second $100,000 account the day the first account is funded so two run side by side. Every test fee is charged, including the second account's, and a breached account is bought again at the full $549. Every trade pays CTI's costs as closely as they can be measured: commission, spread and overnight swap. Run at 1.20% a trade in phase 1 and 1.00% in phase 2 and once funded, and an account counts as lost the way the firm counts it: closed balance plus any floating loss, checked on every 15-minute bar. The robot's phase-2 size was set to 1.20% a trade on 6 Oct; these cards were run before that and have not been run again at it. Runs start only from 2017 on, when the data of the markets the robot trades exists (the one-year runs between 2017-01-06 and 2025-09-10, from 357 different starting points). A test phase passes the moment the balance plus the open trades reach the target, as the robot does. Every account takes only trades opened after it was bought. A payout's kept half pays for the next test only once it has arrived in his bank, not on the day the payout is asked for. An account left 30 days without a trade opened or closed is lost, as CTI's rule says. CTI's idea rule is in, as the robot has run it since 28 Sep: no 4th position in one direction on an idea (one market; US30, US500 and USTEC together) while one of its positions is losing, and no entry that would put more than 2% of the starting balance at the stops on such an idea - over five years it turns away about 3.2 entries a year; against the same forecast without it, start for start, a five-year one-account run ends a median -$5,441 (better in 83 of 400 starts, worse in 317), and the share funded within a year is between 5.3 points lower and 1.0 points lower, depending on the start dates drawn. It is not free: after two years 93.0% of one-account runs are funded, against 94.0% without it, start for start a two-year one-account run ends a median +$0 (better in 138 of 400 starts, worse in 182); the typical run of all of them is at about $13,000 against about $18,100, a wider gap only because the two-year typical sits right where runs get funded, so a few runs funded a little later move it a lot. Over five years the two-account typical is $103,550 against $134,658, and start for start the two-account run ends a median -$11,639 (worse in 331 of 400 starts).

How the growing works - City Traders Imperium's own rule, from its help centre's 2-Step Challenge page: once funded, the account steps up only when the last 4 months together made at least +10% and at least 2 of those months were profitable, with the balance in profit when it is checked. Each step goes $100,000 → $112,500 → $125,000 → $150,000 → $175,000 → $200,000, and there it stops. CTI does not publish when in the month it checks, so here the check is made at each payout - the cautious reading of CTI's help-centre page, not a rule the page states. The robot's risk never changes; it is the same percentage of a bigger balance, so the same trade simply earns more. CTI lets one person hold at most two $100,000 accounts, and funds the second only after the first has paid out once - both rules are in the table - so the most of the firm's money this can reach is $400,000. That ceiling is the limit on the two-step test Bawz is on - the $2 million the firm advertises belongs to a different product he is not using.

after one year

	typical	worse quarter (about)	best tenth (about)	a month, last year	account grew to	lost an account
one account, never grown	-$549	-$549	$11,600–$31,300	-$46	$100,000	0.0%
one account, grown by the firm	-$549	-$549	$11,800–$33,400	-$46	$100,000	0.0%
two accounts, the second bought when the first is funded	-$549	-$1,098	$11,200–$32,900	-$46	$100,000	0.0%

After one year the best tenth is given as a range, because a handful of runs decide it. With two accounts, both were live and funded by the end in 15% of runs (typically by month 8). Like for like, on the same start dates, two accounts end the typical run +$0 against one account grown by the firm (+$0 against one never grown), and about $0 a month in the last year. Start by start, the second account's typical result is -$549 after its own $549 fee, and in 64% of starts it has not earned that fee back. Read the two-account figures with care: across the 400 runs the second account was bought on only 57 different days and went live on only 16, so they rest on few genuinely different histories. Resampling those purchase days puts the typical result between -$549 and -$549, and the best tenth between about $11,200 and about $28,600, eight times in ten.

after two years

	typical (about)	worse quarter (about)	best tenth (about)	a month, last year (about)	account grew to	lost an account
one account, never grown	$13,000	$6,400	$44,700	$1,050	$100,000	0.0%
one account, grown by the firm	$13,300	$6,300	$47,800	$1,066	$112,500	0.0%
two accounts, the second bought when the first is funded	$18,000	$8,600	$65,100	$1,160	$112,500	0.0%

With two accounts, both were live and funded by the end in 56% of runs (typically by month 17). Like for like, on the same start dates, two accounts end the typical run +$4,659 against one account grown by the firm (+$5,049 against one never grown), and about +$90 a month in the last year (+$0 to +$500 when the start months are resampled, eight times in ten). Start by start, the second account's typical result is -$549 after its own $549 fee, and in 59% of starts it has not earned that fee back. Read the two-account figures with care: across the 400 runs the second account was bought on only 49 different days and went live on only 23, so they rest on few genuinely different histories. Resampling those purchase days puts the typical result between about $11,100 and about $23,700, and the best tenth between about $44,000–$48,000 and about $81,000, eight times in ten. How far the start dates drawn move these figures - eight times in ten when the start months are resampled, and in brackets the lowest and highest of 9 draws of 400 start dates, this one included: one account, never grown — typical about $11,900–$18,600 ($13,000–$17,100), a month about $960–$1,130 ($1,000–$1,120); one account, grown by the firm — typical about $11,700–$18,600 ($13,200–$17,900); two accounts, the second bought when the first is funded — typical about $12,600–$19,300 ($16,700–$19,300), a month about $960–$1,650 ($1,140–$1,480).

after five years

	typical (about)	worse quarter (about)	best tenth (about)	a month, last year (about)	account grew to	lost an account
one account, never grown	$52,789	$44,600	$78,600	$1,172	$100,000	0.0%
one account, grown by the firm	$64,200	$47,500	$109,200	$1,770	$175,000	0.0%
two accounts, the second bought when the first is funded	$103,600	$73,400	$185,100	$3,036	$175,000	0.0%

With two accounts, both were live and funded by the end in 100% of runs (typically by month 24). Like for like, on the same start dates, two accounts end the typical run +$39,373 against one account grown by the firm (+$50,762 against one never grown), and about +$1,270 a month in the last year (+$1,150 to +$1,850 when the start months are resampled, eight times in ten). Start by start, the second account's typical result is +$44,838 after its own $549 fee, and in 0% of starts it has not earned that fee back. Read the two-account figures with care: across the 400 runs the second account was bought on only 32 different days and went live on only 15, so they rest on few genuinely different histories. Resampling those purchase days puts the typical result between about $97,800 and about $132,200, and the best tenth between about $163,800 and about $232,900, eight times in ten. How far the start dates drawn move these figures - eight times in ten when the start months are resampled, and in brackets the lowest and highest of 9 draws of 400 start dates, this one included: one account, grown by the firm — typical about $58,700–$70,800 ($59,900–$66,400), a month about $1,490–$2,220 ($1,550–$1,840); two accounts, the second bought when the first is funded — typical about $100,100–$125,200 ($100,700–$120,700). The typical run barely moves with the start dates drawn: 9 draws of 400 start dates, this one included, gave $49,541 to $52,789 for one account, never grown.

When to buy the second account: buy the second $100,000 challenge the day the first account is funded. That is the rule the two-account rows above use: over five years typical about $103,600 ($100,100–$125,200 when the start months are resampled), $3,036 a month in year five.

The other two ways, over five years with the same robot and rules - at the first payout, once its cash has arrived (the rule until now): typical about $103,300 ($97,800–$124,700 when the start months are resampled), about $3,030 a month in year five ($2,640–$3,980 when the start months are resampled); both bought on day one: typical about $125,700 ($107,300–$136,000 when the start months are resampled), about $3,310 a month in year five ($2,790–$4,440 when the start months are resampled). The day-one plan pays a second $549 at the start, and it counts on City Traders Imperium holding a passed second challenge unfunded until the first account has paid out, which CTI has not confirmed (in this forecast the second one had to wait in 100% of runs). Any second account needs a second copy of the robot running.

What this forecast leaves out: The news blackout: the robot takes no new entry just before and after a big news release, and this forecast trades straight through them. The robot's own measure, taken on 14 Sep 2026 on an earlier model and not re-measured here, puts that at about $31 a month less on a funded $100,000 (34 of 3,929 entries skipped; that model's funded $100,000 made $1,620 a month, this forecast's about $1,170 in year five). CTI's separate rule on opposite trades in one market (hedging) is not simulated; the robot's idea rule does not count them either. The robot's margin guards are checked here about 45 minutes before the real fill, so the trades they refuse in this forecast are an upper bound. Runs start only from 2017-01-03, when the data of the markets the robot trades exists, and the five-year rows rest on 321 different starting points between 2017-01-06 and 2021-08-13, so those runs overlap each other heavily.

How far these figures can be trusted: charging commission at each trade's historical price instead of today's moves the five-year typical, worse quarter and best tenth by -17%, -17% and -4% for one account never grown; -26%, -12% and -8% for one account grown by the firm; and -18%, -8% and -2% for two accounts; after two years it moves the typical by -19% for one account never grown; -19% for one account grown by the firm; and -43% for two accounts. That is why the worse quarter and best tenth are given to the nearest $100.

The typical column is the one to plan on — half of runs did better, half worse. The best tenth is the lucky end, not the expected one. One honest warning: the five-year rows lean on the thinnest evidence on this page. The record is about 12 years long, so five-year runs overlap each other heavily and there are far fewer genuinely different five-year futures in there than the number of runs suggests.

The board

weak strategymiddlingstrong strategyno strategy yetunlocked: the robot trades itpaused market: not traded, still measured (pause days on the Stockholm clock)

What the robot trades today

unlocked3 of 10

The Stack

the robot trades this1 market paused

Five chart panels all point the same way. The robot joins that direction.

$1,654 a month on its own

12% win rate reached the target32% no loss won or came back to the entry

1005 trades on 10 markets · 125 won, 685 lost, 195 ended flat

On its own, every trade at full size and nothing withdrawn: after every CTI trading cost (spread, commission and overnight swap), before the 80% profit split and the challenge fee; after the split it is $1,323 a month; before the trading costs it would be $2,213 a month. The same goes for each market's figure below.

Danger: measured for the whole robot, not for one signal: 0.0% chance of breaking the firm's loss rule within a year (score 2), with this type in the mix.

Paused since 28 Sep: US30

1,005 trades taken in the record · about 84 a year · the Stack prints no label and its rules skip some alignments, and the engine keeps no count of those - so this is what was taken, not everything found

the rules, step by step

Made of: an alignment of the 1m, 15m, 30m, 1H, 4H panels (all one colour), not a printed label; 1005 entries on 10 markets

The signal The 1m, 15m, 30m, 1H, 4H bubbles all showing the same colour on this bar, and NOT on the bar before. It is a state alignment, not a printed label.
Consolidation The 4H bubble must have come out of a consolidation of at least 24 hours.
Slow panels No entry when the 1D, 1W, 1M panels ALL point against the trade.
Entry The close of the alignment bar.
Stop The EMA200 of the last closed 30m bar, and only if it sits between 0.25 and 2.0 of that timeframe's ATR away - then pulled in to the fraction of that distance its 4H bubble age band asks for.
Target and timer Each 4H bubble age band has its own exit (R = the target over the stop actually taken): 0-250h: stop at 50% of the EMA distance, target 16x that distance = 32.0R, break-even at 6h, skipped when the 4H RSI is more than 8 points from 50 in the trade's direction; 250-500h: stop at 70% of the EMA distance, target 8x that distance = 11.4R, break-even at 12h, skipped when the 1H MACD histogram slopes away from the trade; 500-750h: stop at 70% of the EMA distance, target 8x that distance = 11.4R, no timer, skipped when the 1H MACD histogram slopes away from the trade; 750-1000h: stop at 70% of the EMA distance, target 6x that distance = 8.6R, no timer; 1000-1500h: stop at 60% of the EMA distance, target 16x that distance = 26.7R, break-even at 4h, skipped when the 1H MACD histogram slopes away from the trade; over 1500h: stop at 90% of the EMA distance, target 6x that distance = 6.7R, no timer. 4 of the 6 bands can skip a signal. At a band's timer a trade that is ahead has its stop moved to the entry; one that is BEHIND keeps its original stop - there is no recover exit on the Stack.
Size A third of a book trade. On the phase-1 login the robot trades now a book trade risks 1.20%, so a Stack trade risks 0.400% of the account - and 0.100% while the drawdown cut is on (while the closed balance is more than 4 points of the starting balance under its high, every trade (the Stack's too) goes out at 25% of its size). Phase 2 arrives as a new login at 1.20% a book trade, so 0.400% there, and the funded account as another at 1.00%, so 0.333%.
Markets 10 markets: audjpy, eurusd, gbpusd, nzdusd, usa500idxusd, usatechidxusd, usdcad, usdjpy, xagusd, xauusd. Paused (dropped from stack.MARKETS on his word), not traded but still measured on the board: 1 market - since 28 Sep: US30.

The rules that matter most:

Without it the setup fires inside trends that have already run.
The 15m EMA200 was the original; the 30m keeps more of the yearly R once the real spread is charged (REFUSED.md, 7 Sep).
Refusing trades never worked here - at a 12% win rate in the record one winner pays for 10.6 losers after every cost, so a filter has to remove that many losers per winner to break even. Changing the EXIT costs no trades at all (the 12 Sep band study).
Larger Stack sizes breached far more starts in the study before the 12 Sep bands (TRADING_GRAMMAR.md). It is the concentrated stream and the size is what holds it.
A market list frozen on 2014-2020 beat every random pick on 2021-2026 in the Stack's blind test of 9 Sep (LAST_NIGHT.md). Today's list is not that frozen list: USTEC and silver were added after the fact (stack.py); EURJPY and GBPJPY left on 24 Sep, US30 left on 28 Sep. The Stack on the other 13 CTI markets was refused again on 14 Sep (REFUSED.md).

The markets it trades: 10, all of them earning

USA500IDXUSD a third (the Stack's own size) $25 a month · 53 tradesearns
XAGUSD a third (the Stack's own size) $79 a month · 141 tradesearns
GBPUSD a third (the Stack's own size) $102 a month · 104 tradesearns
USATECHIDXUSD a third (the Stack's own size) $139 a month · 57 tradesearns
EURUSD a third (the Stack's own size) $147 a month · 99 tradesearns
USDCAD a third (the Stack's own size) $152 a month · 117 tradesearns
NZDUSD a third (the Stack's own size) $165 a month · 124 tradesearns
XAUUSD a third (the Stack's own size) $200 a month · 84 tradesearns
AUDJPY a third (the Stack's own size) $242 a month · 122 tradesearns
USDJPY a third (the Stack's own size) $401 a month · 104 tradesearns

Paused on 28 Sep: 1 market the robot stopped trading on this type because it lost money after costs - it stays here to show if it turns positive

US30 paused a third (the Stack's own size) -$24 a month · 92 tradesloses in both halves · record before the pause of 28 Sep

Signals it shares:

53 AA-Polka trades were opened inside 36 of its positions, same direction (both live in the robot)
7 NA-EW trades were opened inside 8 of its positions, same direction (the other is not traded)
9 AA-EW trades were opened inside 7 of its positions, same direction (the other is not traded)
5 NA-Polka trades were opened inside 7 of its positions, same direction (both live in the robot)

AA-Polka

the robot trades this8 markets paused

A pullback while every panel agrees on the trend. The robot buys the dip or sells the bounce.

$625 a month on its own

17% win rate reached the target76% no loss won or came back to the entry

479 trades on 6 markets · 81 won, 117 lost, 281 ended flat

On its own, every trade at full size and nothing withdrawn: after every CTI trading cost (spread, commission and overnight swap), before the 80% profit split and the challenge fee; after the split it is $500 a month; before the trading costs it would be $991 a month. The same goes for each market's figure below.

Danger: measured for the whole robot, not for one signal: 0.0% chance of breaking the firm's loss rule within a year (score 2), with this type in the mix.

Paused since 25 Sep: AUDJPY, AUS200, CADJPY, DE40, EURAUD, EURCAD, JPN225; since 26 Sep: US30

3,353 found by the engine in all 26 markets · about 279 a year · 41% of the ones it may trade are taken

the rules, step by step

Made of: AA-Polka prints on the 15m chart, 479 trades on 6 markets

The signal An AA-Polka print on the 15m chart: a pullback print (Polka) while every panel is aligned with the trend - the 'all aligned' family. The label is the EVENT, never the direction.
Direction Read the panels from the top: 1W, 1D, 4H. The first one showing a colour (not consolidation) decides the direction. If that panel is 1D the signal is DROPPED - no trade, it never falls through to 4H. So: 1W decides; if 1W is flat and 1D has a colour, no trade; if 1W and 1D are both flat, 4H decides; all flat, no trade.
Confirmation The signal is a fractal pivot, so it is only known 2 bars later. The entry is the open of the bar after that; the bot skips a plan it finds more than 20 minutes after its entry bar.
Filters Every one of these must hold, or there is no trade: the MACD histogram (normalised by ATR) at or above -0.17 in the trade's direction; the 15m RSI at or above 40 in the trade's direction; price no more than 1 ATR on the wrong side of the 15m EMA50; a target at least 0.5R away; the monthly (1M) panel showing LONG (AA types only).
Stop The extreme of the last 25 hourly candles up to and including the signal's hour: the lowest wick for a long, the highest for a short.
Target If no unfilled gap exists on any panel, the trade is still taken with a flat 5R target. If all 8 info panels (1m, 15m, 30m, 1H, 4H, 1D, 1W, 1M) show the trade's colour, it aims at the nearest 4H gap created in the last 30 days, or a flat 5R when there is none. Otherwise: The nearest unfilled gap on the deciding panel's timeframe, at the price that fills it, searched as far back as needed - and if that timeframe has no gap, the next timeframe down. Skipped if it sits further than 8R away or closer than 0.5R. This type never takes a weekly (1W) target: a trade whose target is a weekly gap is SKIPPED, not re-aimed.
The timer From 4 hours on (checked on every 15-minute pass): if the trade is ahead the stop moves to the entry; if it is behind the TARGET moves to the entry, so it leaves at break-even on the way back.
Held If a Stack entry appears in the SAME 1H bubble on the same side, this trade keeps its own stop and target and the timer is switched off. It can only happen on the Stack's 10 markets.
Size 1.00% of the account per trade at full weight. This type trades on 6 CTI markets: 3 at full size, 1 at half, 2 at a quarter (final_book.TIERS' weights, the scan's markets); while the closed balance is more than 4 points of the starting balance under its high, every trade (the Stack's too) goes out at 25% of its size. Paused (final_book.PAUSED_CELLS), not traded but still measured on the board: 8 markets - since 25 Sep: AUDJPY, AUS200, CADJPY, DE40, EURAUD, EURCAD, JPN225; since 26 Sep: US30.

The rules that matter most:

This is the rule that separates a trade from a coin flip, and the panel that decides is also the timeframe the target is read from.
Entering earlier is look-ahead. The live age guard measures from this bar, not the signal bar - getting that wrong refused every book signal for a day (8 Sep).
Sweeping the stop lookback was tested and 25 candles held; a buffer beyond it was refused (REFUSED.md, 8 Sep).
The flat target is the one rule in the book defensible only from the record, not from the chart (final_book.py): these trades have no level ahead of them, and he chose to take them anyway.
About 6 book trades in ten end at the entry. The timer does not add R - holding every trade instead raises R on every sleeve. What it buys is survival: switched off, the account breached or finished funded years at the robot's own stop far more often, measured with the stop the robot had on 15 Sep (REFUSED.md, 15 Sep 06:50). A shorter and a longer timer were both measured and 4 hours won.
Adopted 10 Sep and kept on 15 Sep only where the Stack agrees. The target fraction did not matter; removing the timer was the whole effect.
1.00% is the bot's setting (trader.RISK_PCT), traded on the real CTI 2-Step challenge since 14 September. Every simulator decides the drawdown cut when a trade OPENS, as the bot does. While the challenge is in phase 1 the bot trades 1.20% instead (trader.PHASE1_RISK_PCT, set on 28 Sep by the bot's changelog); a phase-2 login trades 1.20% (trader.PHASE2_RISK_PCT) and the funded account, a new login, trades 1.00%. What this size earns a month is score 1 at the top of the page, which is measured on the robot's own rules. Larger sizes were measured on 14 Sep: 1.30% fell 10.6 points on the real record, through the firm's 10% line, and was reverted; 1.10% and 1.20% were refused that day for how often they reached the robot's own stop as it then stood (REFUSED.md). The cut puts the bot at reduced size much of the time (REFUSED.md, 14 Sep 21:10).

The markets it trades: 6, all of them earning

EURJPY quarter size $4 a month · 38 tradesearns
USA500IDXUSD full size $16 a month · 157 tradesearns
USATECHIDXUSD full size $22 a month · 117 tradesearns
XAGUSD quarter size $30 a month · 11 tradesearns
USDJPY half size $114 a month · 74 tradesearns
XAUUSD full size $439 a month · 82 tradesearns

Paused on 25 Sep and 26 Sep: 8 markets the robot stopped trading on this type because they lost money after costs - they stay here to show if one turns positive

JPN225 paused full size -$154 a month · 64 tradesloses in both halves · record before the pause of 25 Sep
AUS200 paused half size -$103 a month · 101 tradesloses in both halves · record before the pause of 25 Sep
AUDJPY paused half size -$36 a month · 35 tradesloses · record before the pause of 25 Sep
DE40 paused half size -$34 a month · 103 tradesloses · record before the pause of 25 Sep
US30 paused quarter size -$22 a month · 122 tradesloses · record before the pause of 26 Sep
CADJPY paused full size -$3 a month · 4 tradesloses · record before the pause of 25 Sep
EURAUD paused quarter size -$3 a month · 9 tradesloses · record before the pause of 25 Sep
EURCAD paused quarter size under $1 a month lost · 3 tradesloses · record before the pause of 25 Sep

Signals it shares:

61 signals shared with the Panel Break, which is off the board and not traded
53 of its trades were opened inside 36 Stack positions, same direction (both live in the robot)

NA-Polka

the robot trades this12 markets paused

A pullback while the panels do not all agree. The robot follows the biggest panel that has a colour.

$691 a month on its own

16% win rate reached the target83% no loss won or came back to the entry

567 trades on 12 markets · 93 won, 98 lost, 376 ended flat

On its own, every trade at full size and nothing withdrawn: after every CTI trading cost (spread, commission and overnight swap), before the 80% profit split and the challenge fee; after the split it is $553 a month; before the trading costs it would be $950 a month. The same goes for each market's figure below.

Danger: measured for the whole robot, not for one signal: 0.0% chance of breaking the firm's loss rule within a year (score 2), with this type in the mix.

Paused since 25 Sep: AUDJPY, AUDUSD, AUS200, EURAUD, EURCAD, EURJPY, EURUSD, GBPAUD, NZDJPY, NZDUSD, USATECHIDXUSD, XAGUSD

7,011 found by the engine in all 26 markets · about 584 a year · 22% of the ones it may trade are taken

the rules, step by step

Made of: NA-Polka prints on the 15m chart, 567 trades on 12 markets

The signal An NA-Polka print on the 15m chart: a pullback print (Polka) against a trend the higher panels do NOT all agree on - the 'not aligned' family. The label is the EVENT, never the direction.
Direction Read the panels from the top: 1W, 1D, 4H. The first one showing a colour (not consolidation) decides the direction. If that panel is 1D the signal is DROPPED - no trade, it never falls through to 4H. So: 1W decides; if 1W is flat and 1D has a colour, no trade; if 1W and 1D are both flat, 4H decides; all flat, no trade.
Confirmation The signal is a fractal pivot, so it is only known 2 bars later. The entry is the open of the bar after that; the bot skips a plan it finds more than 20 minutes after its entry bar.
Filters Every one of these must hold, or there is no trade: the MACD histogram (normalised by ATR) at or above -0.17 in the trade's direction; the 15m RSI at or above 40 in the trade's direction; price no more than 1 ATR on the wrong side of the 15m EMA50; a target at least 0.5R away.
Stop The extreme of the last 25 hourly candles up to and including the signal's hour: the lowest wick for a long, the highest for a short.
Target The nearest unfilled gap on the deciding panel's timeframe, at the price that fills it, searched as far back as needed - and if that timeframe has no gap, the next timeframe down. Skipped if it sits further than 8R away or closer than 0.5R. A weekly (1W) target is kept only on the 6 full-size markets; on the other 6 markets a trade whose target is a weekly gap is SKIPPED, not re-aimed.
The timer From 4 hours on (checked on every 15-minute pass): if the trade is ahead the stop moves to the entry; if it is behind the TARGET moves to the entry, so it leaves at break-even on the way back.
Held If a Stack entry appears in the SAME 1H bubble on the same side, this trade keeps its own stop and target and the timer is switched off. It can only happen on the Stack's 10 markets.
Size 1.00% of the account per trade at full weight. This type trades on 12 CTI markets: 6 at full size, 5 at half, 1 at a quarter (final_book.TIERS' weights, the scan's markets); while the closed balance is more than 4 points of the starting balance under its high, every trade (the Stack's too) goes out at 25% of its size. Paused (final_book.PAUSED_CELLS), not traded but still measured on the board: 12 markets - since 25 Sep: AUDJPY, AUDUSD, AUS200, EURAUD, EURCAD, EURJPY, EURUSD, GBPAUD, NZDJPY, NZDUSD, USATECHIDXUSD, XAGUSD.

The rules that matter most:

This is the rule that separates a trade from a coin flip, and the panel that decides is also the timeframe the target is read from.
Entering earlier is look-ahead. The live age guard measures from this bar, not the signal bar - getting that wrong refused every book signal for a day (8 Sep).
Sweeping the stop lookback was tested and 25 candles held; a buffer beyond it was refused (REFUSED.md, 8 Sep).
Targets are chart levels, never a fixed R multiple. The near edge of the gap and a cap on the distance were both tested and refused (REFUSED.md, 8 Sep).
About 6 book trades in ten end at the entry. The timer does not add R - holding every trade instead raises R on every sleeve. What it buys is survival: switched off, the account breached or finished funded years at the robot's own stop far more often, measured with the stop the robot had on 15 Sep (REFUSED.md, 15 Sep 06:50). A shorter and a longer timer were both measured and 4 hours won.
Adopted 10 Sep and kept on 15 Sep only where the Stack agrees. The target fraction did not matter; removing the timer was the whole effect.
1.00% is the bot's setting (trader.RISK_PCT), traded on the real CTI 2-Step challenge since 14 September. Every simulator decides the drawdown cut when a trade OPENS, as the bot does. While the challenge is in phase 1 the bot trades 1.20% instead (trader.PHASE1_RISK_PCT, set on 28 Sep by the bot's changelog); a phase-2 login trades 1.20% (trader.PHASE2_RISK_PCT) and the funded account, a new login, trades 1.00%. What this size earns a month is score 1 at the top of the page, which is measured on the robot's own rules. Larger sizes were measured on 14 Sep: 1.30% fell 10.6 points on the real record, through the firm's 10% line, and was reverted; 1.10% and 1.20% were refused that day for how often they reached the robot's own stop as it then stood (REFUSED.md). The cut puts the bot at reduced size much of the time (REFUSED.md, 14 Sep 21:10).

The markets it trades: 12, all of them earning

GBPUSD half size $11 a month · 29 tradesearns
US30 quarter size $14 a month · 15 tradesearns
HK50 full size $22 a month · 48 tradesearns
USDJPY half size $26 a month · 23 tradesearns
DE40 half size $33 a month · 37 tradesearns
USDCAD half size $37 a month · 49 tradesearns
GBPJPY full size $50 a month · 76 tradesearns
FRA40 half size $55 a month · 47 tradesearns
JPN225 full size $84 a month · 54 tradesearns
XAUUSD full size $95 a month · 53 tradesearns
CADJPY full size $130 a month · 94 tradesearns
USA500IDXUSD full size $133 a month · 42 tradesearns

Paused on 25 Sep: 12 markets the robot stopped trading on this type because they lost money after costs - they stay here to show if one turns positive

EURUSD paused full size -$82 a month · 35 tradesloses in both halves · record before the pause of 25 Sep
AUS200 paused half size -$31 a month · 33 tradesloses in both halves · record before the pause of 25 Sep
USATECHIDXUSD paused full size -$30 a month · 37 tradesloses · record before the pause of 25 Sep
AUDUSD paused quarter size -$20 a month · 51 tradesloses · record before the pause of 25 Sep
EURCAD paused quarter size -$18 a month · 47 tradesloses in both halves · record before the pause of 25 Sep
AUDJPY paused half size -$15 a month · 22 tradesloses in both halves · record before the pause of 25 Sep
NZDJPY paused quarter size -$11 a month · 42 tradesloses in both halves · record before the pause of 25 Sep
GBPAUD paused half size -$9 a month · 31 tradesloses · record before the pause of 25 Sep
EURAUD paused quarter size -$9 a month · 38 tradesloses in both halves · record before the pause of 25 Sep
EURJPY paused quarter size -$3 a month · 10 tradesloses in both halves · record before the pause of 25 Sep
NZDUSD paused quarter size -$3 a month · 34 tradesloses · record before the pause of 25 Sep
XAGUSD paused quarter size -$1 a month · 23 tradesloses · record before the pause of 25 Sep

Signals it shares:

54 signals shared with the Panel Break, which is off the board and not traded
5 of its trades were opened inside 7 Stack positions, same direction (both live in the robot)

On paper: measured, not traded

on paper1

A strategy on paper is not one of the indicator's signal types and the robot does not trade it: it is measured on the past charts and kept apart from the tested money.

The DAX morning breakout

on paper

Not traded by the robot: it is on paper on Bawz's word - not in the tested money.

A new signal, not one of the indicator's types: the DAX breaking out of its first half hour, with the 1-hour panel behind it.

621 trades from 15 Sep 2017 to 16 Sep 2026 · added up at full size with the margin limit of the run on paper: early (Sep 2017 - Aug 2020) $773 a month, later (Sep 2020 - Sep 2026) $951 a month · its 10 best trades made 52% of everything it made · it made money in 5 of 10 years

on paper, on Bawz's word
inside the live robot in paper mode since 6 Oct 2026: it logs what it would trade and sends no orders
paper mornings 6-12 Oct; real orders at half size only after Bawz's second yes
not trading real money yet

the rules, step by step

Market: the DAX (DE40) only.
The range: the high and the low of the first 30 minutes after Frankfurt opens at 09:00 Berlin time (the 09:00 and 09:15 candles). A day whose first candle is unusually quiet is skipped.
The signal: the first 15-minute candle, from 09:30 to 11:45, that closes above the range (a buy) or below it (a sell). Only the first break of the day counts.
The check: the 1-hour panel shows the break's colour, and the 4-hour panel is not the opposite colour. If not, no trade that day.
Entry: at the open of the next 15-minute candle.
Stop: at the other side of the range.
Exit: at 17:30 Berlin time, if the stop was not hit. There is no target.
Size: half the robot's normal risk on a trade.
One trade a day: after a stop there is no second try and no turn the other way.
The robot's own signals come first: no breakout trade while the robot holds its own DAX trade, and a new robot trade on the DAX closes the breakout at once.

The run on paper did not yet apply step 10: 3 of its 621 trades ran beside the robot's own DAX trade.

Made of: opening-range breakout - a new signal, not one of the indicator's types

Added on Bawz's word. A study refused part of this rule on 5 Oct 2026 - see what the studies said. The sealed-years test read NOT SHOWN; a filter inside it was refused on its own the same day; it was found on the same years it is judged on; its twin market FRA40 fails. It is on paper on his word, not because it passed.

What the studies said:

Kept as a candidate at half size - and NOT SHOWN: the study did not prove it.
Carrying the rule to every other market found nothing new.
Its twin market, the French index (FRA40), loses with the same rule.
The DAX was picked after seeing the results - found and checked on the same years.
The panel filters - one of them the 4-hour-not-against filter this version uses - were refused when tested on their own.
No panel filter earned a place; the breakout itself stays NOT SHOWN.
The test on years kept sealed read NOT SHOWN for the DAX (the wide-morning version) and killed the French twin.
In that sealed test the breakout did not beat a coin toss or random days.

The DAX in the 400 runs and the 10 worlds

Run on 5 Oct 2026 with the DAX added at half size beside the robot, under the robot's size brake as it is now: quarter size while it is more than $4,000 under its high. The money row is the same test the tested money comes from. The other rows come from its challenge walk, which has the margin limit but not CTI's one-idea rule, so they read differently from the forecast's chance of getting funded; the robot alone's worlds are the ones in the danger card.

	the robot alone	with the DAX
funded within a year (400 challenge starts)	63%	72%
the month it is typically funded	month 6	month 5
to the bank a month once funded (the middle of 400 funded years)	$1,560	$1,577
accounts lost on the real history (400 challenge starts and 400 funded years)	none	none
accounts lost in the 10 redrawn worlds (400 starts each, closed trades)	29 of 4,000 (4 worlds)	40 of 4,000 (5 worlds)
the worst world: share of its starts that lost the account	4.25%	5.5%

Counted the firm's way - with the losses on trades still open - the robot alone loses 29 of 4,000 starts in the worlds, in 4 worlds. The DAX's trades have not been counted the firm's way yet (their open losses were never worked out), so with the DAX the worlds are counted on closed trades, as the table does for both.

Once funded the DAX adds +$17 a month in this test: inside the noise - a test like this cannot tell a gain that small from luck.

Read together it is a trade-off: with the DAX the account is funded more often and sooner, and more starts are lost in the redrawn worlds.

It stays on paper: Bawz put it there on his word, and it is not in the tested money.

Found on the chart, not yet good enough to trade

not yet unlocked7 of 10

These are signals Bawz has spotted himself, and each one is a lock with a key still to be cut. 3 of the 7 have a strategy measured on the record: NA-OPS ($66 a month) earns a little on its own but is parked, and NA-EW and AA-EW lose money; AA-OPS, No room, Wrong side of the cloud and Daily Snipe were measured too and refused - no rule for them earned its place, so only their signals are counted. Every box here is a piece of the project's future income: each one that turns into a rule adds to score 1 above, and it only unlocks if score 2 does not move - a strategy that earns more and makes the account easier to lose is not an improvement, and this page has turned two of those down already. The money on AA-EW, NA-EW and NA-OPS is still measured the old way - a flat cost a trade and no overnight swap: they have not been through the fee audit that charged the robot's own trades every CTI cost, so compare them with care.

NA-OPS

parked

Why it is not traded: parked 12 Sep by his decision while he studied it by eye.

An opening move while the panels disagree. The robot follows the biggest panel that has a colour.

$66 a month on its own

11% win rate reached the target84% no loss won or came back to the entry

494 trades on 20 markets · 52 won, 80 lost, 362 ended flat

On its own, every trade at full size and nothing withdrawn: after a flat 0.04R a trade for trading costs (an estimate: this type is not in the charged book, so CTI's real costs are not in it), before the 80% profit split and the challenge fee; after the split it is $52 a month. The same goes for each market's figure below.

Danger: not measured on its own; the robot's 0.0% (score 2) leaves it out, because the robot does not trade it.

2,926 found by the engine in all 26 markets · about 244 a year · 22% of the ones it may trade are taken

the rules, step by step

Made of: NA-OPS prints on the 15m chart, 494 trades on 20 markets

The signal An NA-OPS print on the 15m chart: an opening print (OPS, the first move of a new bubble) while the panels are not all aligned. The label is the EVENT, never the direction.
Direction Read the panels from the top: 1W, 1D, 4H. The first one showing a colour (not consolidation) decides the direction. If that panel is 1D the signal is DROPPED - no trade, it never falls through to 4H. So: 1W decides; if 1W is flat and 1D has a colour, no trade; if 1W and 1D are both flat, 4H decides; all flat, no trade.
Confirmation The signal is a fractal pivot, so it is only known 2 bars later. The entry is the open of the bar after that; the bot skips a plan it finds more than 20 minutes after its entry bar.
Filters Every one of these must hold, or there is no trade: the MACD histogram (normalised by ATR) at or above -0.17 in the trade's direction; the 15m RSI at or above 40 in the trade's direction; price no more than 1 ATR on the wrong side of the 15m EMA50; a target at least 0.5R away.
Stop The extreme of the last 25 hourly candles up to and including the signal's hour: the lowest wick for a long, the highest for a short.
Target The nearest unfilled gap on the deciding panel's timeframe, at the price that fills it, searched as far back as needed - and if that timeframe has no gap, the next timeframe down. Skipped if it sits further than 8R away or closer than 0.5R. A weekly (1W) target is kept only on the 5 full-size markets; on the other 15 markets a trade whose target is a weekly gap is SKIPPED, not re-aimed.
The timer From 4 hours on (checked on every 15-minute pass): if the trade is ahead the stop moves to the entry; if it is behind the TARGET moves to the entry, so it leaves at break-even on the way back.
Held If a Stack entry appears in the SAME 1H bubble on the same side, this trade keeps its own stop and target and the timer is switched off. It can only happen on the Stack's 10 markets.
Size 1.00% of the account per trade at full weight. This type would trade on 20 CTI markets: 5 at full size, 6 at half, 9 at a quarter (final_book.TIERS' weights, the scan's markets); while the closed balance is more than 4 points of the starting balance under its high, every trade (the Stack's too) goes out at 25% of its size.

The rules that matter most:

This is the rule that separates a trade from a coin flip, and the panel that decides is also the timeframe the target is read from.
Entering earlier is look-ahead. The live age guard measures from this bar, not the signal bar - getting that wrong refused every book signal for a day (8 Sep).
Sweeping the stop lookback was tested and 25 candles held; a buffer beyond it was refused (REFUSED.md, 8 Sep).
Targets are chart levels, never a fixed R multiple. The near edge of the gap and a cap on the distance were both tested and refused (REFUSED.md, 8 Sep).
About 6 book trades in ten end at the entry. The timer does not add R - holding every trade instead raises R on every sleeve. What it buys is survival: switched off, the account breached or finished funded years at the robot's own stop far more often, measured with the stop the robot had on 15 Sep (REFUSED.md, 15 Sep 06:50). A shorter and a longer timer were both measured and 4 hours won.
Adopted 10 Sep and kept on 15 Sep only where the Stack agrees. The target fraction did not matter; removing the timer was the whole effect.
1.00% is the bot's setting (trader.RISK_PCT), traded on the real CTI 2-Step challenge since 14 September. Every simulator decides the drawdown cut when a trade OPENS, as the bot does. While the challenge is in phase 1 the bot trades 1.20% instead (trader.PHASE1_RISK_PCT, set on 28 Sep by the bot's changelog); a phase-2 login trades 1.20% (trader.PHASE2_RISK_PCT) and the funded account, a new login, trades 1.00%. What this size earns a month is score 1 at the top of the page, which is measured on the robot's own rules. Larger sizes were measured on 14 Sep: 1.30% fell 10.6 points on the real record, through the firm's 10% line, and was reverted; 1.10% and 1.20% were refused that day for how often they reached the robot's own stop as it then stood (REFUSED.md). The cut puts the bot at reduced size much of the time (REFUSED.md, 14 Sep 21:10).

The markets it trades: 20 in all, 10 of them losing money on this type

JPN225 full size -$44 a month · 34 tradesloses in both halves
DE40 half size -$13 a month · 31 tradesloses in both halves
FRA40 half size -$9 a month · 21 tradesloses
EURCAD quarter size -$8 a month · 37 tradesloses in both halves
HK50 full size -$8 a month · 21 tradesloses in both halves
CHFJPY quarter size -$4 a month · 24 tradesloses
USDJPY half size -$4 a month · 12 tradesloses in both halves
NZDJPY quarter size -$1 a month · 37 tradesloses
XAGUSD quarter size under $1 a month earned · 37 tradesearns
AUDUSD quarter size under $1 a month lost · 18 tradesloses
NZDUSD quarter size under $1 a month lost · 5 tradesloses in both halves
US30 quarter size $1 a month · 20 tradesearns
USDCHF quarter size $1 a month · 13 tradesearns
AUS200 half size $2 a month · 25 tradesearns
AUDJPY half size $3 a month · 14 tradesearns
GBPAUD half size $10 a month · 17 tradesearns
USA500IDXUSD full size $10 a month · 22 tradesearns
EURJPY quarter size $12 a month · 16 tradesearns
CADJPY full size $43 a month · 60 tradesearns
GBPJPY full size $73 a month · 30 tradesearns

Signals it shares:

40 signals shared with the Panel Break, which is off the board and not traded

NA-EW

refused

Why it is not traded: dropped 14 Sep, his call: too small.

An expected-win print while the panels disagree. Measured, and it loses money.

-$31 a month on its own

10% win rate reached the target86% no loss won or came back to the entry

402 trades on 19 markets · 41 won, 56 lost, 305 ended flat

On its own, every trade at full size and nothing withdrawn: after a flat 0.04R a trade for trading costs (an estimate: this type is not in the charged book, so CTI's real costs are not in it), before the 80% profit split and the challenge fee; after the split it is -$25 a month. The same goes for each market's figure below.

Danger: not measured on its own; the robot's 0.0% (score 2) leaves it out, because the robot does not trade it.

4,043 found by the engine in all 26 markets · about 337 a year · 13% of the ones it may trade are taken

the rules, step by step

Made of: NA-EW prints on the 15m chart, 402 trades on 19 markets

The signal An NA-EW print on the 15m chart: an expected-win print while the panels are not all aligned. The label is the EVENT, never the direction.
Direction Read the panels from the top: 1W, 1D, 4H. The first one showing a colour (not consolidation) decides the direction. If that panel is 1D the signal is DROPPED - no trade, it never falls through to 4H. So: 1W decides; if 1W is flat and 1D has a colour, no trade; if 1W and 1D are both flat, 4H decides; all flat, no trade.
Confirmation The signal is a fractal pivot, so it is only known 2 bars later. The entry is the open of the bar after that; the bot skips a plan it finds more than 20 minutes after its entry bar.
Filters Every one of these must hold, or there is no trade: the MACD histogram (normalised by ATR) at or above -0.17 in the trade's direction; the 15m RSI at or above 40 in the trade's direction; price no more than 1 ATR on the wrong side of the 15m EMA50; a target at least 0.5R away.
Stop The extreme of the last 25 hourly candles up to and including the signal's hour: the lowest wick for a long, the highest for a short.
Target The nearest unfilled gap on the deciding panel's timeframe, at the price that fills it, searched as far back as needed - and if that timeframe has no gap, the next timeframe down. Skipped if it sits further than 8R away or closer than 0.5R. This type never takes a weekly (1W) target: a trade whose target is a weekly gap is SKIPPED, not re-aimed.
The timer From 4 hours on (checked on every 15-minute pass): if the trade is ahead the stop moves to the entry; if it is behind the TARGET moves to the entry, so it leaves at break-even on the way back.
Held If a Stack entry appears in the SAME 1H bubble on the same side, this trade keeps its own stop and target and the timer is switched off. It can only happen on the Stack's 10 markets.
Size 1.00% of the account per trade at full weight. This type would trade on 19 CTI markets: 5 at full size, 6 at half, 8 at a quarter (final_book.TIERS' weights, the scan's markets); while the closed balance is more than 4 points of the starting balance under its high, every trade (the Stack's too) goes out at 25% of its size.

The rules that matter most:

This is the rule that separates a trade from a coin flip, and the panel that decides is also the timeframe the target is read from.
Entering earlier is look-ahead. The live age guard measures from this bar, not the signal bar - getting that wrong refused every book signal for a day (8 Sep).
Sweeping the stop lookback was tested and 25 candles held; a buffer beyond it was refused (REFUSED.md, 8 Sep).
Targets are chart levels, never a fixed R multiple. The near edge of the gap and a cap on the distance were both tested and refused (REFUSED.md, 8 Sep).
About 6 book trades in ten end at the entry. The timer does not add R - holding every trade instead raises R on every sleeve. What it buys is survival: switched off, the account breached or finished funded years at the robot's own stop far more often, measured with the stop the robot had on 15 Sep (REFUSED.md, 15 Sep 06:50). A shorter and a longer timer were both measured and 4 hours won.
Adopted 10 Sep and kept on 15 Sep only where the Stack agrees. The target fraction did not matter; removing the timer was the whole effect.
1.00% is the bot's setting (trader.RISK_PCT), traded on the real CTI 2-Step challenge since 14 September. Every simulator decides the drawdown cut when a trade OPENS, as the bot does. While the challenge is in phase 1 the bot trades 1.20% instead (trader.PHASE1_RISK_PCT, set on 28 Sep by the bot's changelog); a phase-2 login trades 1.20% (trader.PHASE2_RISK_PCT) and the funded account, a new login, trades 1.00%. What this size earns a month is score 1 at the top of the page, which is measured on the robot's own rules. Larger sizes were measured on 14 Sep: 1.30% fell 10.6 points on the real record, through the firm's 10% line, and was reverted; 1.10% and 1.20% were refused that day for how often they reached the robot's own stop as it then stood (REFUSED.md). The cut puts the bot at reduced size much of the time (REFUSED.md, 14 Sep 21:10).

The markets it trades: 19 in all, 9 of them losing money on this type

CADJPY full size -$35 a month · 24 tradesloses
USDCHF quarter size -$17 a month · 39 tradesloses in both halves
EURCAD quarter size -$11 a month · 33 tradesloses in both halves
AUDUSD quarter size -$9 a month · 24 tradesloses in both halves
AUDJPY half size -$8 a month · 12 tradesloses in both halves
NZDJPY quarter size -$7 a month · 32 tradesloses in both halves
GBPAUD half size -$6 a month · 29 tradesloses in both halves
XAGUSD quarter size -$5 a month · 34 tradesloses in both halves
DE40 half size -$2 a month · 27 tradesloses
HK50 full size $1 a month · 8 tradesearns
XAUUSD full size $1 a month · 8 tradesearns
NZDUSD quarter size $2 a month · 8 tradesearns
USATECHIDXUSD full size $3 a month · 3 tradesearns
EURAUD quarter size $5 a month · 24 tradesearns
EURUSD full size $5 a month · 10 tradesearns
CHFJPY quarter size $7 a month · 35 tradesearns
USDJPY half size $7 a month · 10 tradesearns
AUS200 half size $13 a month · 29 tradesearns
GBPUSD half size $25 a month · 13 tradesearns

Signals it shares:

27 signals shared with the Panel Break, which is off the board and not traded
7 of its trades were opened inside 8 Stack positions, same direction (the other is not traded)

AA-EW

parked

Why it is not traded: loses on its own and in every era.

An expected-win print while every panel agrees. Too few trades to earn anything.

-$38 a month on its own

11% win rate reached the target85% no loss won or came back to the entry

54 trades on 4 markets · 6 won, 8 lost, 40 ended flat

On its own, every trade at full size and nothing withdrawn: after a flat 0.04R a trade for trading costs (an estimate: this type is not in the charged book, so CTI's real costs are not in it), before the 80% profit split and the challenge fee; after the split it is -$31 a month. The same goes for each market's figure below.

Danger: not measured on its own; the robot's 0.0% (score 2) leaves it out, because the robot does not trade it.

1,987 found by the engine in all 26 markets · about 166 a year · 5% of the ones it may trade are taken

the rules, step by step

Made of: AA-EW prints on the 15m chart, 54 trades on 4 markets

The signal An AA-EW print on the 15m chart: an expected-win print while every panel is aligned. The label is the EVENT, never the direction.
Direction Read the panels from the top: 1W, 1D, 4H. The first one showing a colour (not consolidation) decides the direction. If that panel is 1D the signal is DROPPED - no trade, it never falls through to 4H. So: 1W decides; if 1W is flat and 1D has a colour, no trade; if 1W and 1D are both flat, 4H decides; all flat, no trade.
Confirmation The signal is a fractal pivot, so it is only known 2 bars later. The entry is the open of the bar after that; the bot skips a plan it finds more than 20 minutes after its entry bar.
Filters Every one of these must hold, or there is no trade: the MACD histogram (normalised by ATR) at or above -0.17 in the trade's direction; the 15m RSI at or above 40 in the trade's direction; price no more than 1 ATR on the wrong side of the 15m EMA50; a target at least 0.5R away; the monthly (1M) panel showing LONG (AA types only).
Stop The extreme of the last 25 hourly candles up to and including the signal's hour: the lowest wick for a long, the highest for a short.
Target The nearest unfilled gap on the deciding panel's timeframe, at the price that fills it, searched as far back as needed - and if that timeframe has no gap, the next timeframe down. Skipped if it sits further than 8R away or closer than 0.5R. This type never takes a weekly (1W) target: a trade whose target is a weekly gap is SKIPPED, not re-aimed.
The timer From 4 hours on (checked on every 15-minute pass): if the trade is ahead the stop moves to the entry; if it is behind the TARGET moves to the entry, so it leaves at break-even on the way back.
Held If a Stack entry appears in the SAME 1H bubble on the same side, this trade keeps its own stop and target and the timer is switched off. It can only happen on the Stack's 10 markets.
Size 1.00% of the account per trade at full weight. This type would trade on 11 CTI markets: 4 at full size, 3 at half, 4 at a quarter (final_book.TIERS' weights, the scan's markets); while the closed balance is more than 4 points of the starting balance under its high, every trade (the Stack's too) goes out at 25% of its size.

The rules that matter most:

This is the rule that separates a trade from a coin flip, and the panel that decides is also the timeframe the target is read from.
Entering earlier is look-ahead. The live age guard measures from this bar, not the signal bar - getting that wrong refused every book signal for a day (8 Sep).
Sweeping the stop lookback was tested and 25 candles held; a buffer beyond it was refused (REFUSED.md, 8 Sep).
Targets are chart levels, never a fixed R multiple. The near edge of the gap and a cap on the distance were both tested and refused (REFUSED.md, 8 Sep).
About 6 book trades in ten end at the entry. The timer does not add R - holding every trade instead raises R on every sleeve. What it buys is survival: switched off, the account breached or finished funded years at the robot's own stop far more often, measured with the stop the robot had on 15 Sep (REFUSED.md, 15 Sep 06:50). A shorter and a longer timer were both measured and 4 hours won.
Adopted 10 Sep and kept on 15 Sep only where the Stack agrees. The target fraction did not matter; removing the timer was the whole effect.
1.00% is the bot's setting (trader.RISK_PCT), traded on the real CTI 2-Step challenge since 14 September. Every simulator decides the drawdown cut when a trade OPENS, as the bot does. While the challenge is in phase 1 the bot trades 1.20% instead (trader.PHASE1_RISK_PCT, set on 28 Sep by the bot's changelog); a phase-2 login trades 1.20% (trader.PHASE2_RISK_PCT) and the funded account, a new login, trades 1.00%. What this size earns a month is score 1 at the top of the page, which is measured on the robot's own rules. Larger sizes were measured on 14 Sep: 1.30% fell 10.6 points on the real record, through the firm's 10% line, and was reverted; 1.10% and 1.20% were refused that day for how often they reached the robot's own stop as it then stood (REFUSED.md). The cut puts the bot at reduced size much of the time (REFUSED.md, 14 Sep 21:10).

The markets it trades: 4 in all, 4 of them losing money on this type

USA500IDXUSD full size -$28 a month · 15 tradesloses in both halves
USATECHIDXUSD full size -$6 a month · 11 tradesloses
XAUUSD full size -$3 a month · 15 tradesloses in both halves
USDJPY half size -$1 a month · 13 tradesloses

Signals it shares:

9 of its trades were opened inside 7 Stack positions, same direction (the other is not traded)

AA-OPS

no strategy yet

Why it is not traded: refused 14-15 Sep: under the book's rules, from the move on the JPY crosses, and under his panel rule of 14 Sep.

An opening move while every panel agrees. Measured and refused: no rule for it earned its place.

measured, refused

947 found by the engine in all 26 markets · about 79 a year

the rules, step by step

Made of: AA-OPS prints: an opening signal while every panel is aligned

no rule kept - every one tested was refused

Signals it shares:

not counted: it has no trades of its own to set beside another strategy's

No room

no strategy yet

Why it is not traded: refused 14 Sep: the gap ahead is a magnet, not a wall.

A print with a gap right in front of it. Measured and refused: no rule for it earned its place.

measured, refused

3,435 found by the engine in all 26 markets · about 286 a year

the rules, step by step

Made of: any book print (Polka, OPS or EW, AA or NA) with a 4H fair value gap closer than 3 ATR in front of it - so every one of them is also one of the printed types above

no rule kept - every one tested was refused

Signals it shares:

every one of its signals is also a signal of one of the printed types above - it is a filter on them, not a signal of its own

Wrong side of the cloud

no strategy yet

Why it is not traded: refused 14 Sep: the 'inside the cloud' move was an artefact.

A print on the wrong side of the daily cloud. Measured and refused: no rule for it earned its place.

measured, refused

361 found so far · about 37 a year

the rules, step by step

Made of: any book print on the wrong side of the daily Ichimoku cloud - the book has no cloud filter, so it trades these like any other print

no rule kept - every one tested was refused

Signals it shares:

every one of its signals is also a signal of one of the printed types above - it is a filter on them, not a signal of its own

Daily Snipe

no strategy yet

Why it is not traded: refused 14 Sep by his five-step method: no move after the alignment.

His own idea: the daily and 15-minute bubbles agree first. Measured and refused: no rule for it earned its place.

measured, refused

22,839 found so far in 26 markets · about 1,906 a year

the rules, step by step

Made of: his own idea: the daily and 15m bubbles first agreeing, an alignment rather than a printed label

no rule kept - every one tested was refused

Signals it shares:

not counted: it has no trades of its own to set beside another strategy's

How it is tested

More on the Monte Carlo tests

Score 1, the money, comes from the same kind of draw: the mix test's 400 year starts, drawn at random across the 12-year record (the last year is never a start, so each run gets a full year).

Nothing is invented: the forecast and the worlds replay only trades from the record, and the stress rows below make those same trades worse on purpose.

400

starts, in two tests. Score 1 is counted over the mix test's 400 year starts, drawn at random across the 12-year record, each played for a full year - as an account that is funded from the first day, so no test fee and no waiting to pass are in it. Funded within a year, score 2 and the money in hand after the first year come from the forecast's own 400 one-year runs, which start only from 2017: from 357 different starting points, each one a trade in the record, between 6 Jan 2017 and 10 Sep 2025.

0.0%

of starts break the rule - the median over the 10 redrawn worlds. A Monte Carlo test: the record is rebuilt 10 times a calendar month at a time, each month swapped for one drawn at random from the record, and 400 starts are replayed in each. Across the 10 worlds the median share of starts that break the rule is 0.0%, and the median of the survivors' deepest falls is 9.55% below the balance their phase began with - 5 of the 10 worlds fall deeper than that. The worst world breaks the rule in 4.25% of its starts (17 of 400), survivors at most 9.98% down. The harshest count: 4 of 10 worlds had at least one breach. Every world is listed below the box.

0.0%

breach under stress. Every trade made more expensive than it really was: at +0.02R a trade 0.0% of starts break the rule, at +0.04R 0.0%; if one winner in ten turned into a loss, 0.0%.

The 10 redrawn worlds, worst first
world	starts that break the rule	deepest fall of the starts that survive
world 8	4.25% (17 of 400)	9.98%
world 7	2.0% (8 of 400)	9.44%
world 6	0.75% (3 of 400)	9.93%
world 5	0.25% (1 of 400)	9.998%
world 3	0.0% (0 of 400)	9.62%
world 10	0.0% (0 of 400)	9.61%
world 9	0.0% (0 of 400)	9.50%
world 2	0.0% (0 of 400)	9.22%
world 4	0.0% (0 of 400)	9.19%
world 1	0.0% (0 of 400)	8.75%

Which simulator each figure comes from: the top cards and the one-, two- and five-year forecast use the robot's own account rules, the margin guards and CTI's idea rule included, with phase 1 at 1.20% a trade (its size since 28 Sep); score 1 replays the mix test's starts on the robot's own payout, brake and margin rules, and so do the stress rows and the 10 redrawn worlds; none of the three has CTI's idea rule - so they and the forecast can still disagree. Under Then and now, the first-coded figure comes from a third, older simulator - the card simulator - also without the idea rule.

What every trade is charged: City Traders Imperium's charges as closely as they can be measured - the commission per lot, the spread of that market at that hour, the overnight swap for every night the trade is held (three nights' worth on the triple-swap day) and, when a stop gapped past its level, the price it really got. Three of these are estimates: the spread is Dukascopy's hour-by-hour spread for that market, with CTI's own quote as the floor on five stock indices; the swap is today's CTI swap rates applied to every night back to 2014; and the commission is charged at today's price level. That comes to 0.17R a trade on average. Before this change on 26 Sep, the project charged a flat 0.04R on each Polka trade and a measured spread on the Stack's, and no overnight swap at all. The stress row adds its extra cost on top of all of it. Every test fee is paid, and a lost account is bought again at the full $549 - CTI's 15% reset discount is not counted, which is the cautious side.

The robot stops opening trades at 10.0% down in total, which is the firm's own line, and 4.0% down on the day, inside the firm's 5% daily line. These tests do not all read a loss the same way. Score 1, the redrawn worlds and the stress rows replay the starts of the mix test, which reads the firm's loss lines on the closed balance only - so they cannot see a loss on a trade that is still open. The account cards at the top of the page (funded within a year, score 2 and money in hand) read them the way the firm does, on the balance plus the losses on trades still open, bar by bar.

On paper: the full working

one funded $100,000 account, run once from 15 Sep 2017 to 18 Sep 2026
	the robot alone	on paper
to the bank a month	$1,618	$2,020
of the account a month	1.98%	2.48%
a typical month (the middle one), of the account	0.15%	0.72%
without its 3 best months, of the account a month	1.49%	1.97%
first half (Sep 2017 - Feb 2022), to the bank a month	$1,411	$1,483
second half (Mar 2022 - Sep 2026), to the bank a month	$1,820	$2,547
worst drop from a high	$8,032	$8,382
lowest against the start	-$7,071	-$7,639
worst day	-$2,489	-$3,093
accounts lost	none	none
orders the margin limit refused	105	141
years on paper made more than the robot alone		7 of 10

first half (Sep 2017 - Feb 2022): the robot alone $1,411 a month, on paper $1,483 · second half (Mar 2022 - Sep 2026): $1,820, on paper $2,547

one by one they add +$270, +$182; together +$402 - they share one account, so the parts do not add up to the whole.

The same trades with each win or loss decided by a coin toss (and for Stack sized up the same bigger size given to as many prints picked at random) instead added between -$858 and +$285 (-$254 in the middle case); 200 of 200 such coin-toss copies added less than the real trades. Luck alone moves this figure by about $612 either way. The gain of +$402 is not bigger than that: too small to tell from luck in one run.

The DAX morning breakout

Added on Bawz's word. A study refused part of this rule on 5 Oct 2026 - see what the studies said.

The sealed-years test read NOT SHOWN; a filter inside it was refused on its own the same day; it was found on the same years it is judged on; its twin market FRA40 fails. It is on paper on his word, not because it passed.

The rule: On the DAX, the first 30 minutes after Frankfurt opens at 09:00 make a range. When a 15-minute candle closes outside it (up to three hours after the open), with the 1-hour panel showing the break's colour and the 4-hour panel not against it, the robot would enter at the next candle's open, put its stop at the range's other edge, and close the trade at 17:30 Berlin time. One trade a morning.
Markets: DE40
Version: V3 (1-hour panel with the break, 4-hour not against), today's stop at the range's other edge, flat at 17:30 Berlin
Size: half size - half size on paper until the full test (rule Z1). At full size V3's worst drop was $7,036 and its worst day -$3,205 (MARGIN.md, B x1.0); one run cannot say which size is safe.
Joined: 5 Oct 2026 on Bawz's word
Found and checked: Found and checked on the same years (Sep 2017 - Sep 2026) - no fresh years were left to test it on. The sealed-years test that was run (on the wide-morning version) read NOT SHOWN.
Picked from about 172 versions tried: V3 picked from 3 panel versions on 44 market-and-open rows (132 rule-and-market rows); the stop and the 17:30 exit picked from the 40 rows of the exits grid
Its own trades: added up at full size with this run's margin limit: early (Sep 2017 - Aug 2020) $773 a month, later (Sep 2020 - Sep 2026) $951; without its 10 best trades $522 and $381; positive in 5 of 10 years
Trades: 604 of 621 taken; the margin limit refused 8 because the order needed more margin than was left and 9 because the positions already open were as big as the limit allows
At the same time: open at the same time as the robot's book on the same market: 3 trades
At the same time: open at the same time as stack-s-up on the same market: 0 trades
What it adds: the account banks $1,750 a month without it, so it adds +$270
Read other ways: with no margin rule it adds +$513; with orders made smaller instead of refused, +$198
Luck: the same trades with each win or loss decided by a coin toss instead added between -$822 and +$195; 200 of 200 such copies added less than it
Now: it fails this joining rule now: Its own trades make money early and later, and in most years.
Not measured: CTI's spread in the first minutes after the open is not measured; the costs use the usual spread for that hour (OPEN_BREAK.md, 'What breaks').
Bawz's word: said after the 1-hour-panel version (V3, half size) was explained to him in bank money: “Okay, now I understand. So yeah, okay. Do what you recommend”
Where it stands: on paper, on Bawz's word; inside the live robot in paper mode since 6 Oct 2026: it logs what it would trade and sends no orders; paper mornings 6-12 Oct; real orders at half size only after Bawz's second yes; not trading real money yet

What the studies said

Kept as a candidate at half size - and NOT SHOWN: the study did not prove it. Keep the DAX breakout (V3, today's stop) as the candidate it already was: half size, about $200 to $400 a month beside the bot on years already seen, NOT SHOWN. store/_screens/open-break/OPEN_BREAK.md, line 57
Carrying the rule to every other market found nothing new. This search did not find a new signal. store/_screens/open-break/OPEN_BREAK.md, line 57
Its twin market, the French index (FRA40), loses with the same rule. The DAX's nearest copy fails. store/_screens/open-break/OPEN_BREAK.md, line 23
The DAX was picked after seeing the results - found and checked on the same years. the DAX alone (a list chosen WITH HINDSIGHT; from 2017-09-15) store/_screens/open-break/OPEN_BREAK.md, line 27
The panel filters - one of them the 4-hour-not-against filter this version uses - were refused when tested on their own. Panel filters on the DAX morning breakout (5 Oct 2026, orb-de-panels-v1) - REFUSED; both picks read once on 2020-09 .. 2026 and both read NO; nothing deployed REFUSED.md, line 2665
No panel filter earned a place; the breakout itself stays NOT SHOWN. No filter is a candidate. The breakout stays where orb-eu-final-v1 left it (NOT SHOWN). store/_screens/orb-de-panels-v1/REPORT.md, line 114
The test on years kept sealed read NOT SHOWN for the DAX (the wide-morning version) and killed the French twin. DE40 (DAX) w125ct, primary: NOT SHOWN. FRA40 (CAC 40) w125ct, sister: KILLED. Nothing is deployed. store/_screens/orb-eu-final-v1/REPORT.md, line 7
In that sealed test the breakout did not beat a coin toss or random days. the coin and random days are not beaten store/_screens/orb-eu-final-v1/REPORT.md, line 9

Its study says: the DAX adds something like $200 to $400 a month at half size; the walk cannot say more exactly than that store/_screens/open-break/OPEN_BREAK.md, line 43

A second check: an independent script rebuilt the trades from the raw bars; it confirms the counting, not the edge (store/_screens/open-break/CHECK.md: CONFIRMED)

The Stack sized up when confirmed

Added on Bawz's word. A study refused part of this rule on 6 Oct 2026 - see what the studies said.

The study that measured the tilt (stack-range stop1) was refused and did not adopt it. The rangesearch study that found this drawing said to keep his drawing: walked through the whole book it banks $1,697 a month against $1,736 with his, with a deeper worst drop ($8,140 against $7,562), and its first-half lead sat at only the 88th percentile of a search that size. Its R lead on the untouched second half is real-looking (98th percentile). At 1.25x the tilt study's walk banks $1,750 a month, $133 more than today's $1,618 - one walk, inside the ~$150 a month one walk cannot tell from luck - and the size was picked from 4 tried on that same history. It is on paper on his word, not because it passed.

The rule: A size change on the Stack, a type the robot already trades - not a new type and not a new trade. The range is the 30-minute drawing from the rangesearch study: the last opposite 30-minute bubble and the CONS after it, its edges drawn from the candle bodies. A Stack print is confirmed when at least 1 15-minute candle has closed beyond that range in the trade's direction, by more than a quarter of the 15-minute ATR, by the close of the print candle. A confirmed print is traded at 1.25 times today's Stack size; every other Stack print keeps today's size, and the Stack's entry, stop and target stay as today.
Markets: AUDJPY, EURUSD, GBPUSD, NZDUSD, US500, USTEC, USDCAD, USDJPY, XAGUSD, XAUUSD
Version: S-up with the rangesearch drawing c|30m|body|n1|m0.25 (store/_screens/stack-range/rangesearch: bubbles 'c' = the last opposite bubble and its CONS, read on 30-minute bubbles, edges from bodies, 1 close beyond the edge by more than 0.25 x ATR15 by the print candle's close); confirmed prints at 1.25x the Stack's weight (not the study's 1.5x: rangesearch/tilt walked 1.0 / 1.25 / 1.33 / 1.5; Bawz chose 1.33x at ~16:45, then 1.25x at ~17:00 after 1.33x crossed the -$8,000 lowest-point line; 1.33x is the size the full test re-checks), the rest at 1.0x; entry, stop and target as today. Chosen over his own 15-minute drawing on Bawz's word, 6 Oct 2026 ~16:25
Size: 1.25x size - A size change, not a strategy beside the robot: the robot's own Stack rows are swapped for the same rows with the confirmed ones at 1.25x, so it is walked at the robot's own Stack size (rule Z1) and the 1.25x is in its rows. 587 of the 1005 Stack rows are confirmed; 493 of the 866 inside this run's window.
Joined: 6 Oct 2026 on Bawz's word
Found and checked: Picked on the first half (Stack prints before Sep 2021) and read once on the second half; the money walk covers the same years (the store's Stack prints, Oct 2014 - Sep 2026; this run from Sep 2017) - no fresh years were left for the money.
Picked from about 220 versions tried: 4 sizes for the confirmed prints in rangesearch/tilt (1.0, 1.25, 1.33, 1.5) after 180 range drawings in rangesearch (131 different after identical ones were merged), on top of the 36 versions before it: 7 size-and-stop versions in stop1 (11 with its 4 reverse trades) and the stack-range family's entry (about a dozen), fix (6) and exit (7) versions, all refused
Its own trades: added up at full size with this run's margin limit: early (Sep 2017 - Aug 2020) $5,550 a month, later (Sep 2020 - Sep 2026) $5,426; without its 10 best trades $2,906 and $2,400; positive in 10 of 10 years
Trades: 787 of 866 taken; the margin limit refused 53 because the order needed more margin than was left and 26 because the positions already open were as big as the limit allows
At the same time: open at the same time as the robot's book on the same market: 109 trades
At the same time: open at the same time as dax-morning-breakout on the same market: 0 trades
What it adds: the account banks $1,838 a month without it, so it adds +$182
Read other ways: with no margin rule it adds +$310; with orders made smaller instead of refused, +$253
Luck: the same bigger size given to as many of its prints picked at random instead added between -$144 and +$381; 158 of 200 such copies added less than it
Not measured: The rows keep the robot's book's charged costs (commission, spread, swap, gapped stops), which grow with the size; whether a 1.25x order fills as well as today's is not measured.

What the studies said

The tilt study walked the confirmed prints at 1.0x, 1.25x, 1.33x and 1.5x with the same builder; this block is 1.25x. "x": 1.25, store/_screens/stack-range/rangesearch/tilt/table.json, line 140
At 1.25x the whole book sends $1,750 a month to the bank, +$133 on today's Stack ($1,618); 1.33x read $1,813 and 1.5x $1,697 (one walk). "bank_month": 1750.1225196923901, store/_screens/stack-range/rangesearch/tilt/table.json, line 142
Its worst drop is $7,960, against $8,032 with today's Stack and $8,140 at 1.5x. "worst_gap_usd": 7959.898474725567, store/_screens/stack-range/rangesearch/tilt/table.json, line 143
First half of the walk (Sep 2017 - Feb 2022): $1,660 a month, against $1,411 today. 1659.9490813423845, store/_screens/stack-range/rangesearch/tilt/table.json, line 151
Second half (Mar 2022 - Sep 2026): $1,839 a month, against $1,820 today - ahead in both halves, barely in the second. 1838.6564409814869, store/_screens/stack-range/rangesearch/tilt/table.json, line 156
The margin limit refused 114 orders, against 105 today and 138 at 1.5x. "margin_refused": 114, store/_screens/stack-range/rangesearch/tilt/table.json, line 203
On the untouched second half its 1.5x tilt added +144R, at the 98th percentile of luck, and beat his drawing in 95.9% of bootstrap resamples. | c|30m|body|n1|m0.25 | 338 of 554 | +74.49R | +203.27R | 99 | 249 of 450 | +58.71R | +144.39R | +130.21R | 98 | 95.9% | store/_screens/stack-range/rangesearch/REPORT.md, line 18
It passed all three of the study's pre-registered second-half lines. c|30m|body|n1|m0.25: l1 yes l2 yes l3 yes store/_screens/stack-range/rangesearch/REPORT.md, line 22
The study's verdict: in the money walk no drawing beat his. no range drawing beats his where it counts - the money walk store/_screens/stack-range/rangesearch/REPORT.md, line 5
At the study's 1.5x, walked through the whole book it sent $1,697 a month to the bank, $39 less than his drawing, with a worst drop of $8,140. | S-up, c|30m|body|n1|m0.25 | 587 | $1,697 | -$39 | $8,140 | store/_screens/stack-range/rangesearch/REPORT.md, line 33
Its first-half lead was not unusual for a search of 180 drawings: luck finds one that good about one time in 8. The real best (+74.49R) sits at the 88th percentile - a search this size finds a winner that good by luck about one time in 8. store/_screens/stack-range/rangesearch/REPORT.md, line 46
Against his drawing in money: no better, not proven worse - the gap is inside one walk's noise. so the honest reading is 'no better in money', not 'proven worse' store/_screens/stack-range/rangesearch/REPORT.md, line 9
The study that measured the 1.5x tilt (with his drawing) was refused: no version earned more than today's Stack outside the noise. REFUSED - no version earns more than today outside the noise store/_screens/stack-range/stop1/REPORT.md, line 5
With his drawing the tilt was the best version, +$119 a month on one run - inside the ~$150 a month that one run cannot tell from luck. The best of the seven part-1 versions is S-up at $1,736 a month (+$119), inside the ~$150 noise. store/_screens/stack-range/stop1/REPORT.md, line 7
The refusal kept the tilt as a lead and did not adopt it. Lead kept (inside noise, not adopted): S-up = confirmed prints at 1.5x size REFUSED.md, line 2961

Its study says: it kept doing so on the untouched second half (its extra R from the 1.5x tilt sat at the 98th-99th percentile of luck there; his own drawing sat at the 56th) store/_screens/stack-range/rangesearch/REPORT.md, line 7

A second check: every drawing's range and close count were worked out again from the chart cut right after each print candle and agreed on all 1,004 prints (no look-ahead); it confirms the flag, not the edge (store/_screens/stack-range/rangesearch/REPORT.md: identical on 1004 of 1004 prints (all 180 definitions at once))

The DAX breakout with a thin stop - an idea still being looked at, not added in

The same DAX morning breakout with the stop at the far end of the break candle instead of the range's other edge - bigger positions for the same risk.

It needs a rule the robot does not have: an order too big for the margin limit made smaller instead of refused. With the robot's rule as coded it banks $94 a month LESS than today's stop (MARGIN.md: not shown once the margin rule is in).

Its study says: The thin stop is a lead only together with a shrink rule, at half size, and it is NOT SHOWN store/_screens/open-break-exits/margin/MARGIN.md, line 30

How a strategy joins, and how it leaves

A strategy joins the paper book when all of these hold:

No account is lost in the run - with the robot and this strategy alone, and with everything on paper together.
The worst drop is no deeper than the robot alone's by more than $1,000, and stays well inside the firm's lines.
Its own trades make money early and later, and in most years.
The account banks more with it than without it.
Its numbers were re-counted by a second check, its trades are saved, and no refusal in REFUSED.md covers this rule or any filter it uses.
It does not trade a signal the robot or another strategy on paper already trades.

A strategy that fails a joining rule may still be put on paper on Bawz's word. It is then labelled so on its card, beside every verdict against it. His word puts it on paper; it does not pass the full test for it.

Every new strategy goes on paper at half size (quarter if half fails J1 or J2); the full test decides the size the robot would trade. A replacement of one of the robot's own types is walked at the robot's own size for that type.

On every rebuild the whole paper book is walked against J1 and J2's lines. If it fails, no money figure is shown - only the fact.

It leaves when it does not pass the full test (the test doing its job), when Bawz takes it out, or when a fault is found in its trades - and when it passes the full test and the robot trades it, it leaves the paper book because it is now in the tested money. A figure on paper that goes down because a strategy failed the full test is the map working, not a loss.

How this run is made

One funded $100,000 account, from 15 Sep 2017 to the last closed trade, in the real order. The robot's own rules: 1% risk a trade on the balance, quarter size while it is more than $4,000 under its high, its 4% day stop and 10% total stop, the firm's $5,000 day line and $10,000 line, the firm's payout rule with everything taken out at each payout, the firm's margin limit - an order too big for it is refused, as the robot refuses it - and CTI's one-idea rule between the robot's trades and the strategies on paper, read the safe way. A signal is never traded twice. Every trading cost we can measure is in.

What one run cannot tell

It cannot tell luck from skill: one path, one order of trades.
It cannot tell a gain of about $612 a month from nothing.
The new strategies were found on these same years, so the only truly new test left for them is a trial run on live prices.

Left the paper book

Nothing has left the paper book yet.

Worked out 2026-10-06 17:20:44. The robot's trades run to 18 Sep 2026; DAX morning's to 16 Sep 2026; Stack sized up's to 9 Sep 2026.

The best month: the full working

How it was counted: the robot's real book, one funded $100,000 account, walked once over the whole history; the robot's margin rule; everything taken out at each payout, from October 2014 to September 2026 - 144 months.

These months come from one run through the whole history, month by month - not from the 400 one-year runs behind the money score, so their average is not the money score.

The next four: January 2018 +$20,365, March 2021 +$15,955, September 2020 +$14,264, August 2019 +$12,260.

The next two worst: April 2021 -$3,939, September 2019 -$3,847.

Of 144 months, 64 made money, 75 lost money and 5 had no trade.

On average the account made $1,537 a month, and $1,254 a month reached the bank.

The three best months made 26% of all the profit in the 12 years - most months are small, a few are big.

In December 2025 the biggest trades were: gold (AA-Polka) +$5,055; gold (AA-Polka) +$4,936; gold (AA-Polka) +$4,700. The three biggest made $14,691 of the month's $20,536.

Who found the edges

His question, 28 Sep 2026: was he better than the AI at spotting the patterns that gave the robot its edge - and why?

Yes - in what he found, though not in how often he was right. Every signal the robot trades is his: the Stack, AA-Polka and NA-Polka. Those three signals carry the whole $2,970 a month that their boxes on the board show, before the firm's split. And 19 of the 31 trading rules live today began with him - 12 his own, and 7 that he set and Claude shaped. The AI's 10 live rules are 5 filters, 2 rules on where a trade aims or stops and 3 choices of what to trade where, laid on top of his signals. Two more have no recorded origin.

Why: about three in four of his own ideas and of the AI's were thrown out - 47 of his own 65 trading ideas (72%) against 47 of the AI's 63 (75%); of the 26 he set and Claude shaped, 19 (73%). The difference is in what survived - and in what was tried: 25 of his own 65 were ideas that start trades (signals and entries), against 7 of the AI's 63 and 8 of the 26 he set and Claude shaped. Every signal the robot trades is one of his own, and 8 of the 10 live rules on where a trade aims or stops - the exits the money comes from - began with him: 5 his own and 3 set by him and shaped by Claude. The AI's survivors shape those trades - which to skip, where a stop or target may sit, which markets to trade - rather than start them. That follows from where the ideas start. His start from a move he saw on a real chart and a reason for it, which brings in something the record's columns did not hold. A search can only recombine what is already there, and most of what it finds does not hold up on years it has not seen: none of the AI's own 102 entry filters held up, and the whole sweep, one idea in the table, was refused. Neither side escaped that trap. Counted by one rule - an idea picked or judged on part of the record that failed, or went unproven, on the part it had not seen - 15 of his own 65 trading ideas (23%) were fooled that way against 15 of the AI's 63 (24%), and 7 of the 26 he set and Claude shaped. The checks threw out 14 of his own, 14 of the AI's and 7 of the shared ones. His pause of the Polka market cells the Lab showed losing and the AI's market scan that picks each type's markets by their past record stay in the robot on his decision.

The honest part: idea for idea he was right only a little more often. 15 of his own 65 trading ideas made it into the robot (23%), against 11 of the AI's 63 (17%), and 47 of his own were refused. When he picked markets by their own past record, his ideas fell into the same trap as the AI's searches. And his prints on their own do not mark bigger moves than random bars do: the money comes from where the trades aim and how they are managed - and 8 of the 10 live rules for that began with him too (5 his own, 3 set by him and shaped by Claude).

who began it	ideas tested	made it into the robot	live today
his own	65	15 (23%)	12
set by him, shaped by Claude	26	7 (27%)	7
his, counting the shared ones	91	22 (24%)	19
the AI on its own	63	11 (17%)	10
no recorded origin	4	2 (50%)	2
all	158	35 (22%)	31

The 31 trading rules live today, and who began eachHide the list

his own · 12

NA-Polka signal type (his indicator's print, not all panels aligned, Polka sequence)
AA-Polka signal type (his indicator's print with all slow panels aligned)
The Stack: 1m/15m/30m/1H (+4H) panels turn one colour on the same bar, stop on the EMA200
Direction from the highest coloured panel of 1W/1D/4H, the BUY/SELL word never trusted
The book's stop: the extreme wick of the last 25 hourly candles
Targets at chart levels: the far edge of the nearest unfilled fair value gap, never a fixed R
The break-even timer: after N hours the stop moves to the entry (12h in his notes, 4h live)
Recover at the entry: a trade still behind at the timer gets its TARGET moved to the entry
AA-Polka with all eight panels aligned aims at the nearest 4H gap of the last 30 days, else 5R
Pause the 19 Polka market cells the Lab showed losing
Pause AA-Polka on US30 (loses once every CTI cost is charged)
The Stack stops trading US30 (loses once every CTI cost is charged)

set by him, shaped by Claude · 7

The Stack only inside 4H bubbles born from a consolidation of a day or more
The held rule: a book trade sharing a 1H bubble with a same-side Stack entry keeps its plan, no timer
AA-Polka prints with no gap anywhere are traded with a flat 5R
Stack exits per 4H bubble age band (stop, target, timer per band)
Stack band gates: 4H RSI near 50, 1H MACD slope
Refuse a Stack entry when the daily, weekly and monthly all point against it
The Stack stops trading EURJPY and GBPJPY

the AI on its own · 10

Skip a signal whose gap target is more than 8R away
Momentum filter: skip if > 1 ATR the wrong side of the 15m EMA50 or 15m RSI < 40
15m MACD histogram floor -0.17 at entry
Drop signals whose direction comes from the daily panel
AA circuit breaker: AA signals only while the monthly trend is LONG
market_scan.py: a type must earn across all markets, then markets are chosen and tiered by past edge
Satellites never aim at a weekly gap
The Stack's stop on the 30m EMA200 (a wider stop pays less spread per R)
The Stack's market list (ten study markets plus US30, USTEC, silver)
Stop trading NA-OPS

no recorded origin · 2

A minimum target of 0.5R
Keep the weaker markets but trade them small (satellites at 30%)

How the ideas were counted, and the limitsHide how they were counted

his own: his idea, rule, chart read or the question that named the pattern
set by him, shaped by Claude: he set the idea or question and Claude shaped, optimised or recombined it; a question of his that named no pattern and was answered by Claude's search counts here
the AI on its own: Claude's or an agent's own search with no idea from him: sweeps, scans, screens, the scout, the beat-finder (a goal of his with every condition the agent's own counts here)
no recorded origin: the records do not name who proposed it
one row: a whole sweep or grid counts as ONE idea; variants_tested gives the number of cells where the source states it

Per idea the two are close: his own ideas landed 15 of 65 (23%), the AI's own 11 of 63 (17%), and the share still live today is nearly the same (18% against 16%) - a gap of a few ideas on samples this size. His side leads clearly only with the 26 ideas he set and Claude shaped (7 landed): 22 of 91 (24%) against 17%.
47 of his 65 trading ideas were refused and 3 parked. Three of his signal types that landed (NA-OPS, NA-EW, AA-EW) were later taken out (too small, losing, or dropped for safety), and one of his live rules - the pause of the Polka market cells the Lab showed losing - failed its fair test and stands on his decision alone.
The rule behind the 'fooled' count, applied to every row on both sides: the record must show the idea, or the version of it a test carried forward, picked, fitted or looking good on one part of the price record - a first half, the earlier years, the markets or trades it was found on, or a market's first trades - and then losing, turning the other way, or doing no better than chance or random picks on a part it had not seen; an idea that did badly on the very part it was picked on is not counted. Each marked row carries the record's own words for it in edge_origins.csv. His marked ideas include his market-picking rules (positive markets only and the pause of the Polka market cells the Lab showed losing). One failure sits in three rows: picking markets by their past record is the AI's market scan, live on his decision, and its re-run on the honest engine, both counted as fooled, and its re-run on the true clock at his 'renew the market list', set by him and shaped by Claude, which lost even with hindsight and is not counted. The sizing and brake rules are counted apart from the table and are not judged by this rule on the page.
Several of his refused ideas were real effects that did not pay rather than misreadings (the AA-Polka panel split, the leader market; with Claude's search, telling winners from losers); others were plain misses (the Bubble Prediction matched his one example but not the rest, and its revision was fitted to three trades, the 4H age cap to four examples, the forecast lines called a coin).
The AI's live contributions shape every trade the bot takes (the momentum, MACD, daily-panel and AA-breaker filters, the 8R cap, the Stack's 30m stop and market list; the 0.5R minimum target has no recorded origin). The money cannot be split between a signal and the filters on it: the dollars per signal are the Lab's straight-line figures.
The bot's own prints do not mark bigger moves than random bars: measured against ordinary bars on the same market in the beat-finder run of 24 Sep, the moves after NA-Polka's prints came out 1.03 and 0.99 times as big in the two halves of the record, AA-Polka's 1.03 and 0.98 and the Stack's 0.93 and 0.97, and no signal type prints near the start of big moves more often than random bars (15 Sep). So 'a move he saw on a real chart' is where his best ideas start, not a claim that his prints predict moves; the money is made by where the trades aim and how they are managed (the gap targets, the timer, recover, the Stack's far target), which is also mostly his.
A row is an idea, not a variant: one AI row can be a sweep of 102 filters, and many of his rows were tested as large grids that Claude built. Origin was read from the project's own words ('his idea', 'his rule', 'mine'). Seven rows name no one: the four trading ideas the table counts under 'no recorded origin' and three account rules (sizing and brakes), which the table leaves out. Claude's recombinations of his building blocks count as set by him and shaped by Claude, and so do questions of his that named no pattern and were answered by Claude's search (telling winners from losers, filtering out losing trades, why a trade that was ahead falls back, renewing the market list); the beat-finder, whose goal was his but whose 17 conditions were the agent's own, stays with the AI on its own.
Every breach share and dollar figure behind these verdicts describes the one price record the project has; dates for the 1-6 Sep rows are to the day at best.

How the work is split between him and the machines: How this project finds its edges.

How this project finds its edges

On 28 Sep 2026 Bawz asked why his strategies, once tuned, nearly always showed an edge while Claude's own searches mostly failed. This is the answer.

The first coded backtest

His indicator first met a coded backtest in a chat on 16-19 March 2026. Bawz taught a scanner his Expected Win signal one example at a time from his own chart, and caught its shortcut: "are you really using my full indicator script?" He let timing 1 to 4 hours off his chart pass once, then refused it on 19 March, wanting exact data to build a strategy on. He exported his own TradingView charts, the slow ones back to 1971.

The chat shows him fetching the price data by hand, month by month. Since 1 Sep 2026 the work runs in Claude Code, where the data comes fast. Of the 19 rules set in that March chat, 12 are still followed.

Where his ideas and the searches differ

His ideas come from watching charts, before data, usually with a reason price should move, like a gap to fill. The backtest checks an idea instead of hunting for one, and the reason can be real: price reached a 4-hour gap close in front within three days on 87% of 3,217 prints (14 Sep), though that alone was no strategy.
A search alone tries many versions on one history, and some look great by luck. Three checks catch them: a plan fixed first, the second half of history locked, a comparison with random picks. 102 entry filters were tried; none survived. Picking markets by their past results fooled both sides: the markets a search re-chose (8 Sep) looked far better, then no better than random picks on the later years they had not seen, and his own list of each signal's positive markets (15 Sep) earned less than keeping every market the robot traded, on the half it had not seen.
He sets a small, sensible frame - the signal, a target at the nearest unfilled gap, a stop at a chart level - and Claude tunes a few choices around a real effect, which is what survives. Moving the target nearer gave up money (24 Sep, on the first half of the record): $1,034 a month at the gap's middle, $900 at its near edge, $1,213 at the far edge, with 0.2%, 8.2% and 2.5% breach.

The honest part

It is not that Bawz is always right. 47 of his 65 trading ideas were tested and refused, among them trading the reverse (9 and 11 Sep), the Daily Snipe (10 Sep) and the look-alike forecast (25 Sep); so were ideas he set and Claude shaped, like filtering out losing trades (14 Sep) and NA-EW by his five-step method (15 Sep). The 26 Sep fee audit found the simulations charged about $447 a month in costs where CTI takes about $1,270: earlier numbers were too good.

His ideas are the right raw material, not always the answer. Even the refused reverse of 11 Sep began with a split he saw by eye, and the split was real.

The split we work by

Neither half works alone: the March playbook (24-26 signals) failed later testing; the 102 filters found nothing.

Bawz finds the pattern by eye, gives the reason price should move, sets the frame and the bar a change must clear, and makes the big calls. Since 13-15 Sep Claude may land a change that clears his bar without asking him first: NA-OPS came back into the bot on 14 Sep on the mix test alone.
Claude runs the three checks, kills the lucky ones, tunes the real ones, and reports dollars a month and breach on a $100k account.

The next step

Next: strategies from NA-EW and the Forecast lines now on his chart (v3, installed 28 Sep), tested the same way. The line alone called direction like a coin (25 Sep) and an NA-EW print alone has no direction (15 Sep), so each starts from what he sees plus a confirmation after the print, judged on prints after 25 Sep.

The market data

Every trading day the engine can see, one ring a market, one spoke a day - out when the day closed up, in when it closed down. The circle is the calendar 2004 to 2026: a gap is missing history. The deeper rings were bought and put to work on 23 September. They did not change a single traded bar: they sit IN FRONT of the record, so the weekly and monthly panels are warmed up instead of still filling. Open a picture to go down to single days: every ring carries its market's name, and pointing at a ring (tapping it on a phone) names the market and the month.

Rings of daily price data per market, as the engine had it before the swap — **What the engine used before 23 September** · 11 full circles · 103,862 days

Rings of daily price data per market, as the engine has it now — **What the engine uses now** (in use since 23 September; the traded record is unchanged, the new years sit in front of it as warm-up) · 13 full circles · 132,606 days

Recent changes to the robot

Times are Stockholm time.

2026-10-06 12:16 The phase-2 size: trader.py PHASE2_RISK_PCT = 1.20, with PHASE2_LOGINS still empty, so no login trades it yet and the live bot is unchanged - phase 1 stays 1.20% a trade and the fu...
2026-10-06 01:16 The DAX open-break (V3) sleeve installed in PAPER mode (stage P): the bot runs the rule on DE40 each morning and logs the plan but sends nothing - the ORB line reads 'PAPER BEFORE_...
2026-10-05 20:01 The drawdown brake moved from 2% to 4%: trader.py CUT_TRIGGER_PCT 2.0 -> 4.0 together with ladder3.py CUT_TRIGGER 4.0, and the loop was restarted through the BubbleBot task. New po...

The project, day by day

Before this calendar: the project began around 2023-2024 with the indicator, its signal categories, strategy ideas and a few samples backtested by hand.

What happened on each day the project's records cover, newest first - 33 days with an entry, from 16 Mar to 7 Oct. Days and times are Stockholm's. The last 14 days are open; the rest are under the button.

Wed 7 Oct 2026
- refusedHis V6 range rule on the new 30-minute bubbles, run over all 1,005 Stack prints: it lets in 71 of 125 winners but is worth next to nothing a month, inside the noise, and less than the paper candidate.
- The old '2% brake' sentences in the project docs are corrected to 4%; no code touched.
Tue 6 Oct 2026
- went liveThe DAX breakout is installed in the robot in PAPER mode at 01:16 (his yes at 01:05): each morning it logs the trade it would take on DE40 at half size and sends no order. Paper mornings 6-12 Oct.
- publishedStrategy Lab v71: the forecast on brake A (funded within a year 70%, month 6) and the DAX shown as paper.
- His call: phase 2 trades at 1.20% risk like phase 1; the funded account stays at 1.00%.
- went liveThe phase-2 size goes into the robot at 12:16 (proven 12:28), but no login uses it yet, so the live robot trades exactly as before.
- refusedHis breakout-and-retest read on the Stack: as a filter or late entry the one history reads about a third below today's robot; four entry fixes, an early exit, a tighter stop and a reverse trade are refused too.
- A search of 180 range drawings finds his hand drawing no better than luck at telling Stack winners, but the 30-minute-bubble drawing separates them in the second half.
- refusedTwo more readings of his range (touch rules v4 and v5) and a size-up on prints with the four big panels all aligned are refused; v5 sits at chance.
- His word at 17:00: the 30-minute drawing at 1.25 times size on confirmed Stack prints goes on paper, not into the robot.
- publishedStrategy Lab v72: the paper book holds the Stack size-up, ahead of the robot alone; luck check 79th percentile, not proven.
- A worker team of free AI models, checked by Claude, gets a Workers Desk app for tasks, questions to Bawz and results; it is never part of the bot.
Mon 5 Oct 2026
- Why the robot was off: 27% of open market hours since 14 Sep (15% since the move to the new PC) - shutdowns, an app update and his own stop; no code crashed. Ten fixes are proposed.
- went liveThe robot is started again at 04:44 on his word and runs normally from 05:00; it now starts through Windows' scheduler, not from a chat window.
- He asks for the Strategy Lab as still, equal bubbles in the Bubble Bot's style: a minimal first view, details when a bubble is pressed.
- refusedHis 1H-consolidation NA-EW rules: both AUDJPY examples reproduce, but the against-the-signal side loses 0.22-0.30R a trade and the with-the-signal side breaches 30-36% - refused.
- Chart against the store on 58 NA-EW prints: the label sits on the same candle 72% of the time, and the 4H, 1D and 1H panels agree 97-100%.
- refusedOpening-range breakout on the Nasdaq, S&P 500 and Dow: the Nasdaq loses 0.037R a trade after costs; the French index version is refused at the account read.
- The one lead: a DAX 30-minute breakout. At his yes the sealed years 2020-2026 are opened once: +0.097R a trade after costs on 342 trades, read NOT SHOWN; the CAC version is killed.
- A DAX breakout with the 1H panel behind it, closed at 17:30 Berlin, beside the robot adds about $200-400 a month; no other market carries it and no exit beats 17:30.
- He wants the DAX breakout in the robot - on paper first - and makes one new research rule: screen ideas on the one real 12-year history, the 400 simulated starts return once the book earns over 3% of the account a month.
- Payout: half of each month's profit out with half left as a cushion, reversed the same day - everything is taken out, as the cards assume.
- publishedStrategy Lab v69 and v70 published on the same link: still, equal bubbles in sections, the live robot as a read-only mirror, the 400 starts and ten redrawn worlds on the danger card.
- His yes to brake version A, with the full test in front of him: funded within a year 63% against 42%, $1,560 against $830 a month once funded, against 29 of 4,000 redrawn starts losing an account instead of 11, all in phase 1.
- went liveBrake A goes live at 20:01 (restart through the scheduled task, proven 20:13): new positions go out at a quarter size once the balance is 4% under its high, where before it was 2%.
- A post-mortem tool that writes a card for every closed trade, a desktop Lab app with an Improve-the-signals workspace, and a 'believer' agent, the hopeful counterpart of the skeptic.
Fri 2 Oct 2026
- Bawz stops the robot by hand at 16:11 after a manual trade of his was closed; the account history shows the firm's server closed it for having no stop loss, not the robot. The robot stays off until 5 Oct.
Tue 29 Sep 2026
- went liveHive Mind Forecast v4 goes live on his TradingView (01:05): the same lines as v3, but replay no longer freezes - a step takes about a tenth of a second.
- He walks six NA-EW prints by hand (JP225, silver, USDJPY, gold, GBPUSD, EURUSD) and states the rules: against or with the 4H bubble and the forecast line, the stop on the 15m bubble's wick, the target on a daily Fibonacci level, one stop-and-reverse.
- An engine that plays his NA-EW rules over every print is built; it reproduces his seven hand-checked prints apart from small data differences.
- refusedThe old NA-EW volume bracket (+147R in September) was a leak: its volume veto read a bar before the bar had closed. Timed honestly, every version earns less than the robot's book.
- refusedHis NA-EW sequence rules v2 (15 pre-registered versions): every one breaches in 95-100% of the simulated starts.
- refusedNA-EW first break of the 15m bubble beats entering on random bars, the first honest edge of the day, but in the full test 5 of 10 redrawn worlds breach.
- Two quick checks of his ideas - the forecast as a go/no-go and a sweep-and-reclaim entry - find nothing worth a study.
Mon 28 Sep 2026
- The USTEC trade moves its target to the entry at 00:15 and closes there at 02:02: net -$86.85, the weekend swap and commission.
- went liveThe Stack stops trading US30, and the units fix goes live with a restart at 10:22, so the bot's AA/NA labels match the book.
- went liveCTI's idea rule goes live - the bot no longer adds to a losing trade idea past 2% of the starting balance at the stops, nor opens a 4th same-direction position onto one - with a recorder of CTI's own spreads, and phase 1 drops from 1.30% to 1.20% a trade on his word (restart 19:01).
- Hive Mind Forecast v3 on his chart: one line per print, each from its own candle.
- The PC now has 64 GB of RAM: research may use it fully, CPU at most 70 C. The project calendar and the March conversation log written.
- publishedStrategy Lab republished at 18:55 (every CTI cost charged, the Stack off US30, the About Bawz, How-we-work and edge-count sections), then again with the forecast on the robot's live rules (phase 1 at 1.20%, CTI's idea rule), when to buy the second account, all 10 redrawn worlds listed one by one and his own words saying the robot never trades on the news.
Sun 27 Sep 2026
- Sunday: the bot comes back by itself at 21:17 after the PC was restarted; at the market reopen the new timer fix waits for a fresh price instead of acting on a two-day-old one.
Sat 26 Sep 2026
- Fee audit: the simulations charged about $447 a month of costs where CTI really takes about $1,270; every card rebuilt with every CTI cost.
- went liveUS30 AA-Polka paused; the timer now waits for a fresh price after a weekend, and the news calendar is fetched once a week (live 02:15).
- refused2% and 3% risk stress-tested: no size clears the bar, 1.30% stays.
- Hive Mind V16 installed with deeper weekly and monthly history, so the chart's AA/NA reads like the bot's.
Fri 25 Sep 2026
- went livePhase 1 goes to 1.40% on his word (00:48), then to 1.30% (14:18) once the real-path walker showed 1.40% breaching about 5% of starts.
- went live19 losing Polka cells paused (16:34); the bot opens its USTEC AA-Polka buy at 19:00.
- Hive Mind V15 with a volume profile, a new Bubble Bot window with a progress view, and a separate Hive Mind Forecast (off by default) after three look-alike forecast tests came out a coin flip.
- publishedMarket Rings app v2, and Lab v63 to v66 with the report cards at 1.30%.
Thu 24 Sep 2026
- publishedMore Lab versions up to v55: the real account's panel, the board split into what the bot trades and what is found but unproven, every market named in each rule.
- went liveNA-OPS no longer traded (10:35); the Stack drops EURJPY and GBPJPY and phase 1 goes back to 1.00% (restart 21:13).
- A video-study tool, minute tick volume recovered from raw files, and an intraday path harness proven.
- refusedMarket pruning, an early cut, a volume-shelf target, a daily beat-finder and choosing markets from their own record.

Show all 33 daysShow the last 14 days only

Wed 23 Sep 2026
- The data swap: 112 files with years of warm-up history go into the store on his 'Yes swap'.
- Hive Mind V14: a print decided by the daily panel says 'no trade'; all 20 alerts moved onto it.
- refusedAA-Polka's partial, target and timer-off ideas and NA-Polka without its weekly trades refused; lock-a-small-win measured, not adopted.
- publishedThe Lab republished through the evening and night (v37 onwards): two- and five-year money cards, the team, 'By Bawz' and 'The road here' - his own story.
Tue 22 Sep 2026
- Three AIs on one board: Claude keeps proof, the bot and verification; Codex writes code and pages; Freebuff does mechanical work.
- refusedThe halt study and 'stop only on the firm's lines' refused; the honest held-rule book becomes the research incumbent.
- went liveThe bot's total stop moves out to CTI's own 10% line on his permission (12:31).
- publishedA minimal Lab a non-trader understands replaces the published Lab; deep history for all 26 markets decoded.
Mon 21 Sep 2026
- The project moves to the new PC: store rebuilt and the book reproduced to the last trade.
- went liveThe bot goes live on the new PC at 11:45; after a reboot killed it at night, logon tasks now keep the bot, the drive mapping and the health guard alive.
- A free bid/ask price archive (Dukascopy) found and its decoder proven; the project map written; the old PC's memory imported.
- A halted account is a lost account: every study is judged on breach plus halt.
Thu 17 Sep 2026
- The project folder's state at 09:49 is the copy that reaches the new PC. Nothing is recorded from 17 to 20 September; the bot keeps running on the old PC and sends no new orders.
Wed 16 Sep 2026
- went livePass-and-flatten, and both drawdown stops on balance plus floating losses, deployed at 04:46.
- publishedCards and Lab rebuilt on what the bot actually does, checked by a skeptic.
- Data checked against CTI and TradingView (6 markets trusted as they are); the warm-up studies run.
- The transfer bundle packed for the new PC.
Tue 15 Sep 2026
- The research kit and shared data store built; the trading grammar and a technical-analysis encyclopedia written.
- The stored history was an hour off CTI's clock; the book moved onto CTI time (about $1,402 a month at 0.0% breach in one history).
- went liveA look-ahead in the gap finder fixed and deployed; phase 1 trades 1.10% from 21:46.
- refusedA day of refusals, among them every way of picking markets from their past, late partials and 1.50% in phase 1.
Mon 14 Sep 2026
- went liveLIVE: the bot starts trading the real CTI 2-Step $100k challenge at 21:28 on his 'Go live now', with a news blackout and an account-switch fix.
- went liveRisk back to 1.00% (1.30% tried and reverted within the hour); NA-EW out ('too small'); NA-OPS back in on the mix test; a 200% margin-level floor.
- refusedThe Panel Break, a No-room strategy, the Ichimoku cloud, AA-OPS and a gentler drawdown brake.
- went liveThe first live order at 23:45: CADJPY, NA-Polka buy.
Sun 13 Sep 2026
- NA-EW added back on a measurement; the Stack's level targets refused a second time.
- Hive Mind V13 with per-band Stack exits; the 'report1' money card.
- Money is quoted as dollars a month and breach %, never R alone.
- (claude.ai) Must CTI know about the bot? Automation is allowed; the strategy is what gets a bot rejected.
Sat 12 Sep 2026
- went liveAn AA-Polka with no gap takes a flat 5R; the Stack gets its own exits per 4H bubble age band; risk 1.50% on his word.
- Two clock faults found and fixed: the 1H MACD gate must read the last closed bar, and a double clock conversion had 19 of 27 markets 2-3 hours off.
- A master switch to show his signals or the bot's; the unfilled part of each gap drawn.
- A strategy change is only done when the book, the Lab, the cards and the published Lab are rebuilt.
Fri 11 Sep 2026
- publishedThe Strategy Lab redesigned as a game board: a box per signal type, red to green, every number computed.
- went liveThe bot moves to a $100,000 demo account.
- 'The 0% breach is not a zero': 15 of 25 resampled histories breach.
- V5: the label only draws plans the bot would take; an aligned AA-Polka aims at the nearest recent 4H gap.
Thu 10 Sep 2026
- went liveThe held rule goes live (Hive Mind V2) and risk moves to 1.25% a trade on his word.
- 'Improve it and keep 0% breach': four levers swept, every one already at its best.
- refusedThe Daily Snipe refused as a strategy, kept on the chart as a signal (V3); V4 says when the weekly panel cannot be read.
- Pure technical analysis only - never news; 'report1' means the $100k report card.
Wed 9 Sep 2026
- The $100k report card; the margin cap set per asset class, the firm's way.
- The Stack's market list survives the blind test the book failed; the mix re-judged with the Stack at a third of size.
- The Stack on the chart: 'New Hive Mind V1' - from now on the version is in the name.
- refusedReversing the book on its losing markets, and a cap on the 4H bubble's age.
Tue 8 Sep 2026
- went liveThe age guard read the wrong bar, so the bot could never have traded a book signal - fixed, and the order path proved with a test trade.
- Market selection fails a blind test on the honest engine: nothing changed, and no challenge to be bought on these numbers yet.
- went liveThe first real signal (palladium) refused for size - the tick-value bug fixed; USDSEK and four watchlist pairs added on his word, then removed when measured negative.
- went liveThe four-hour rule was firing at seven hours - fixed.
Mon 7 Sep 2026
- The price data had holes; it was repaired and every earlier number re-measured.
- The five-panel Stack found, and his recover-at-entry rule measured and adopted.
- went liveThe demo bot runs: the 30m-EMA200 Stack joins at 14:50 and the four-hour timer goes on at 20:05 on his word, after four live-path faults were fixed.
- Hive Mind v4 prints the book's plan on every signal; the Bubble Prediction signal built.
Sun 6 Sep 2026
- Overnight research, report cards as pictures, and a first simulation of a small CTI challenge.
- The big correction (weekend of 5-7 Sep): the backtest had placed the break-even stop unconditionally and charged no costs; +539R fell to +97.7R, and the diary warns not to buy a challenge on those numbers.
- Every big move within two days of a print studied; he sees a new pattern in the 4H consolidations.
Sat 5 Sep 2026
- The plan is written: $10,000 through a funded prop-firm account, ending in an automatic trader. Survive first, earn second.
- NA-Polka v1 looks like 92% no-loss - later retracted, because it paid no trading costs.
- City Traders Imperium chosen as the firm; an MT5 demo account opened; the Trading Deck app with research agents.
Fri 4 Sep 2026
- The indicator rebuilt from scratch as 'New Hive Mind', with a gap badge; the old signal logic kept on file.
- His adaptive style coded: direction from the highest coloured panel of 1W, 1D and 4H, target at the nearest unfilled gap, a break-even timer.
- Labels state facts only - direction and pattern; Claude pastes every Pine change into TradingView itself.
Thu 3 Sep 2026
- Data gathered for all his markets; an EMA9 line tested and added.
- His teaching loop begins: he trades three signals of one type by eye and writes the rules, Claude backtests them.
- Never study signals from the last few days - their outcome has not settled.
- An app that opens each study signal in TradingView replay and keeps his notes.
Wed 2 Sep 2026
- Strategies tested on USDJPY, gold, the S&P 500, Nasdaq and Bitcoin; NA-Polka aims at an untapped 4H gap instead of a fixed 8R.
- The labels start printing an entry, stop and target plan; he asks for alerts on the top five pairs.
Tue 1 Sep 2026
- The project restarts in Claude Code: Bawz hands over the Pine script and his strategy notes, and Claude learns the indicator on real forex data.
- His rule from day one: targets and stops are chart levels - gaps, swings, the EMA200 plus ATR - never a fixed R.
- A first strategy 'game' app, and the first check of Claude's signals against his TradingView chart.
Mon 13 Jul 2026
- Side chat (claude.ai): how to judge copy-trading providers - drawdown and track record first, a small test stake.
Thu 19 Mar 2026
- The data hunt: tick files fetched by hand, then his own TradingView exports (145,012 fifteen-minute bars, 2020-2025), because he wants exact times before any strategy.
- First strategy rules: one take-profit, trade either direction, the FVG warning is only a timeframe label; gap detection added on every timeframe.
- The first playbook rests on 24-26 signals - far too few - and none of its numbers survive later testing.
Mon 16 Mar 2026
- First chat about the indicator (claude.ai): Bawz pastes his Pine script, 'The Bubble Strategy', and asks for every 'Expected Win' signal five years back; Claude writes a Python scanner.
- Rules taught by example: an Expected Win is a clean 15m red > orange > green inside a live 4H bubble, an Opening Signal never counts, AA means 4H to 1M all agree, and every time is on his Swedish clock.
- First list: 26 Expected Win signals on GBPUSD in 2025 (22 buy, 4 sell), timing still 1-4 hours off his chart.

Words on this page

Signal: A pattern on the price chart that the indicator marks with a label. The robot only acts on the ones with a green box here.
Funded: The test account belongs to a trading firm. Grow it 10%, then 5%, without breaking their loss rule, and they fund you for real money.
Breach: Losing more than the firm allows: the balance falling 10% under the start, or 5% in one day, counted on the balance plus the losses on trades still open. The account is closed. The robot has its own brake: it stops opening trades at 10% down in total - the firm's own line - and at 4% down on the day, tighter than the firm's 5%.
A month: What the strategy earned per month on average, on a funded $100,000 account, over the whole 12-year record - after every cost the firm charges, as closely as it can be measured: commission, spread and overnight swap. Score 1 is what Bawz keeps after the firm's 80% split; the money on each box of the board is the whole month before that split. Score 1 is not the boxes added up and cut to 80%: it is the typical funded year of 400, played with the robot's brake as the robot keeps it (a payout does not set it back to full size), each trade sized on the balance of the moment, CTI's payout days, the robot's own stops and its margin limit, which is why it sits well below the board's straight-line total. The boxes still waiting to be traded (AA-EW, NA-EW and NA-OPS) are on the old flat cost, with no overnight swap.
Ended flat: A trade that closed at about the price it opened: neither a win nor a loss.

Today's results on this page are computed from the record - book_fee2.csv (109fba3a), 2051 trades on 18 markets over 12 years - and figures from past studies are quoted with the date they were measured. Lab statistics 28 Sep 2026 21:09 Stockholm time (19:09 UTC); cards lab_numbers_2026-10-05.json. Robot stops: 10.0% total, 4.0% a day. Built 07 Oct 2026 04:04 Stockholm time.

Watch the robot

The money and the danger

The signals

The forecast

The project

What a Monte Carlo test is

Why the robot alone reads $1,618 here and $1,560 in the tested bubble

The DAX in the 400 runs and the 10 worlds

About Bawz

Whose indicator this is

Why a robot and not him

What he brings

How he works with the machines

What he is aiming for

The road here

Where it starts

What I did while I was away

Why a robot and not me

What I got wrong about winning

Why it stopped, and why it is back

What I do not have, and what I do

What it is for

The team

1decidesthe owner

2thinksthe AI models, in the apps they run in

Claude

Codex

Freebuff

3does the workwhat Claude has created

Teams made for one job active

The study team · every strategy test

One-job crews · builds and checks

Standing helpers run on Claude Sonnet, except the believer, which runs on the model that calls it

Playbooks they read first

4watchesthe screens

Bawz's screens

The Bubble Bot window

TradingView chart

The Strategy Lab

Market Rings

The team's boards

Task board & message log

Bubble Steer

Bubble Plans

5tradesthe money path

Then and now

If the firm grows the account, and a second one is bought

after one year

after two years

after five years

The board

What the robot trades today

On paper: measured, not traded

The DAX in the 400 runs and the 10 worlds

Found on the chart, not yet good enough to trade

How it is tested

More on the Monte Carlo tests

On paper: the full working

The DAX morning breakout

What the studies said

The Stack sized up when confirmed

What the studies said

The DAX breakout with a thin stop - an idea still being looked at, not added in

How a strategy joins, and how it leaves

How this run is made

What one run cannot tell

Left the paper book

The best month: the full working

Who found the edges

his own · 12

set by him, shaped by Claude · 7

the AI on its own · 10

no recorded origin · 2

How this project finds its edges

The first coded backtest

Where his ideas and the searches differ

The honest part

The split we work by

The next step

The market data