These describe what the market is doing — they are not
advice about what to do. Your own method decides that.
Every reading here is our own measurement, made with settings
we chose, from public gold prices. They are observations and experiments, not forecasts, not claims
about what is true, and not advice. Tap any tile to see the working — and to see where we could
reasonably be argued with.
Bots' money, last days
The chart
Gold, the chart
What's going on
Our score, and what gold did
We publish this because we could not make it work. Across every reading
so far, no relationship between these scores and what gold did next survived
testing — and three of the six barely move at all. The chart is here so you can
see that for yourself instead of taking our word for it. If it ever starts
working, this is where it will show first.
These readings blend into one score the bots trade with. News is what
people are saying; Econ data is the numbers actually released, scored
against what forecasters expected; Smart money is where big funds and
gold ETFs are positioned. A quiet day here is normal — most days are quiet.
The BuilderPlus
Compose a bot of your own from simple blocks — when to enter,
how much to risk, when to exit — and replay it over a real year
of gold in seconds. No coding. The stop loss is always on.
While Touchstone grows, the Builder is free for invited testers.
Ask us for a code.
Opens your mail app with the request ready — we reply with your
personal code, usually within a day.
Twelve characters, in three groups. Dashes and capitals do not
matter. Checking takes a moment on purpose.
What is inside
The seven entry blocks a bot can be built from, and the other
names people use for them. Reading is open to everyone; trying
one needs a code.
What are you trying to build?
Describe it in your own words, any length. If we cannot build it
yet, your words are what tell us what to build next.
Only the words you type here are sent, with the app version and
language. Nothing else — no saved bots, no device details. Your mail
app opens with it written out, and you press send.
A single result never tells you whether a strategy is good — only whether
it was good, once. These pages ask the harder questions: would it
still work if the settings were slightly different, if the same trades had
arrived in a different order, if a broker took its cut, and in which
conditions it actually earns its money.
Is this luck?1,000 alternative years
1,000 years is enough to judge. Bigger counts only smooth the
chart and take longer — they do not change the answer.
What this does. Your year happened in one
particular order. This deals the same kinds of trade again, thousands of times, in
thousands of different orders — as percentages of the account, compounded from the same
starting money, exactly as the real year worked. What it reports is the range of
outcomes those settings could plausibly have produced, including the bad runs that simply
did not come up this time. It is a measure of how much of your result was the method and
how much was the order the cards fell in — not a forecast of next year.
Deep luck testthe same year, re-dealt
The quick test above re-orders the trades. This one re-deals the year itself in whole days and lets the strategy trade each version from scratch — same candles, same start, ending near the same place. It asks a harder question: is the edge in the rules, or in the exact tape you tested on?
For invited testers while it is new — it runs a full year hundreds of times on your phone. Unlock the Builder to use it.
One honest limit: every re-dealt year is built from this year's own days, so it inherits this year's character. A strong trend re-dealt is still a strong trend — this cannot show what the rules do in a year gold goes nowhere.
The luck testevery setting, not just yours
Re-runs the whole year at every value of
each slider. A plateau means the idea works across a range and your exact
number doesn't matter much. A lonely spike means you found a fluke.
Changed since these curves were drawn
These curves were measured around your earlier settings. Run the test again to
redraw them around the one you just picked.
When it worksby market condition
Trades are grouped by what the market was doing when each
one opened. A strategy that only earns in one regime is a strategy with a
switch you haven't built yet.
Over timerolling 3-month result
One yearly total can hide a strategy that died in March.
Quarters and a rolling line show when it worked and when it stopped.
If you took money outsame trades, minus a monthly withdrawal
Every dollar withdrawn stops compounding the moment it leaves.
A bot can be a salary or a snowball — this shows the price of each.
By session and weekday
A saved bot keeps a forward record: from the day you saved it, it is
replayed over every new candle since — the same honest test as a backtest,
except it could not have been tuned to fit. Backtests flatter; forward
records tell the truth. Everything is stored on this phone only.
Forward records, side by side
You are composing a bot of your own. Answer the three questions
below — when to enter, how much to risk, when to get out — and every
change replays a real year of gold on your phone. The hard stop loss
is always on. Broker costs from Tune's Realism card apply here too.
Save it, and it starts an honest forward record under My bots.
Past results never promise future ones.
The Lab replays a bot over a real year of 30-minute spot gold — the same
replay the official backtests use. Move a slider and the whole year
recomputes on your phone. News score held at 0; no spread or slippage
charged, so these numbers are friendlier than real trading.
Past results never promise future ones.
Realismwhat a broker takes
How your broker charges
Read these three off your own trading app. Gold is priced per ounce,
so the gap it shows between the buy and sell price already IS
dollars per ounce — 4265.10 to sell and 4265.45 to buy is a spread of
0.35. Slippage is how much worse than the screen price you
usually get filled. Commission is per 1.00 lot, in and out, if your
broker charges one — many gold accounts do not.
Cost per round trip
This is the only number the engine charges, in dollars
per ounce. 0 matches the paper engine and every published backtest. A real
broker usually takes $0.30–0.60 in normal hours, and several dollars around news.
Overnight financing (swap)
Enter what your own broker charges — a positive number is a
cost, a negative one a credit (short positions sometimes pay you). Charged
only for nights actually held: rollover happens at 5pm New York on trading
days, so Saturday and Sunday are free, and one day counts triple because
the position's settlement date jumps the weekend. For gold and FX that day is
Wednesday.
Performance, in detail
Funded-account check
Funded-trading firms judge traders by two numbers: the most you may lose in one day, and the most you may ever be below your starting money. Set the limits — the check replays your current settings against them.
This only says whether these settings would have passed these limits over this past year of data. Every firm's rules differ, and past results never promise future ones.
Your settings vs the live rules
Trades under your settings
Your record
You vs a botsame days, same market
The bot is replayed over exactly the days your log covers,
starting from the same $10,000. It is not a competition — it is a way to see
whether your judgement added anything to a mechanical rule.
Trade log
Economic-data report card
One row per data morning: the econ score when the bots
first knew, and where gold stood 4 and 24 hours later. Months of rows —
not feelings — will decide whether the econ layer keeps its voice.
"na" means the market was closed at that mark.