MT5 Wasn’t Built for This: Inside NEXUS DOM Heatmap V25 and Its Local AI Copilot

MetaTrader 5 was designed to display markets. NEXUS DOM Heatmap V25 is trying to make it explain them.

Most retail trading platforms still revolve around the same primitive abstraction:

Price becomes a candle.

Then traders stack indicators on top of that candle and attempt to reconstruct what happened underneath.

Moving averages.

Oscillators.

Support.

Resistance.

Volume.

Momentum.

But by the time all of that appears on the screen, the actual battle may already be over.

Liquidity was added.

Liquidity disappeared.

Aggressive buyers crossed the spread.

Passive sellers absorbed them.

An iceberg replenished.

A wall failed.

Another wall held.

The order book changed shape.

Then price moved.

NEXUS DOM Heatmap V25 was built to investigate that hidden layer directly.

🎥 Watch the new V25 video:
https://youtu.be/VLkcgouzfJ8

🌐 More information:
https://metaquantuniverse.com/nexusdomheatmapmt5/

This Is No Longer Just a Heatmap

The original concept was straightforward:

Take genuine Depth of Market / Level 2 data inside MetaTrader 5, preserve it through time, and transform it into a usable liquidity heatmap.

But over successive versions, that idea evolved into something much larger.

V25 is now closer to a complete order-flow workstation inside MT5.

The system combines:

DOM history.
High-frequency liquidity capture.
Trade flow.
Time & Sales.
Absorption.
Sweeps.
Icebergs.
Pulling and stacking.
Liquidity memory.
Cumulative Delta.
Volume Profile.
Market-state detection.
Signal scoring.
Telemetry.
Historical review.
And now a local AI copilot.

The objective is no longer simply to display the book.

The objective is to understand what the book is doing.

The Order Book as a Film, Not a Photograph

A traditional DOM is essentially a snapshot.

At this instant, there are X contracts here.

Y contracts there.

Then everything changes.

What matters in microstructure is not only where liquidity exists.

It is:

how long it existed,
how it changed,
whether it was executed,
whether it reloaded,
whether it disappeared,
and what price did afterward.

V25 treats DOM history as a continuous time-film.

The heatmap sampling rate can now be adjusted live between 5 and 30 Hz. At 30 Hz, that means one stored heatmap column every 33.3 milliseconds. Missing book events can be carry-forward filled so the visual timeline remains continuous, while capture and rendering remain decoupled.

That distinction matters.

A screen does not necessarily need to redraw 30 complete frames every second to preserve 30 DOM states per second.

The data capture and the visual publication are separate problems.

That is one of the architectural changes that makes V25 much more interesting than a conventional MT5 indicator.

Milliseconds Matter

Order flow becomes considerably more difficult when different elements of the interface are running on different clocks.

This is exactly the kind of problem V25.1 addresses.

Earlier versions could position the heatmap using a local millisecond clock while trades, bubbles, absorption markers, candles and labels inherited whole-second server-time behavior.

At aggressive zoom levels, the result could become visible: the heatmap moved smoothly while some overlays appeared to remain frozen and then jump.

V25.1 maps server time into the local millisecond clock and uses millisecond information for executions and events so that bubbles, markers and the heatmap remain spatially synchronized. The implementation notes describe live verification at roughly 47 pixels per second, with the trade-bubble chain maintaining its distance relative to the heatmap during a burst.

This sounds like a small technical detail.

It is not.

When studying market microstructure, timing is part of the signal.

Absorption: When Aggression Stops Moving Price

Imagine aggressive sellers repeatedly hitting the BID.

You would normally expect price to continue lower.

But it does not.

Trades continue.

Volume accumulates.

The level survives.

That is where absorption becomes interesting.

NEXUS attempts to identify exactly these types of interactions.

Instead of treating traded volume in isolation, the engine can combine execution with resting liquidity, replenishment behavior, market context and Data Quality.

The same logic applies in reverse.

Aggressive buyers repeatedly lift the ASK, yet price stops advancing.

That may indicate passive selling interest absorbing the attack.

The key concept is simple:

Aggression alone does not determine the result.

The interaction between aggression and passive liquidity does.

Icebergs: When the Visible Size Is Not the Real Size

One of the more interesting additions in the NEXUS intelligence layer is the attempt to distinguish displayed liquidity from executed liquidity.

Suppose a level displays 40 contracts.

Then 70 contracts execute.

The level remains.

Another 30 execute.

The displayed quantity replenishes again.

What appeared to be a small wall may actually represent considerably more liquidity than the visible book initially suggested.

V25 contains explicit iceberg logic using reload counts and an executed-volume-to-displayed-volume relationship. The default configuration shown in the source requires multiple reloads and compares executed flow with the maximum displayed quantity.

The earlier “Answers” layer also describes an Iceberg X-Ray concept: executed volume is attributed to the resting level that absorbed it, then compared with the largest displayed quantity to estimate whether hidden size may be present.

This is fundamentally different from looking at a static DOM.

The question becomes:

How much business actually happened behind that number?

Spoofing-Like Behavior Without Pretending Every Cancellation Is Manipulation

Spoofing is one of the most abused words in retail order-flow trading.

A large order disappears.

“SPoOfInG!”

That is not enough.

Modern order books naturally contain huge quantities of cancellations because market-making systems, execution algorithms and other automated participants continuously update quotes.

NEXUS therefore approaches this as spoofing-like behavior, not instant legal certainty.

The intelligence engine includes configurable conditions involving peak displayed volume, repeated pull cycles, minimum display time and a low execution ratio.

The interesting pattern is not simply:

order appears → order disappears.

It is something closer to:

Large liquidity appears.
It persists long enough to matter.
The visible structure changes.
Little of it actually executes.
It gets pulled.
The sequence repeats.

That becomes a behavioral signature worth investigating.

Sweeps: When Aggression Crosses Multiple Levels

Not every aggressive execution is meaningful.

One tiny market order crossing the spread is noise.

A genuine sweep is different.

The V25 intelligence layer contains parameters for a very short sweep window, minimum executed volume, minimum trade count and minimum number of price levels crossed.

That means the engine can reason about execution sequences, rather than isolated prints.

This becomes particularly useful in fast futures markets.

A burst of aggressive flow traveling through several price levels in 100-200 milliseconds is fundamentally different from scattered individual transactions occurring over several seconds.

Same total volume.

Completely different microstructure.

Pulling and Stacking

Liquidity is not important only when it trades.

Sometimes what disappears is more important than what executes.

Imagine a large BID appearing below the market.

It looks supportive.

Price begins approaching it.

Then the BID vanishes.

Nothing “hit” the wall.

The wall simply stopped existing.

The short-term liquidity landscape has changed.

That is liquidity pulling.

Stacking is the reverse.

New liquidity accumulates at one or several nearby levels and begins altering the visible structure of the order book.

NEXUS keeps these transitions as explicit components of its liquidity model and can also preserve remembered liquidity levels over time. The V18 liquidity-memory system assigns different weights to absorbed, pulled, swept and stacked liquidity and can persist those levels across sessions.

This opens up a much more interesting question than:

“Where is the biggest order?”

The better question is:

How has this liquidity behaved through time?

Liquidity Memory

Markets have memory even when conventional charts do not.

A level that absorbed enormous aggression five minutes ago may still matter.

A location where liquidity repeatedly vanished may matter.

A zone where a sweep failed may matter.

NEXUS V25 maintains a dedicated Liquidity Memory.

The implementation can remember where absorption, pulling, sweeps and stacking occurred, apply a half-life to those observations, persist them across restarts and keep only the most relevant levels visible.

That transforms the heatmap from a pure visualization into something closer to a behavioral memory of the book.

Cumulative Delta and Divergence

NEXUS also tracks session Cumulative Delta from tape activity.

But raw CVD is only the beginning.

The engine includes a divergence detector comparing price displacement with delta displacement over a configurable observation window.

For example:

Price pushes higher.

Aggressive buying should presumably dominate.

Yet CVD fails to confirm.

Or price holds near the lows while aggressive selling becomes increasingly negative.

These discrepancies can reveal interesting areas where aggression and price response stop agreeing.

Again, this is not a magical BUY/SELL button.

It is another piece of context.

Volume Profile Inside the Microscope

V25 also integrates Volume Profile directly inside the microstructure environment.

The profile can operate on the visible window, the session or a rolling time range and includes POC, VAH and VAL representation.

That means execution distribution and order-book behavior can be studied together.

You can investigate:

Where did business actually occur?

and simultaneously:

How did liquidity behave around those prices?

That intersection is considerably more useful than treating market profile and DOM as unrelated tools.

Walls That Actually Have Statistics

One of my favorite ideas inside the V22/V25 lineage is extremely simple:

Do not call something “support” merely because it looks large.

Measure whether it actually held.

The system records wall outcomes by side and refill bucket.

Importantly, walls that price never reaches are excluded rather than being falsely counted as successful defenses.

The engine can report results such as:

“held 68% (n=112)”

but withholds the rate until enough samples exist.

That last part is important.

Without the sample count, a percentage is decoration.

With the sample count, it begins to become evidence.

Data Quality Is Part of the Signal

Another important architectural choice is that NEXUS does not pretend all market feeds are equally trustworthy.

Signal confidence can be quality adjusted.

The V13.1 layer explicitly allows event confidence to be discounted when Data Quality deteriorates, with a configurable minimum multiplier.

This matters enormously.

If your broker publishes incomplete, stale or thin DOM information, the software should not behave as though it is looking at pristine institutional data.

Bad data should reduce confidence.

Not create false precision.

Signals Without Indicator Soup

The result of all these layers is a very different signal architecture.

Instead of:

RSI crossed 70 → SELL

NEXUS can reason around:

execution,
resting liquidity,
refills,
sweeps,
pull cycles,
display lifetime,
iceberg ratios,
Data Quality,
market state,
CVD behavior,
and remembered liquidity.

Composable alert rules can also impose score and Data Quality thresholds before certain events generate alerts. The included defaults, for example, contain separate rules for sweeps and BID/ASK absorption.

The goal is not more signals.

It is fewer meaningless ones.

A Heatmap That Actually Looks Like Liquidity Terrain

V25 also changes the visual representation of depth.

Earlier 2.5D prism-style relief required sufficiently wide columns to look correct.

But high-frequency DOM sampling produces extremely narrow columns.

So instead of forcing fake geometry onto sub-pixel data, V25 introduces hillshading.

Liquidity becomes altitude.

Real liquidity transitions become illuminated ridges.

Strong structures gain specular highlights and cast visual shading underneath.

The effect is computed during the bake process rather than continuously rebuilding expensive geometry.

It is not merely aesthetic.

The goal is to let the eye recognize liquidity topology faster.

V25 Was Also About Speed

None of this matters if MetaTrader becomes unusable.

The V25 notes document live profiling on MT5 build 6140 with an AMD Ryzen 7 7735HS.

Several expensive object-management paths were identified, the bake queue was resized for high-frequency capture, unnecessary invalidations were removed and chart overlay behavior was optimized.

In that measured environment, thread load reportedly fell from approximately 97% to around 20-40%, while frame cost dropped from roughly 50-60 ms to 10-15 ms.

Those are measurements from that specific test environment, not universal performance guarantees.

But they illustrate the engineering direction clearly:

high-frequency capture without turning MT5 into a space heater.

And Then There Is the AI

This is where V25 takes a particularly unusual direction.

NEXUS DOM Heatmap now connects to NEXUS LLM Bridge V4.

Its default AI architecture is local-first.

The bridge is configured to run LiquidAI LFM2.5-1.2B through Ollama directly on the local computer.

No API key is required for the local provider, and the default AUTO routing attempts the local engine first, while Groq and Cerebras remain available as cloud fallbacks.

The bridge configuration uses a local Ollama endpoint at:

127.0.0.1:11434

and exposes model, thread, temperature, timeout and token-budget settings.

But this is where an important design decision appears.

The LLM Does Not Analyze the Market From Scratch

Throwing raw order-book numbers at a small language model and asking:

“What is happening?”

is a bad architecture.

The V25 development notes explicitly mention this.

During testing, the 1.2B model could phrase answers well but interpret raw order-flow numbers badly.

So the architecture was changed.

NEXUS analyzes first.

It builds a structured FACTS PACK containing information such as:

key level,
book behavior,
refills,
iceberg actor,
tape delta,
events,
defended levels,
invalidation,
and Data Quality.

The LLM then receives that structured interpretation and turns it into natural language.

That leads to one of the cleanest descriptions of the whole system:

The model writes. NEXUS analyzes.

That separation matters.

The deterministic engine handles microstructure.

The language model handles language.

Why Local AI Is Interesting Here

Running the model locally introduces several advantages.

Latency becomes predictable.

There is no mandatory cloud API dependency.

The system can operate without sending the prompt to an external provider when the local engine is selected.

And because the model is relatively small, it can coexist with MT5 on ordinary hardware rather than requiring a data-center GPU.

The bridge even defaults to four local inference threads because the development notes found that configuration more efficient beside MT5 than using eight threads for prompt processing.

This is less about building an omniscient AI trader.

It is about giving the workstation a natural-language interface to its own deterministic intelligence.

That is a much more realistic use of an LLM.

Ask the Market a Question

This creates an interesting interaction model.

Instead of reading dozens of raw measurements manually, the trader can eventually ask questions such as:

What is happening at this price?

Is this BID actually defended?

Is the ASK absorbing buyers?

Where is the strongest hidden liquidity?

What changed during the last 30 seconds?

Why did NEXUS flag this level?

The underlying answer does not need to be invented by the LLM.

The deterministic system already has the observations.

The AI becomes an interpreter.

Instant Replay for the Human Brain

Another elegant feature inherited into the current architecture is the Instant Replay concept.

The R key can summarize what happened during the recent seconds of buffered order-flow activity rather than forcing the trader to manually rewind and inspect everything. The default configuration shown in the source uses a 30-second rewind window.

This addresses a very real problem.

High-frequency order flow can simply move faster than human perception.

Sometimes you know something important happened.

You just do not know exactly what.

Replay makes the machine answer:

What did I just miss?

This Still Depends on Real Data

There is an important limitation that no marketing copy should hide.

A DOM heatmap can only visualize the depth information actually received.

If your broker or data provider gives MT5 genuine Depth of Market data, NEXUS can capture and analyze that structure.

If the feed does not contain real depth, no algorithm can magically reconstruct orders that were never transmitted.

This is why NEXUS is especially relevant on markets and brokers where usable Level 2 / Market Depth data is available.

The quality of the tool cannot exceed the quality of the feed.

And V25 explicitly attempts to acknowledge that through Data Quality scoring rather than pretending otherwise.

MetaTrader 5, But Used Differently

That is perhaps what makes NEXUS DOM Heatmap V25 most interesting.

MT5 was never primarily designed as a high-frequency market-microstructure laboratory.

Yet V25 now layers onto it:

high-frequency DOM capture,
historical depth,
liquidity heatmaps,
millisecond Time & Sales,
absorption,
sweeps,
iceberg logic,
spoofing-like detection,
liquidity memory,
Volume Profile,
CVD,
wall statistics,
market-state intelligence,
signal scoring,
replay,
telemetry,
and a local AI copilot.

The result barely feels like a conventional indicator anymore.

It is becoming a workstation.

Final Thought

Candles are not useless.

They are simply compressed.

They reduce thousands of executions, additions, cancellations, refills and algorithmic decisions into four numbers:

Open.
High.
Low.
Close.

NEXUS DOM Heatmap is built around the opposite philosophy.

Expand the market again.

See the liquidity.

See the execution.

See the aggression.

See the absorption.

See what stays.

See what disappears.

See what actually trades.

Then let the machine organize that information without pretending certainty where none exists.

And now, with V25:

let the deterministic engine analyze it, and let a local AI explain it.

🎥 Watch NEXUS DOM Heatmap V25 in action:
https://youtu.be/VLkcgouzfJ8

🌐 Discover NEXUS DOM Heatmap for MT5:
https://metaquantuniverse.com/nexusdomheatmapmt5/

Price shows the result.

NEXUS shows the battle underneath. ⚡🧠