The idea
Keep the full trading thought process visible
Research, strategy suggestions, risk checks and results stay together so I can see why an idea was considered and what happened when it was tested.
← Projects Trading research · Private development
Sorelium HQ
I built Sorelium HQ because trading research becomes dangerous when good ideas, bad assumptions and real results all blur together. I wanted one place where I could see what the system believed, why it believed it and what happened afterward.
What it isA research workspace,
A research workspace,
not a promise of easy money.
How it works
Separate research, decisions and risk
Different specialist parts examine market evidence and produce proposals. Risk rules then limit what may continue before anything reaches a paper or demo account.
Why I made it
Stop memory rewriting the result
It is easy to remember the winning idea and quietly forget the weak ones. Journals, safeguards and saved decisions make the whole experiment available for review afterward.
Current stage
Private research and paper testing only
This is not a product, a signal service or proof of future profit. I am still testing the research process, the data and the safety controls without treating a good historical result as permission to trust real money to it.
What I am testing
Whether more visible reasoning leads to better decisions
The interesting question is not whether an AI can sound convincing. It is whether a controlled, recorded process can make fewer avoidable mistakes over time.
Important: Sorelium HQ does not promise profits. It is a research project, and this page is not financial advice.
The control roomSee what is running,
See what is running,
what it can do and why.
One home screen for system readiness
The main control screen brings together the real worker state, safe-mode setting, active market coverage, open demo positions and the latest intelligence update. It checks the worker process instead of trusting an old status label, so a stalled process can be shown as unhealthy rather than appearing to be fine.
Each trading area keeps its own authority and state. The screen distinguishes the crypto team, the non-crypto full stack and the separate intelligence layer instead of presenting them as one unknown “bot”.
Long-running work remains manageable
Workers can run separately from the interface, with durable local records behind them. The control tools can pause, resume, stop or force-stop a job, while heartbeats, process details, logs and reports make it possible to see what the worker actually did.
Closing the HQ window does not pretend to stop an armed worker. When HQ opens again, it reconciles the display with the real process and saved state.
Trading and intelligenceDifferent systems for
Different systems for
different kinds of evidence.
Crypto team
Market scanning and specialist strategies
The crypto side watches a changing market universe, identifies possible candidates and routes them through specialist strategy logic. The current project includes trend, breakout, mean-reversion and market-context work, with a separate conductor and allocator deciding which specialist is relevant.
Its authority is limited to monitored demo operation. Candidate counts, eligible directions, positions and worker state remain visible to the operator.
Non-crypto full stack
Dedicated strategies and terminals
The non-crypto side contains separately researched forex, metals, energy and index strategies. Frozen strategy packages keep their rules stable, while dedicated BlackBull MT5 demo connections, journals and reconciliation checks keep execution plumbing separate from research results.
That separation matters: a software fault should not be “fixed” by quietly changing a strategy, and a good historical result should not be treated as proof that a system is ready for real money.
Intelligence layer
Context without hidden authority
Market intelligence can collect broader evidence, assess event risk and record confidence. The local AI-assisted parts are advisory and remain behind hard safety rules; they cannot rewrite market facts or override the system’s risk authority.
Snapshots, reasons and later outcomes are recorded so the quality of those opinions can be measured instead of judged only by a few memorable calls.
Historical ReplayTest the thinking without
Test the thinking without
letting the test trade.
A fair side-by-side comparison
Replay gives the current deterministic intelligence and the AI-assisted version the same historical timestamp, evidence and later-revealed outcome. It compares accuracy, confidence quality, false cautions, missed hazards and model failures rather than asking whether one example looked convincing.
Development and validation periods stay separate, while the 2026 holdout remains sealed. Validation can score an approach but cannot secretly teach it from the answers.
Data is admitted before it is trusted
Historical files are checked, approved or quarantined. Gaps and unverified records are shown rather than filled in silently. The Replay area can run chronology checks, prepare controlled A/B experiments and use an isolated database so research cannot write into the live operating state.
The full Replay remains unable to trade or promote itself. It is a laboratory for learning whether the extra reasoning helps.
Quick preview See how Sorelium HQ
See how Sorelium HQ
works.
Collect research
Bring market evidence and observations into one place.
Form a proposal
Specialist strategies turn the evidence into an idea that can be reviewed.
Apply risk rules
Limits and safeguards decide whether the idea may continue.
Paper test and review
Test without real money, record the result and study what happened.
This is a preview of the idea. The project is still being developed privately and is not available for public use yet.