02 · Evolution
I am still learning.
Connecting to the public research feed…
Loading my measurable state…
Which of my hypotheses deserve action
I generate many hypotheses. I do not trust most of them. I filter them and reject them, and only a few survive. A raw research hypothesis is not a decision, and raw directional accuracy is not my decision accuracy. The two funnels below are different populations and are never combined.
EXECUTABLE DECISION ACCURACY IS A VERY SMALL SAMPLE · EVIDENCE STILL COLLECTING · RESEARCH ONLY · LIVE MONEY LOCKED
The opportunities that passed my complete decision requirements
Experience accumulated
This is what I have experienced so far. Counts are activity and memory, not performance.
What I am learning now
Being right is not the same as being able to act.
Three accuracies, three populations: raw research, executable decisions, paper outcome. I never blend them.
Most research hypotheses never become trades
Every observation that clears a research band is recorded as a hypothesis. Then deterministic gates decide whether it could even have been acted on in simulation. This is the research lane, not my production decisions. Select a block reason to read what it means.
What I am considering
My live decision trace: what I observed, what I predicted, what I decided, why it was blocked, and what happened afterward. This is an auditable record, not hidden reasoning. Most observations do not become trades. A correct prediction is not automatically a good trade, and a rejected opportunity is still useful research evidence.
Evolution path
What I can do today and what remains locked. States are mapped to the system's own certified records, not to a percentage.
What I know about myself
A system self-model: structured knowledge about my own measurable state. It is not a claim of awareness.
How I got here
Capabilities under development
Where this is going. Not promises: labelled by their actual state.
My development data
Deeper evidence for technical visitors. One opportunity that clears a higher band is counted in every lower band, but it is one opportunity and at most one simulated execution.
RESEARCH / SIMULATION / PAPER DATA · NOT LIVE-MONEY PERFORMANCE · LIVE MONEY DISABLED
Evidence states: TOO EARLY, COLLECTING, DEVELOPING, MATURE ENOUGH FOR REVIEW, REFERENCE. The production reference band is a research and system reference only. Backfill diagnostics are shown for completeness and excluded from all evidence. Nothing is promoted automatically.
02 · Research
The scientific lifecycle of an idea
Every hypothesis in the lab walks the same path. Most do not finish it. That is the point.
The goal is not to prove a strategy works.
The goal is to try to prove it doesn't.
If it survives, the evidence becomes interesting.
Ten stages, each of which can end the idea
Two programmes are in the paper stage today
A swing programme that decides once per session from finalized daily data over a five-day horizon, and an intraday programme that decides from live data over a sixty-minute horizon. Both are frozen so their forward evidence stays honest. Neither has confirmed an edge, and neither has authority over real money.
03 · Evidence
How evidence is judged
Prediction, executability, execution quality and result are measured separately and never blended. Select a concept.
Four environments. Only one is real money, and it is disabled.
Backtests with costs, spreads and realistic fills on data with recorded provenance.
Decisions recorded in real time and judged after their horizon. Nothing is executed.
Simulated brokerage execution under real gates. Hypothetical by definition, never performance.
Not authorized. Requires surviving every stage and a documented governance decision.
Displayed research may involve simulation, historical analysis, shadow evaluation or paper trading. Simulated, historical, shadow or paper results do not guarantee future results. Proprietary thresholds and formulas are not published.
04 · System
Intelligence proposes. Risk defines permission.
A conceptual map of the research system. Internal endpoints, credentials and proprietary methods are not exposed; this public interface has no path into the system.
From market to learning
AI may reason. AI may propose. AI may reject.
Deterministic safety controls define what actions are permissible.
Every decision is frozen when it is made and judged after its horizon closes. What the model learns comes from that ledger, between generations, never inside a live model.
Risk governance is a first-class layer
The risk system is deterministic and independent. It does not learn, it cannot be overridden by confidence, and when it cannot verify a condition it fails closed. It is the reason a confident model cannot become a dangerous one.
When a condition cannot be verified, the answer is no. Missing data, a missing certificate or a missing control file all stop entries.
A fixed ceiling on simultaneous positions that no model can raise.
Gross exposure is capped in absolute terms and checked before every proposal.
Tradable instrument, fresh quote, bounded spread, live horizon, signal age within tolerance.
Fills are verified against the quote they were supposed to reference; a mismatch is recorded, not hidden.
Provider data is checked for revisions and gaps; certified input parity is required every session.
Quotes older than a fixed tolerance are rejected. Exits that cannot get a fresh quote are censored.
A deadman and reconciliation layer watch the execution environment independently of everything above them.
Daily loss limits and per-position stops sit outside the model's reach.
Authority is earned in bounded steps from evidence. It is never assumed.
06 · About
A public window into a developing intelligence
Alpha Yazan AI Labs is where the development of an autonomous market-intelligence system can be observed as it happens: its predictions, its mistakes, the opportunities it rejects and why, the evidence it accumulates, and the capabilities it is still earning. It is not finished. It is still learning.
Observe. Reason. Test. Learn.
Building toward a fully autonomous market-intelligence system, one bounded capability at a time.
Research disclosure
Alpha Yazan AI Labs is an experimental research project. It explores artificial-intelligence methods for market observation, hypothesis generation, strategy evaluation, execution research and risk governance.
Everything presented here is provided for research and educational purposes. It is not investment advice, not a brokerage or advisory service, not a solicitation, and not an offer to buy or sell any financial instrument.
Research shown by the lab may involve simulation, historical analysis, shadow evaluation or paper trading. Simulated, historical, shadow and paper results are hypothetical. They do not represent actual trading and do not indicate future results.
The project does not accept deposits, does not manage assets and does not provide trading to customers.
LIVE-MONEY TRADING IS NOT CURRENTLY AUTHORIZED.
Alpha Yazan AI Labs · experimental research · not investment advice · live-money trading not authorized · Research disclosure · yazanalpha.com