Active leads, controls, successors, and honest negative results remain visible.
Researching signal in a noisy market
A clearer view of tomorrow's opportunities.
AlientAI is a research-first market intelligence project. We build point-in-time datasets, test machine-learning models honestly, and study whether a small number of high-quality stock opportunities can survive costs, changing regimes, and unseen future data.
What we have built
From an idea to a disciplined research system.
The project now spans data collection, leakage-safe model development, prospective journaling, outcome measurement, and an independent safety boundary. Every experiment keeps its own identity and evidence trail.
Separate AI/semi-17 and Nasdaq-101 studies across six intraday horizons.
A searchable, provider-pure map of existing market archives and research inputs.
Research evidence is deliberately separated from execution authority.
The foundation
Four layers, each designed to fail safely.
Prediction is only one layer. Data integrity, frozen evaluation, deterministic risk controls, and append-only evidence matter just as much.
Point-in-time data library
Adjusted daily and intraday candles, options history, dated news, earnings material, market-regime context, and immutable source manifests.
Independent model laboratories
LightGBM ranking, technical and relative-strength studies, barrier probabilities, ticker-aware research, and horizon-specific experiments.
Prospective evidence journals
Selections and abstentions are frozen before entry, then matured later under unchanged costs and outcome rules.
Deterministic safety boundary
Models may rank opportunities; a separate risk layer decides whether an observation is even eligible to advance.
Early evidence board
Some models turned positive. The sample size stays attached.
These are research results after the project's frozen cost assumptions. Historical evidence and genuinely prospective runs are shown separately, with losses and limitations kept visible.
Sealed historical test20-session horizon
Autonomous transparent Nasdaq-101
+3.168%
mean net return per signal after a 0.25% cost assumption
590signals
55.59%positive
-6.075%max drawdown
55.59% positive signals · all four observable non-overlap cohorts had positive means
The test was opened once after validation passed and was not retuned. It still carries fixed-universe survivorship bias and simulated-fill limitations, so it is a research lead—not a forecast of future performance.
Prospective checkpoint
First four chronological runs
Average net return after costs
01
Technical + unusual calls
Contextual selective model
+1.030%4-run average
3/4 profitableworst run -1.544%
02
AI17 premarket
Original intraday model
+0.690%4-run average
2/4 profitableworst run -1.544%
03
AI17 unusual calls
Original intraday model
+0.492%4-run average
2/4 profitableworst run -1.488%
One run is one non-overlapping decision cohort—not one ticker. The first four were fixed chronologically; no favorable window or loss was discarded.
Kept visible
Not every early checkpoint passed. The Alpha Nasdaq QQQ-relative model's first four runs averaged -0.982%, with one profitable run and a -6.846% worst run.
Positive results are leads. Negative results remain evidence.
What we have found
Progress includes learning what does not work.
01
Selectivity looks more promising than constant activity.
Our strongest research leads tend to abstain often and wait for unusual combinations of technical, market, and options context. Early samples are encouraging in places, but they are not proof.
02
Costs and market regime can erase a headline result.
Models are judged after estimated trading costs, against market controls, and across changing conditions—not by raw accuracy or total profit alone.
03
More complexity is not automatically better.
Many sophisticated variants failed validation, concentration, drawdown, calibration, or matched-control tests. Those failures are preserved instead of hidden.
04
Confidence must be calibrated, not assumed.
A rank is not a probability. Future autonomy requires independently validated win, failure, uncertainty, and path-risk estimates before any risk gate can rely on a model.
How the research moves
Evidence before confidence.
- 01Validate dataFreshness, identity, timestamps, and completeness
- 02Build featuresOnly information available at the decision time
- 03Rank candidatesSeparate models for separate horizons and jobs
- 04Apply risk gatesAbstention is valid; cash is a position
- 05Freeze the journalNo rewriting after the future becomes visible
- 06Measure outcomesCosts, controls, drawdown, uncertainty, and regime
Where we are going
Build the evidence base. Prove the safety layer. Earn every next step.
AlientAI is not racing toward automatic trading. The path forward is to enlarge the historical library while continuing every legitimate future test, improve calibration and path-risk estimates, and run the complete decision system in shadow mode.
Expand the research library
Preserve candles, options, transcripts, news, earnings, fundamentals, and market context while premium access is available.
Accumulate prospective cohorts
Keep every valid selection, abstention, loss, and win under frozen definitions and realistic costs.
Operate in live shadow mode
Produce exactly what the system would propose without sending an order, while testing data and risk failure modes.
Review limited paper trading
A separate human decision, after enough evidence and proven safeguards. Live trading is not authorized.
Current project status
Active research. Honest uncertainty. No performance promise.
AlientAI is continuing prospective model testing and expanding its private research-data library. Some early leads are worth continued study; none has established reliable profitability.
This website describes a private research project. It is not financial advice, an investment solicitation, or a trading service. No public, paper, or live order is created through this page.