What We Don't Know Yet
Science begins with uncertainty.
Videira researches Cognitive Overload Reduction, Decision Intelligence, evidence synthesis, calibration, explainability and human-machine decision interaction. Openly acknowledging uncertainty strengthens trust.
Instead of claiming certainty where none exists, we measure, validate, calibrate, and continuously improve. This page documents the questions we are actively investigating.
Tomorrow's discoveries begin here.
Trust is earned by showing
We do not hide uncertainty. We publish it — because that is how trust, and intelligence, compound.
What we know
Measured, validated, and reproducible.
What we think we know
Emerging — multiple observations suggest a relationship.
What we are testing
Active experiments with defined hypotheses.
What surprised us
Findings that contradicted our assumptions.
What changed our minds
Where measurement disproved earlier beliefs.
Every topic has a state
We never force certainty. Each research question is classified into one of five honest states.
Confirmed
Supported by sufficient evidence.
Emerging
Multiple observations suggest a relationship.
Investigating
Evidence currently insufficient.
Conflicting Evidence
Results disagree.
Unknown
Not enough data exists yet.
What we are actively investigating
Each card expands into the full research record: evidence, limitations, missing data, the next experiment, and a target.
Conviction metrics shift materially across regime transitions.
Regime labels are partly subjective and retrospective.
Continuous, forward-looking regime classifications.
Track conviction stability across labeled regimes over a rolling window.
Q4 2026
Unexpected discoveries
Where the data surprised us. Each discovery records the assumption it overturned and how we plan to validate it.
Historical similarity behaved opposite to our initial expectation
More historical similarity would monotonically improve predictions.
Similarity past a threshold reduced accuracy.
Excess similarity overfits; the benefit is bounded.
Replicate across regimes and asset classes.
Breakout probability appears neutral in current data
Breakout probability would be directionally informative.
Near-neutral relationship in the current sample.
Breakouts may require richer context to inform direction.
Re-test with conditional features.
Liquidity showed stronger explanatory power than expected in ETF analysis
Liquidity would be a secondary factor.
Liquidity dominated several ETF outcomes.
Liquidity is a primary driver in ETF settings.
Extend to a broader ETF universe.
Trend alignment appears more informative in Forex than Equities
Trend alignment would behave similarly across asset classes.
Stronger in Forex, weaker in Equities.
Signal value is asset-class dependent.
Build per-asset-class signal weighting.
What we cannot yet claim
Publishing limitations is not a weakness. Each one is paired with how we are addressing it.
Certain historical variables were not captured during early platform versions.
Addressed by — Backfilling key variables where reconstructable; gaps are flagged in records.
Some market regimes require additional data.
Addressed by — Actively collecting regime-specific samples.
Outcome capture is still expanding.
Addressed by — Broadening outcome tracking across products.
Research Genome coverage continues to improve.
Addressed by — Extending the Genome to new domains incrementally.
Legacy records contain fewer analytical dimensions than current records.
Addressed by — Layering new dimensions onto legacy records where possible; incomplete records are clearly marked.
How a question becomes an improvement
A disciplined loop. Every completed experiment improves the platform.
Every completed experiment improves the platform. The loop never closes — it sharpens.
Evidence changed our thinking
We celebrate these. Changing our minds in response to measurement is the whole point — never something to hide.
More historical similarity is always better.
Benefit is bounded; excess similarity overfits.
We bound similarity depth.
Model sophistication drives calibration.
Calibration mattered more than sophistication.
We invest in calibration, not just models.
Breakout probability is directional.
Neutral in current data.
We don't over-weight breakouts.
Signals transfer evenly across asset classes.
Liquidity and trend vary by asset class.
We weight signals per domain.
Future investigations to explore
The next frontiers — with their priority, expected impact, current progress, dependencies, and the evidence we expect to need.
Adaptive calibration across asset classes
Dynamic holding horizon
Uncertainty-aware recommendations
Explainability vs prediction quality
What is running right now
A live view of active experiments, recent calibration, newest discoveries, and research completed this month.
Calibration stability across regimes
Signal family out-of-sample test
Similarity-depth error map
Confidence recalibrated for volatile windows
Liquidity dominance in ETFs
Counterfactual overconfidence finding
Disciplined uncertainty is stronger than artificial certainty.
Every unknown identified today becomes tomorrow's competitive advantage. We measure reality instead of defending assumptions — and we publish what we find, honestly.