NyalaiContact
The calibration engine for probabilistic systems in quantitative finance

The problem is no longer producing strategies.
It is knowing which deserve to be trusted.

Know, don't believe.

Nyalai returns binary verdicts on probabilistic systems in quantitative finance. Every candidate is submitted to five adversarial validation gates. Every verdict is signed, timestamped, and publicly citable. The methodology is published in full under Creative Commons Zero. The signature is not.

The industry we operate in

The reproducibility crisis, in four numbers.

Quantitative finance produces published strategies at industrial scale. Very few of them survive an honest out-of-sample test. Nyalai exists because the gap between what is published and what is real has become large enough to be its own asset class.

0
Published equity risk factors
Documented in top-tier finance journals as of 2016. The majority are estimated to be statistical artefacts of trial selection, not real economic effects.
Harvey, Liu & ZhuAnd the cross-section of expected returns · Review of Financial Studies · 2016
> 0%
Probability of backtest overfitting
Estimated rate at which the best-performing in-sample strategy fails to rank among the top strategies out-of-sample. Standard result on published quant portfolios.
Bailey & López de PradoThe deflated Sharpe ratio · Journal of Portfolio Management · 2014
~$0T
Global systematic AUM
Assets under management by systematic hedge funds, quantitative CTAs, and rules-based ETFs. Every allocation decision inside this figure depends on an implicit claim of predictor calibration.
PreqinSystematic and quantitative hedge funds tracker · 2025
12
Independent binary validators
Number of institutional infrastructures that render a public CALIBRATED or NOT-CALIBRATED verdict on a probabilistic financial system with a reproducible audit trail and CC0 methodology. Until Nyalai.
NyalaiRefusal Doctrine v0.3 · 2026
What a Nyalai verdict looks like

Two verdicts, one page.
Pedagogy through contrast.

A refused system (our own meta-calibration engine, Synapse v0.3) side by side with an approved system (a reference case). The rust wax seal signals refusal. The green wax seal signals approval. A reader understands the mechanism in five seconds.

nyalai.com/verdicts/001-synapse-v03
Verdict-form 001
NOT-
CALIBRATED
Synapse v0.3
meta-calibration engine
Ledger N
239 predictions
Evaluated
2026-07-01
Doctrine
v0.3
I
Statistical fragilityBSS = −0.079
FAIL
II
Cross-regime robustness4 of 4 negative
FAIL
III
Guided-search safetyp = 0.773
FAIL
IV
Probabilistic calibrationECE = 0.141
FAIL
V
Bootstrap confidenceCI = [−.147, −.025]
FAIL
nyalai.com/verdicts/reference-R-01
Reference case R-01
CALIBRATED
Reference Predictor R-01
illustrative demonstration
Ledger N
512 predictions
Evaluated
2026-07-06
Doctrine
v0.3
I
Statistical fragilityBSS = 0.087
APPROVED
II
Cross-regime robustness4 of 4 positive
APPROVED
III
Guided-search safetyp = 0.008
APPROVED
IV
Probabilistic calibrationECE = 0.042
APPROVED
V
Bootstrap confidenceCI = [.021, .152]
APPROVED
···
Verdict-form 002 · Pending
Reference Predictor R-02 · under active adversarial evaluation
Ledger of 481 resolved predictions received on 2026-07-05. Gate III (guided-search safety) currently in progress. Verdict expected within 72 hours.
PENDING · IN REVIEW
Deployment vocabulary

Model Trust Levels.
From MTL-1 Untested to MTL-5 Live.

A shared vocabulary for how much trust to place in a probabilistic prediction system. Numbered ascending. Each level defined by empirical passage of specific gates. Each level attached to a publicly declared deployment context. Strict nesting. Binary per level. Publicly contestable. Time-bounded.

MTL1
Untested
No independent validation performed. No public methodology. No track record.
Pre-production research · Not for capital allocation
MTL2
Nominally Calibrated
Passes Brier Skill Score and Expected Calibration Error on a documented ledger of at least 200 resolved observations.
Advisory signal · Never sole allocation signal
MTL3
Robustly Calibrated
Passes MTL-2 plus cross-regime robustness under a Hidden Markov Model. Failure modes documented publicly.
Multi-signal portfolio · Measured exposure with limits
MTL4
Adversarially Validated
Passes MTL-3 plus guided-search safety (Aronson & Masters 2013) and bootstrap confidence. Sovereignty rule enforced.
Primary allocation signal · Institutional scale
MTL5
Continuously Live-Validated
Passes MTL-4 plus rolling live verdict updates and automatic drift detection. Full audit trail to LPs on request.
Mission-critical live trading · Scaled deployment
Where verdicts live

The Verdicts newsroom.

Every verdict is a long-form artifact. Signed. Versioned. Archived. Publication cadence: bi-weekly starting September 2026. Doctrine documents, method notes, and case studies live alongside.

NOT-CALIBRATED
VerdictJul 1, 2026

Synapse v0.3: NOT-CALIBRATED across all five gates

A meta-calibration engine developed by the same author is submitted to CalibrationJudge on a ledger of 239 resolved predictions. Every gate fails at published thresholds. The first Nyalai Verdict-form is a refusal of our own system. That is the intended outcome of an author who takes the sovereignty rule seriously.