RESEARCH TRANSPARENCY / V1
Every chart should have a traceable observation, a documented model assumption and an honest limit on interpretation.
Price observations originate from the whitelisted Yahoo Finance chart endpoint (unofficial), or from CoinGecko Demo where configured and permitted. Users may choose their own date,close CSV. Provider failures are displayed instead of replaced with invented data.
Observed daily closes are grouped by common UTC Sunday–Saturday periods. The final actual close of each week is used; non-trading equity days are never forward-filled. Returns are 100 × ln(close(t) / close(t−1)).
A Gaussian mean-shift Bayesian online change-point recursion models posterior run lengths using a constant hazard. Its output is a model-conditional posterior reset probability, not the probability of a verified market event.
Nine variations explore three hazard settings and three fixed observation-variance scales. Agreement of nearby posterior peaks quantifies sensitivity to these parameters. Agreement is not calibrated certainty or independently verified predictive accuracy.
Pairwise Pearson correlations use only common observed weekly periods and require at least twelve overlaps. Annualized volatility is sample weekly return standard deviation multiplied by √52. Shared shift windows show temporal proximity, not causation.
The model estimates observation variance from the full return sample and a prior mean from initial observations. This introduces hindsight and does not constitute a leakage-free online forecast. Alternative models, academic source snapshots and walk-forward benchmarks are required before scientific validation.