Research evidence explorer¶
This curated research map links to existing original and independently audited documentation; filtering changes navigation only. No pooled coverage estimate or calibration judgment is recomputed in the browser.
Evidence status
General nominal-90% whole-function W3 calibration and field missing-gaze MNAR identification remain unqualified. Stable Python 1.2.0 package release status does not change these statistical limits (#289, #296).
9 of 9 entries shown
R76 · Independent leverage support
Functional-band inclusion under leverage stress; participant support and whole-function inference remain separate questions.
R78 · Temporal covariance
Generator-specific time dependence and uncertainty behavior; not global 90% coverage qualification.
R80 · Heavy-tailed independent units
Student-t5 stress design with original source plots and methodological caveats.
R82 · Retention versus support
High sample retention need not imply adequate independent-unit evidence.
R84 · Independent external validation
What separately recovered missing outcomes can identify under an explicitly defined ascertainment design.
R88 · Selective audit return
Assumed return-process sensitivity; randomized invitation alone does not guarantee outcome recovery.
R93 · Exact observational equivalence
Distinct latent populations may induce the same observed law; invitation rates are not complete identification.
R94 · Externally assumed completion floor
Conditional sharp sets given a real-world completion guarantee, not an estimated return rate.
R95 · Return direction
Monotone response-direction assumptions can narrow a conditional population identified set without revealing the true mechanism.
Full R76–R95 chronology · Source archive · Evidence vs qualification