partial-identification

Bounding Treatment Effects by Pooling Limited Information across Observations

Cross-asset price impact is set-identified, not point-identified: the confounding gap has rank at most K + rank(B). A diagonal truth still yields a dense cross-impact matrix, and the trades immune to the resulting cost error form a large but measure-zero subspace. Preregistered and reproducible.

Causal Inference in Case-Control Studies

Reproducibility artifact for 'Sharp Target-Domain Certificates for Quantum-Kernel Advantage under Distribution Shift': sharp finite-batch certificates, controlled quantum-vs-classical kernel benchmark (EMBER, UNSW-NB15, ToN-IoT, TableShift), frozen specifications, prediction locks, manuscript sources. Zenodo v1.1.6.

[Experimental] Federated Partial Identification for Causal Inference with OMOP CDM

Stata package for difference-in-discontinuities (Picchetti–Pinto–Shinoki 2026): local polynomial estimation when a confounder sits at the same cutoff, plus two validity tests and partial-identification bounds.

Claim-relative identified sets across benchmark evaluation specifications

A unified Python framework for causal effect bounding algorithms

Implementation for the working paper: Symmetry-Informed Causal Partial Identification

Estimating the Effect of Persuasion in Stata

Archival code, results, and reproducibility bundle for common-witness counterfactual image audits; see KNOWN_ISSUES.md before rerunning.

Mémoire d'actuariat (ENSAE Paris, Institut des actuaires) : le besoin de capital lié à la non-conformité au règlement DORA, modélisé par une cascade dirigée entre ses cinq piliers. Chaque nombre du mémoire est vérifié contre la sortie du script qui le calcule.

Sharp performance bounds for early-warning evaluation with interval-uncertain event onset.

Replication files for Jun and Lee (2022)

Estimating the Effect of Persuasion in Stata

A new kind of world model: the PSD coupling kernel K(T,T') of admissible possible worlds. Diagonal = prediction, off-diagonal = counterfactual coupling. Built in public.

Reproducible code and derived data for testing how present-day stellar constraints identify cumulative EUV and wind histories in TOI-700 and LHS 1140.

What 12 million powerlifting attempts can and cannot say about attempt selection. Paper, code and a free attempt calculator.

Tight best- and worst-case fill bounds for a passive order backtested on L2 data

Abstention and bias scores in BBQ-style benchmarks: code, data and results