Reviews pull requests for the shortcuts AI coding agents take to look done without being done: relaxed tests, swallowed errors, fake renames, 11 checks in all. Flags them for a human by default, or blocks the merge if you turn that on. Can also turn a goal into a checklist and only accept a patch once every check passes.

causal-falsify: A Python library with algorithms for falsifying the unconfoundedness assumption in a composite dataset from multiple sources.

Pressure-test research claims with falsifiable evidence plans, adversarial checks, frozen verifiers, and proof ledgers.

Code to reproduce the experiments from the paper "Self-Compatibility: Evaluating Causal Discovery without Ground Truth"

The Why! World Health Year 2025. Let’s join forces and realize that World War III is already ongoing and that the Lie is the main weapon. Let’s create processes and tools to fix this. Based on falsifiying to exclude non-truths rather than as Elon suggested to make a truth-telling tool (very difficult Mr Musk, you should know this....)

Finding Property Violations through Network Falsification: Challenges, Adaptations and Lessons Learned from OpenPilot

A multi‑country synthetic identity generator that produces full, internally consistent life profiles (personal data, family, employment history, historical context, etc.) designed for OPSEC, security research, and testing scenarios, never for impersonation of real individuals or any unlawful use.

Public-safe falsifier lab for observer-aware AI research: synthetic demos, evidence gates, negative controls and witness logs.

Calibration and falsification toolkit for AI claims: null models, evidence grades, claim boundaries and observer-aware workflow examples.

Evidenced-negative taste object — failed the geometry-bearing fit gate (metric_fit 0.207 vs 0.6 threshold). The committed negative reference is preserved as the lab record. No positive taste codec is claimed.

PressureX is an engineering evaluation package for a passive layered structural mitigation concept using shear-thickening fluid behavior to broaden impulsive loads and reduce peak transmitted response in high-vibration aerospace structures. Targets are design-intent until validated.

MIT synthetic simulation sandbox for observable runs, falsifier examples and public-safe research boundary demos.

Source-available IX-Function causal-transfer harness: tests cross-domain causal structure reuse through pre-outcome prediction, reality-delta scoring, uncertainty preservation, falsification, negative controls, model review, donor handoffs, and bounded Wave 6 review gates. No AGI claim.

Falsification Chain of Thought — post-hoc verification of AI judgments using falsificationism to counter sycophancy bias

Candidate constant auditing across observation contexts. Measures constant / drift / rupture boundaries for OMNIA structural measurement.

Certified log-concavity of the Riemann-Jacobi kernel with Arb/FLINT ball-arithmetic certificates. Source-critical audit of the Polya-type real-zero criterion. No unconditional RH claim.

Validation, falsification, artifact traceability, reproducibility, and result-regression layer for OMNIA structural measurement. Evidence, limits, and failures; not a truth oracle.

Indus-script anchor application in the Zer0pa Gnosis Portfolio: clean-room search-without-decode runtime + Phase 4 conditional catalogue at k=70 + Phase 5 non-decipherment posture. No decipherment claimed. Useful now, improving continuously.

Calibrated falsification harness for retrieval & ranking. Four-null gate (incl. gold-marginal-matched random, novel) + SHA-256/git-commit integrity lock. Catches predictors that look right but aren't.

Perception receipts for AI video pipelines. Cross-writer bit-exact under default settings (SHA-256 stable across writers in any language). Zero runtime dependencies; pure stdlib core. ~1.1 KB per video; per-frame CRC32 + schema + versioning. Useful now, improving continuously.