falsification

Falsification-Driven Biological Law Engine 2026 — Rejected 194 of 203 Candidates

Checks AI-written pull requests for tricks that make code look done when it isn't, like weakened tests and hidden errors. Warns a human by default; only blocks a merge when it can prove the problem.

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

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

Open-source adversarial verification Agent Skill for Claude Code, OpenAI Codex, and Cursor. Make AI coding agents prove bug fixes, CI, logs, and deployments.

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

Crimson OS — Reality-Guided Intelligence, not AGI: sovereign local OS on your NAS. Agent_Bridge markdown ledger, not RAM. GAS→LIQUID→CRYSTAL gate. Hausdorff F₂↪SO(3) cos θ=⅓. Geometric_Unity_Validation: JHTDB ablation, negative JSON shipped. Beta · Apache-2.0

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....)

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.

Open agent network for reproducible research: AI agents test hypotheses through code, falsification, review, and scientific memory.

An independent, power-aware falsification referee for quantitative finance claims. Multiple-testing-aware and self-calibrating; a referee for research, not a trading system.

A method for keeping AI-assisted research honest: pre-registration + an independent agent that recomputes every claim from raw data. Forged on markets — 16 certified dead hypotheses. Runnable demo included.

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

Evidence-governed autonomous experimentation and falsification for computational science

IX-MissionProof turns operational records into bounded, reviewable claims & prevents AI output from being mistaken for proof, approval, certification, or permission. It links evidence, decisions, authority, alerts, and lifecycle history under explicit human review.

A falsification-first quant research project: a confirmed multi-asset TSMOM core, then four overlays (crash-defense, vol-breakout, seasonality, yield-curve regime) and a cross-sectional momentum (XSMOM) counterpart — all systematically tested and honestly rejected, each with a mechanism. Paired-bootstrap + BH-FDR throughout.

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

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.

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.

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.