replication

A science blogger attempts to replicate the paper 'Achieving optical transparency in live animals with absorbing molecules' by soaking chicken breast in tartrazine (the food dye in Doritos), achieving partial, gummy-bear-like transparency with a modified low-concentration protocol.

This paper audits graph neural network-based road-level crash prediction by replicating a recent uncertainty-aware model across three London boroughs, finding that most reported design effects fail to replicate or even reverse sign. The authors show that a parameter-free baseline ranking road segments by past crash counts matches or beats the GNN models at horizon-matched lookbacks, arguing reported gains are largely an artefact of short-horizon baselines and calling for horizon-matched baselines and multi-seed reporting.

This paper investigates replication, measurement sensitivity, and persistence in behavioral evaluations of hosted LLMs, demonstrating how conclusions vary across configurations and identifiers using the Regent Chess environment.

Apache Kafka introduces native cluster mirroring to embed cross-cluster replication directly into the broker, simplifying data movement between clusters with byte-for-byte transfer and exact offset preservation.

Gricea is an open-science platform for conversational AI research that enables configurable and deployable research artifacts to support replication, extension, and cumulative knowledge building.

A researcher reproduced AI benchmarks but found that verifying business claims was challenging due to metric aggregation and model integration issues, emphasizing the need for careful scrutiny of AI performance claims.

WalShadow is an open-source tool that replicates Postgres data to ClickHouse with sub-second latency by directly consuming physical WAL, enabling real-time analytics with high throughput and schema evolution support.

Astra is a tool for checking scientific papers that uncovered numerous coding errors in replication packages, some overturning central results in high-ranking journals, and revealed that many models were not run properly.

The article discusses the read-your-own-writes issue in database replicas and highlights PostgreSQL 19's new WAIT FOR LSN command as a solution to improve consistency and reduce lag.

OpenRun introduces built-in Litestream support for SQLite apps, enabling automatic replication to object storage and seamless deployment on Docker and Kubernetes with GitOps integration.

A tweet listing essential database concepts for understanding scaling, including B+ Trees, LSM Trees, and distributed systems topics.

An account of a MySQL upgrade that resulted in data inconsistency due to AUTO_INCREMENT behavior during replication, with insights on binlog formats affecting table references.

Snowflake's engineering blog details the new Postgres data mirroring feature that pushes changes via a snowflake_cdc extension into Iceberg tables, aiming for resilient, low-lag transactional replication.

A paper introducing predictive speculative KV replication to handle bursty LLM inference workloads, with code available on GitHub.

The author shares how they recreated Claude Dynamic Workflow in one afternoon and published a long Chinese review of DeerWorkflow. The project earned 125 stars on GitHub in less than 24 hours.

Matt Shumer celebrates a community replication using his prompt, where Zhuo Tao successfully ran opus 5 to one-shot a game on an M5 MacBook Pro in about 2 hours.

This preregistered replication tests whether the monotonicity effect on label agreement in NLI generalizes from selected low-agreement items to unselected populations, finding that the effect reverses and is small, suggesting the earlier finding was conditional on selection.

Drake Anthony reproduces the effect, discovered in 1675, of mercury glowing when shaken in a vacuum glass tube. By filling the tube with neon gas, he creates a mechanical light bulb with astonishing brightness, demonstrating a plasma light-emitting device that requires no external electric field.

Veritas is a general-purpose replication tool that uses CLI coding agents to automatically replicate scientific research from papers and code, extracting claims and evaluating them, achieving state-of-the-art performance on benchmarks.

Thundersnap is an experimental distributed filesystem from Tailscale that uses btrfs snapshots and Tailscale networking to provide undo-like functionality and efficient replication across computers. It is currently a toy, not production-ready.