Tag
reproducibility
- Aug 27, 2026 Execution Counts and the Hidden State of a Notebook Kernel Why the cells stored in a notebook file were not necessarily executed in the order the file shows, what the execution counter measures, and which parts of the state never reach the document. scientific computing
- Jul 10, 2026 Sources of Nondeterminism in Build Output Timestamps, absolute paths, directory iteration order, and parallelism leak into object files and archives. How each is detected by repeated builds and removed at the source. devops
- Dec 29, 2025 Content-Addressed Cache Keys for Incremental Asset Builds Why a timestamp is the wrong invalidation signal, what belongs in a hash that keys an asset build, and how a stale cache entry survives a settings change. computer graphics
- Dec 22, 2025 Reproducing Training Runs from Recorded Metadata What a training run must record so that a second execution yields the same weights, and where nondeterminism survives even a perfect record. machine learning
- Jul 19, 2025 Hyperplane Structure in Linear Congruential Generators Why successive outputs of a congruential generator lie on a small number of lattice planes, how the spectral test scores that defect, and what state size buys in a modern generator. scientific computing
- Jun 12, 2025 Fused Multiply-Add, Single Rounding, and the Cost of Contraction What a fused multiply-add computes internally, why one rounding instead of two halves the error bound, and how compiler contraction makes identical source produce different bits. scientific computing
- Jan 27, 2025 Executing Provenance Records to Certify That a Result Still Reconstructs Turning a recorded dependency graph into a re-runnable pipeline that hashes outputs and compares them to stored digests, so a result carries continuous evidence that it is still reproducible. scientific computing
- Dec 13, 2024 Feature Definitions as Versioned Artifacts Why a feature needs a name, a query, an owner, and a version, and how change control on feature definitions prevents silently retrained models. machine learning
- Jul 23, 2024 Model Artifact Identity: Why a Checkpoint Hash Is Not a Deployment The behaviour a serving fleet exhibits depends on far more than tensor values, so deployment identity has to cover the tokenizer, the preprocessing path, and the runtime configuration as one unit. inference systems
- Jul 11, 2024 Reassociation, Excess Precision, and the Limits of Bitwise Reproducibility Why identical source can produce different bits on two machines, which compiler transformations are responsible, and how ordering or exactness restores a stable result. scientific computing
- May 29, 2024 Rebuilding a Pinned Artifact Without Its Original Environment Keeping the recipe instead of the bytes only works if the toolchain, dependencies, and clock can be reconstructed. What a pinned environment has to name and what cannot be rebuilt at all. devops
- Apr 22, 2024 Environment Parity as a Byte-Level Claim: What Divergence Between Stages Actually Costs Why staging and production diverge even when both are built from the same configuration, which differences change behaviour, and how to make parity a measurable property. devops
- Feb 3, 2024 Lockfiles as Reproducibility Contracts and Why They Drift What a lockfile must record to make an install repeatable, which inputs it cannot capture, and how lockfiles fall out of sync with manifests without either side being wrong. language runtimes
- Apr 7, 2022 Reproducibility, Replicability, and the Boundaries Between Them The distinct claims implied by re-running an analysis, rebuilding it from raw data on new instruments, and getting the same conclusion from an independent study, and which evidence each one requires. scientific computing