Popular repositories Loading

1. Sruthi0510-data Sruthi0510-data Public

2. merge-order-sensitivity merge-order-sensitivity Public

   Proves which keys in a CDC merge have an order-dependent final state, and which the delivery model can actually reach. 8.0% of keys are order dependent and 3.5% only across batch boundaries, where …

   Python

3. rag-failure-attribution rag-failure-attribution Public

   Attributes RAG retrieval failures to what actually caused them: the ones a better retriever can fix against the ones only re-chunking can. Finds the questions your recall metric scores as wins whil…

   Python

4. dq-detectability-audit dq-detectability-audit Public

   Audits a data quality suite for rules that are statistically incapable of catching the defect they were written to catch, and for the damage a block-sampling clause does to the ones that could. 15 …

   Python

5. llm-label-churn llm-label-churn Public

   Shows what an LLM prompt change actually did, which an accuracy delta cannot. Accuracy is up 1.3 points and the paired test cannot distinguish it from noise; 2,248 labels moved for a net gain of 13…

   Python

6. stream-replay-divergence stream-replay-divergence Public

   Measures what a replay of a stateful streaming job cannot reproduce. 62% of the output rows survive reprocessing, the session table goes from 205 rows to 12, and the bound you can compute from the …

   Python