https://matrixcognition.com/ An AI engineering site: hands-on tutorials, paper readings with the arXiv id, model and tooling news with sources, and evaluation code you can run. 2026-09-11T00:00:00Z https://matrixcognition.com/posts/golden-test-set-chatbot.html https://matrixcognition.com/posts/golden-test-set-chatbot.html How to build a golden set for a support bot and score it with exact match, token F1, ROUGE-L, and per-item assertions. Two bots, 48 items, code and data included. 2026-09-11T00:00:00Z https://matrixcognition.com/posts/agent-tool-use-loop.html https://matrixcognition.com/posts/agent-tool-use-loop.html A tool-use loop built as an explicit state machine with six stop conditions and a token budget, measured over 2,000 simulated episodes. Code included. 2026-09-08T00:00:00Z https://matrixcognition.com/posts/structured-output-json-schema.html https://matrixcognition.com/posts/structured-output-json-schema.html A validate-and-retry loop for model JSON, with a small schema validator, a tolerant extractor, and measured retry rates from a 2,000-run simulation. 2026-09-06T00:00:00Z https://matrixcognition.com/posts/minimal-retrieval-pipeline.html https://matrixcognition.com/posts/minimal-retrieval-pipeline.html A complete retrieval pipeline in the standard library, run on 216 arXiv abstracts, with Recall@k and MRR measured for whole-document versus chunked indexing. 2026-09-06T00:00:00Z https://matrixcognition.com/posts/kv-cache-arithmetic.html https://matrixcognition.com/posts/kv-cache-arithmetic.html Bytes of KV cache per token from a config.json, worked for Qwen3-8B and Qwen3-32B at four context lengths on an 80 GB device, with GQA and int8 KV as the two levers. 2026-09-06T00:00:00Z https://matrixcognition.com/posts/gemini-3-8-flash-release.html https://matrixcognition.com/posts/gemini-3-8-flash-release.html Google's 2 September 2026 release of Gemini 3.8 Flash, dated from the announcement, with the model page limits, the pricing expiry, and the eval table's footnotes. 2026-09-06T00:00:00Z https://matrixcognition.com/posts/bootstrap-eval-deltas.html https://matrixcognition.com/posts/bootstrap-eval-deltas.html A paired bootstrap for the accuracy gap between two systems on 500 items, with McNemar's test, coverage checked over 1,000 replays, and the code and per-item data. 2026-09-06T00:00:00Z https://matrixcognition.com/posts/earlyeval-cheaper-agent-evaluation.html https://matrixcognition.com/posts/earlyeval-cheaper-agent-evaluation.html A reading of arXiv:2609.02783, which trains classifiers to halt agent runs once the outcome is predictable, and what its Table III actually shows. 2026-09-05T00:00:00Z https://matrixcognition.com/posts/covert-indirect-prompt-injection-icoa.html https://matrixcognition.com/posts/covert-indirect-prompt-injection-icoa.html A reading of arXiv:2608.30362, which splits attack success into covert and overt outcomes on AgentDojo and shows the covert share can be engineered. 2026-09-05T00:00:00Z