https://evoclock.github.io/fieldnotes/ Writing and evidence: agent systems, models, evaluation and computational biology. en-GB Fri, 23 Oct 2026 00:00:00 +0000 https://evoclock.github.io/fieldnotes/articles/60-percent-of-the-time-it-works-every-time.html https://evoclock.github.io/fieldnotes/articles/60-percent-of-the-time-it-works-every-time.html Fri, 23 Oct 2026 00:00:00 +0000 Is published research data ready to train models? I revisit four circadian ChIP-seq studies and what their raw data can support under consistent reanalysis. comp-bio evaluation https://evoclock.github.io/fieldnotes/articles/is-fast-decoding-the-be-all-and-end-all.html https://evoclock.github.io/fieldnotes/articles/is-fast-decoding-the-be-all-and-end-all.html Fri, 09 Oct 2026 00:00:00 +0000 Four serving recipes for GLM-5.3-Flash on two DGX Sparks. The recipe that answers fastest is not the one that generates fastest, and none of the four escapes the constraints that matter for real agent work. models evaluation https://evoclock.github.io/fieldnotes/articles/herdr-natural-language-agent-automation.html https://evoclock.github.io/fieldnotes/articles/herdr-natural-language-agent-automation.html Sat, 05 Sep 2026 00:00:00 +0000 How the Agentic Driver extension set uses Herdr and Pi to route tasks by role, model and machine—and keep persistent sessions within reach from a laptop, phone or remote terminal. agent-systems https://evoclock.github.io/fieldnotes/articles/wrangling-qwens-long-thinking-runs.html https://evoclock.github.io/fieldnotes/articles/wrangling-qwens-long-thinking-runs.html Wed, 26 Aug 2026 00:00:00 +0000 How I manage Qwen's tendency to go off on a long reasoning run, why completion limits are not enough, and where quantisation creates a second serving problem. models agent-systems https://evoclock.github.io/fieldnotes/articles/primitives-were-already-there.html https://evoclock.github.io/fieldnotes/articles/primitives-were-already-there.html Fri, 21 Aug 2026 00:00:00 +0000 Why AI infrastructure keeps rediscovering established primitives, and how to distinguish useful standardisation from inflated novelty claims. agent-systems https://evoclock.github.io/fieldnotes/compbio/circadian-chipseq-audit.html https://evoclock.github.io/fieldnotes/compbio/circadian-chipseq-audit.html Sun, 09 Aug 2026 00:00:00 +0000 A method reconstruction, sensitivity analysis and local ENCODE-equivalent comparison for public mouse liver circadian factor ChIP-seq. No tested condition reproduced both the deposited peak counts and the peak sets. comp-bio evaluation https://evoclock.github.io/fieldnotes/notes/seat-benchmarking.html https://evoclock.github.io/fieldnotes/notes/seat-benchmarking.html Sun, 09 Aug 2026 00:00:00 +0000 Choosing a model for a specific job, and deciding what to change when it fails. evaluation https://evoclock.github.io/fieldnotes/notes/memory-management.html https://evoclock.github.io/fieldnotes/notes/memory-management.html Sun, 09 Aug 2026 00:00:00 +0000 Model state, session history and external retrieval belong to different parts of the agent system. agent-systems https://evoclock.github.io/fieldnotes/notes/watts-per-token.html https://evoclock.github.io/fieldnotes/notes/watts-per-token.html Sun, 09 Aug 2026 00:00:00 +0000 A 6.6 W fan, a 35 W desk fan, and the cost of keeping a model running. evaluation https://evoclock.github.io/fieldnotes/articles/local-multi-model-workforce.html https://evoclock.github.io/fieldnotes/articles/local-multi-model-workforce.html Thu, 30 Jul 2026 00:00:00 +0000 From finishing my PhD to keeping knowledge, model choice and authority in my own systems. agent-systems https://evoclock.github.io/fieldnotes/publications/project-brief.html https://evoclock.github.io/fieldnotes/publications/project-brief.html Wed, 29 Jul 2026 00:00:00 +0000 The project brief: a task-bound Implementer, what it learned, and what the coding tests showed. models https://evoclock.github.io/fieldnotes/publications/technical-report.html https://evoclock.github.io/fieldnotes/publications/technical-report.html Mon, 27 Jul 2026 00:00:00 +0000 Two starting checkpoints, executable training evidence, and fifteen paired coding-retention runs. models evaluation https://evoclock.github.io/fieldnotes/evals/fable-run5-granite.html https://evoclock.github.io/fieldnotes/evals/fable-run5-granite.html Fri, 10 Jul 2026 00:00:00 +0000 Run 5: a small-model signal on a storage-budget trap, with a same-length sham control. evaluation https://evoclock.github.io/fieldnotes/evals/cruxeval-o-ab-184.html https://evoclock.github.io/fieldnotes/evals/cruxeval-o-ab-184.html Fri, 10 Jul 2026 00:00:00 +0000 Seven local models, 184 output-prediction problems, and a prompt change that helped some and hurt others. evaluation models https://evoclock.github.io/fieldnotes/evals/screen_eval_run4.html https://evoclock.github.io/fieldnotes/evals/screen_eval_run4.html Wed, 08 Jul 2026 00:00:00 +0000 Run 4: Qwen created a little headroom during calibration, but all three arms passed the follow-up test. evaluation https://evoclock.github.io/fieldnotes/evals/cruxeval-o-results.html https://evoclock.github.io/fieldnotes/evals/cruxeval-o-results.html Wed, 08 Jul 2026 00:00:00 +0000 Seven models over 100 CRUXEval-O problems, with prompt, parser and infrastructure failures kept visible. evaluation models https://evoclock.github.io/fieldnotes/evals/screen_eval.html https://evoclock.github.io/fieldnotes/evals/screen_eval.html Tue, 07 Jul 2026 00:00:00 +0000 Sonnet 5, three prompt conditions, three difficulty levels and 27 responses: testing error detection separately from presentation. evaluation https://evoclock.github.io/fieldnotes/evals/trap_eval.html https://evoclock.github.io/fieldnotes/evals/trap_eval.html Tue, 07 Jul 2026 00:00:00 +0000 Run 2: holding the model constant separated presentation changes from trap detection. evaluation