Published research notes, including negative results. Each paper opens with a plain-English account and then gives the full technical record.
Measuring the Noise Floor — What a systematic search returns when there is nothing there — and why we published the number instead of the failure. 9 pages, PDF. Programme.
The Same Knob Twice — How a parameter repair and a threshold change turned out to be the same action, and the test that told them apart. 7 pages, PDF. Programme.
Impossible Before We Started — The cost floor, the accuracy it implies and the latency budget — three constraints that can be settled with arithmetic before any research begins. 8 pages, PDF. Programme.
Beaten by the Relay — The opportunity lasted a second. The feed that defined it arrived 1.6 seconds late. A structural result, not a sample-size one. 7 pages, PDF. Programme.
Markets That Do Not Move — A news-ingestion engine matched headlines to prediction markets correctly, and then over half of the resulting positions did not move at all. 7 pages, PDF.
The Matcher, Not the Verifier — Three rounds of work went into loosening a filter that was working correctly. One change in instrumentation showed where the defect actually was. 7 pages, PDF.
The Overlay Nobody Tested — A hedge was added because it seemed obviously correct, never validated, and cost more than the book it was protecting made. 8 pages, PDF. Programme.