Research, in writing.
Papers, notes and essays from Forticia Research Institute. Every piece states what it claims and what it does not, and shows its sources.
- Published
- 9
- Disciplines
- 4
- Latest
- 6 Oct 2026
Archive
9 of 9 papers
- Paper
How many independent proteins are in a benchmark? Error correlation across sequence similarity and the cost to error bars
Model errors on related proteins are correlated even at the weakest significant hits. We measure it and what it does to benchmark confidence intervals.
Biology21 min read - Paper
Where identical code gives different answers: a reproducibility budget for a protein-sequence pipeline
One protein pipeline, 192 runs, one factor changed at a time: seeds dominate, thread and BLAS noise is invisible, and one library release flips a verdict.
Biology17 min read - Paper
Canaries in the approval queue: what injected known-bad requests buy a fatigued human reviewer
In simulation, injected known-bad requests help a fatigued reviewer only above a spare-attention threshold, and recognisable canaries hide the failure.
AI15 min read - Paper
What an agent audit log can answer: measuring incident answerability across logging schemas
Simulated approval incidents and a hash-chained log show which fields answer which questions, what defaults cost, and how anchoring sets tamper detection.
AI16 min read - Paper
Gate stacks are not products: measuring the rubber-stamp risk of a research pipeline
In a simulated research pipeline, correlated gates protect up to 18 times less than their pass rates imply, and a one-draw placebo is a coin flip.
Quant19 min read - Paper
When a failure record lies: the economics of shared negative results in a research swarm
In a simulated swarm, sharing falsified ideas saves a third of compute, but neighbourhood rules close off real effects unless records carry their test power.
Quant17 min read - Paper
Never Tune a Corpse
When may a rejected idea be reopened? Simulated burial and exhumation rules, their false discovery cost, and how the graveyard can audit the court.
Culture16 min read - Paper
When Approval Stops Carrying Information
A conditional-information measure of rubber-stamping, the defect rate where review rationally collapses, and which levers restore it. A simulation study.
Culture18 min read