Method
Changelog
Every public change to the method, the program that runs it and this site, newest first. Each method entry carries the version and the start of the sha256 of the files it describes, so a change can be checked rather than taken on trust. Failed tests stay in the list.
Method and program
The checks for each run are written and sealed before the data are opened. A later change is a dated amendment, sealed before it is run; the earlier result is never overwritten.
Change detector: sealed re-test passed, 8 of 8
Public dataWe fixed the method under an amendment sealed before the run, then ran it once on public files it had never seen. All four streams with a change were flagged at the right place, and none of the four streams without one was flagged. These are flow-cytometer files with whole-instrument swaps, which are large changes; the detector has not been run on any scheme's data.
Change detector for EQA schemes: first test on raw instrument files failed
Public dataOn public flow-cytometer files, the detector for undeclared instrument changes flagged both streams in which nothing had changed. Of the two streams with a change, it placed the change correctly in one. Its threshold assumed the noise between samples was better behaved than real samples are. The result stays on the record.
Seal record brought up to date without editing the original
The first seal pinned the v0.2 program files, which v0.2.1 replaced. A dated record now pins the current files; the first seal is left exactly as it was written.
Program v0.2.1: the same results in any row order
Public dataIn v0.2, re-exporting the same file in a different row order changed some re-test shares. v0.2.1 puts the pairs in a fixed order before any computation. The change was sealed as an amendment before it was run; its run gave the same totals and the same verdict on every check.
Program v0.2 run on public data: all five checks passed
Public data34 method comparisons at 14 stand-in sites, from 12 open-access studies: 10,579 paired results. This is the pre-registered result and it stays on the record as written.
Checks fixed and sealed for program v0.2
Before any public data set was run, we wrote down the five checks the run had to pass and sealed them together with the program and its configuration.
This site
Changes a visitor can see, dated by when they were made.
Manufacturer and EQA sections; this changelog
The page for diagnostics makers and EQA schemes now says what a conversion pilot delivers and who sees what, and how a scheme can work with us. This changelog was added.
Clearer difference label; privacy detail
The mean difference is labelled as a percentage of the old result wherever it appears. The privacy page names who hosts the site and the mailbox, and how transfers are covered.
One page per public method comparison
Each comparison in the change explorer has its own page with its figure, every number the program wrote, and the full citation of its study. Comparisons, stand-in sites and studies are now counted the same way on every page.
Registry and MedEval-1 pages
A page describing the designed Analyser-Change Registry, which no laboratory has joined yet, and a page on MedEval-1, the held-out evaluation for medical AI built by NakedSignal.
Guides published
Reference guides on bias at decision limits, crossings by chance, reagent lot changes, comparability across analysers and sites, point-of-care HbA1c, creatinine methods, and measurement change in trials and medical AI, each with its sources.
Band counts, attribution and sample scope corrected
Full and indicative re-test bands are counted separately; every public study is cited with its authors, source and licence; the sample report lists which parts of the full report it shows and why the others are not shown.
New nakedsignal.com
The site now describes what an instrument change does to results, with the change explorer, a sample report, the method, and how we handle data. Every result carries a label saying where it came from.