Change log and corrections

how the model has changed, everything that has gone wrong, and the decisions and reviews behind them

The methodology describes what the forecast does now. These pages are its record: every change and test result by date, every defect found and what it cost, the decisions taken and the reviews that tested them. The plain-text review folder for outside reviewers is at /review.

• How the model has changed

  Every dated change, test result, correction and review note, newest first, with the evidence each one cites.

  Generated from the dated record

• Every test run on past elections

  Each written-up test and check of the forecast models and their data, by its own title.

  Generated from the test write-ups

• Everything that has gone wrong

  A numbered log of every defect found in this system: what it was, what it cost, how it was caught, and what now stops it recurring. This is the longest document here.

  Public defect log

• The adjustment register and the standard test

  The standard test every adjustment faces, its amendments and their dates, and the register of every adjustment with its test record.

  Full technical document

• Decisions and why

  The architecture decision records. Each one states the alternatives that were rejected and what would have to be true to revisit it.

  Full technical document

• Adversarial review

  The document sets out the findings from attempts by separate AI model sessions to break the forecast's conclusions, the answers given to those findings, and the objections still open; no person outside the project has done such a review yet.

  Full technical document

• Data sources

  Every source, its license, what is taken from it, when it was checked, and the ones that are deliberately not used with the reason.

  Full technical document

• What it is supposed to do

  The specification, and a row per requirement saying where it is implemented and what demonstrates that it works.

  Full technical document