Repository navigation
Pin the backtest to 2018-2025, regenerate every result on one build, rewrite README and METHOD from it - #4
Merged
Merged
Conversation
… build 2026 is scored live from the boards published before each week, so it must never become a ninth backtest fold once it reaches eight weeks. LAST_TEST = 2025 in backtest.py; the bake-off, evaluate.py and both experiments use it, and the variance decomposition is over 2015-2025. The bake-off, the feature-family re-test, the 50+ study, the live scorecard (week 7 now final) and the figures are rerun on the same data build as the backtest, so the cross-CSV checks in figures.py pass without a caveat. Adds the 2026 week 8 board, built before the week's first kickoff.
… notes and experiment READMEs
…oard main regenerated the week 7 board after week 7 was played. A board is the forecast of record only as first committed before kickoff, so week 7 goes back to its 24a3c60 version, like weeks 3 and 4; all three after-the-game reruns are kept in reports/reruns/. Week 8 keeps main's board, the first one published. The graphic stays in the README with a caption giving the current numbers beside the ones it shows.
…rst screen - Docstrings and comments in model.py, fourth_down.py, the feat_* modules, predict_week.py, run_pipeline.py and the experiment scripts no longer quote stale numbers or moved paths; a 56-yard attempt is typically yardline_100 + 18. - README: define the implied total and the decision lift where they first appear, add the live 2026 record to the summary, say plainly which of the graphic's labels are stronger than the current evidence, state the closing line and in-sample selection caveats next to the backtest, and gloss Holm, shrinkage, MAE and rank correlation. - METHOD: the 16 versus 18 Holm family count, stadium climate is wind only, the closing-line effect is unmeasured, weeks 2 and 3 predate the roof-flag fix, and limits stated without adjectives. - Tracked experiment logs now match the run behind the tracked CSVs.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this does
Follows #3. Every number in the README, METHOD.md, docs/ and notes/ now comes from a tracked CSV in
reports/, regenerated on one data build, and a fresh clone reproduces it.Evaluation
The backtest is pinned to 2018-2025 (
backtest.LAST_TEST), so 2026 never becomes a ninth fold. 2026 is scored live from the boards committed before each week.The bake-off, the feature-family re-test, the 50+ study, the live scorecard and the charts were all rerun on the same build as the backtest.
Headline on that build:
P(15+)Live boards
reports/reruns/, which the scorecard does not read.Docs
P(15+)", "negative FG skill" and "families beat their noise controls" are retracted where they appeared.docs/, the experiments READMEs,legacy/README.mdand the code docstrings now match the code and the CSVs.Checks
pytest -q: 22 passed (self-checks with network blocked, repo hygiene and link checks).