idleon-automation — docs
Companion notes and tools for the screen-reading bots at
github.com/jandersson/idleon-automation.
Predictors
- predictors.md — refresher notes on the
prediction algorithms in
common/predictor.py: KNN with
inverse-distance weighting, bivariate OLS, GP regression (the
default), the GP classifier behind hoops’ make_prob
candidate-ranking, and the planned darts E[stripe] model with its
2D wind encoding.
- predictor_playground.html —
“the bot’s eye”: all four models fitted live in the browser on the
real shot record. Hover to watch KNN pick its neighbours, scrub the
GP lengthscale, see the σ map glow where the bot is guessing, and
run the bob loop through the make-probability field to watch it
pick its firing moment.
Investigations
- hoops_findings.md — the June 2026 miss
investigation: why misses were never aim error (structure clanks +
launch-velocity physics), the per-hoop direction policy, multi-signal
make detection, and the measured outcome (25.5% → 34.8% baseline,
then the make_prob classifier beyond it).
Backlogs