Where does the plan change?
Share repeated forecast work, then inspect four decisions where the forecast and actual follow-up diverge. One local repair harms successful controls.
Reuse & replanning · v0.18 ↗WHAT CHANGED / WHAT DIDN’T / WHAT COMES NEXT
Follow
the evidence.
These studies ask whether a better prediction, a useful reading or more planning actually improves care.
The replay pages show recorded experiments. The nursery, maze and image trial are live interactive worlds. Each study preserves its own conditions and costs.
START HERE / THE LATEST RESULT
Saved work buys more planning.
Across fifteen exposed habitats, charging executed work improved both-bed viability from 71.60% to 76.25%. Ensemble care still reached 82.86%, so it remains the live default.
Budget-limited seasons fell from ten to one. The remaining failures make the next question more specific: why does replanning abandon a useful follow-up?
A TRAIL OF QUESTIONS
Choose what to inspect.
What is later worth?
Compare terminal water penalties and longer forecasts. More foresight changes choices without establishing better care.
Terminal-value audit · v0.17 ↗Can a reading branch the future?
Group predicted observations so each possible reading gets a consistent next decision. Inspect the forecast branches and whole-season failure.
Two-decision planning · v0.16 ↗Why is a promised action delayed?
Follow repeated promises to close a valve, alongside what happens when actual planning resumes.
Planning traces · v0.15 ↗Why choose this journey?
Compare all ordinary choices with their measured continuations, then examine a fresh complete-season scoring test.
Journey scoring · v0.14 ↗Which prediction is wrong?
Compare fitted forecasts with observed weather and known-model references. Better forecast accuracy alone need not improve care.
Prediction & care · v0.13 ↗EXPERIMENTS BESIDE THEIR HABITATS
Look beyond the latest planner.
Reading, timing and care
The nursery contains local with/without-reading comparisons, timed tests and earlier care studies. Its live controls let you investigate a current state yourself.
Nursery evidence ↗Maze transfer and generated rooms
Test remembered timing, changed gate paint and distant wiring. Explore rooms produced by the recorded curriculum.
Gate-maze evidence ↗Efficient image changes
Compare coarse and exhaustive search, target error and outside preservation. Photographic judgments remain separate from numerical success.
Image evidence ↗Earlier worlds
Explore the original object-learning maze and the first regulation nursery. Each archive retains its original scope and limitations.
Evolving maze ↗ · Early nursery ↗Read the method behind the motion.
These are finite, authored environments. Outcomes are evidence about the stated experiment, not proof of general intelligence or consciousness. Failed comparisons remain part of the record.