Findings
Four results define the current evidence: developmental histories vary along broad gradients; their order matters; recent state dominates next-day prediction; and matched mismatches identify what to measure next.
Evidence boundary. Every number below belongs to a named cohort. The source archive, primary morphospace, event-weather models, transition models, and matching analysis do not share one universal N.
Walk through all five main figures → Start with the validation challenges
The scientific argument
Fires follow different developmental pathways.
The main coordinates describe broad differences in when growth was allocated: earlier versus later, and concentrated versus more distributed. Five axes summarize 89.4% of standardized shape variance.
See the morphospace → Finding 02 · Null sample N = 4,000The order of growth contains information.
Rearranging the same observed increments changes developmental traits and space coverage. Real histories allocate 54.1% of growth to the first half on average, versus 50.0% after shuffling.
See the ordering test → Finding 03 · 87,944 transitions from 31,700 firesRecent fire state explains much more than weather alone.
The region-held-out autoregressive state model reaches R² = 0.448. Weather adds ΔR² = 0.005; the full weather × state model adds ΔR² = 0.018.
Compare the evidence → Finding 04 · 9,212 eligible complete firesSimilar weather can accompany different histories.
Observed mismatch among weather-matched pairs is 49.7%, close to the 50.4% conditional-permutation mean. The examples locate useful contrasts without claiming excess mismatch or a mechanism.
See the matched fires →Read the evidence in sequence
The findings move from representation to ordering, then from prediction to unresolved mechanism. This sequence prevents weather or endpoint outcomes from defining the response they are later asked to explain.