Derive difficulty from right and wrong.
Every word in the 617-word list carries a predicted difficulty built from grapheme structure, sound, morphology, root pattern and lexical load (spellclimb-difficulty-v0, predicted-unfitted). Predictions are only a start. Judge them, or answer against them — right and wrong responses are what turn a prediction into a measurement.
Predicted difficulty is unfitted — a component heuristic, not yet a measurement. Responses also sync anonymously to the SpellClimb API; local records remain the offline buffer. Answer honestly: reveal, then mark whether you had it.
Say the meaning out loud — or write it — before you look.
Derived from right/wrong so far
No responses yet. Answer a round in Respond mode and empirical difficulty appears here, hardest first.
Residual = empirical − predicted (0–1 scale). Large positive: harder in practice than predicted. Treat n < 3 as noise.
Export for the pipeline
Responses feed node scripts/calibrate-difficulty.mjs (Rasch fit when names are recorded). Judgements refit the component weights. Files download to this device; nothing is uploaded.
How the derived difficulty works
In Respond mode each Had it / Not yet is a right/wrong observation. Per word the page keeps count and derives empirical difficulty as a smoothed logit of the proportion correct — words everyone gets sit low, words everyone misses sit high, and small samples stay visibly uncertain. Export the responses and run node scripts/calibrate-difficulty.mjs to fit the Rasch model and rank exactly where the component prediction disagrees with practice.
In Judge mode you mark each prediction Right, Too hard or Too easy. That is the fastest way for one expert to refit the weights before any learner meets the tiers. Records sync anonymously (device token, no names) to the SpellClimb API and are also kept in this device’s local storage until you export them.
The word list itself is on the full scholarship vocabulary list, and the drill this calibration feeds is the Vocab Sprint. Method and limits: docs/spellclimb-difficulty-scope.md.