No generic memory games. Meta-analysis of 145 working-memory-training comparisons found no reliable far transfer — trained tasks improve, everything else doesn't (Melby-Lervåg, Redick & Hulme 2016, Perspectives on Psychological Science). Spatial ability is broadly trainable (Uttal et al. 2013, Psych Bulletin, g≈0.47), yet adult chess skill correlates only weakly with general visuospatial ability (Waters, Gobet & Leyden) — the skill lives in chess-specific structures. Every drill here is a real position for that reason.

Skilled imagery is chunk-driven, not photographic. Blindfold-chess experiments show strong players encode positions functionally — what pieces do, not what they look like — and recall functionally significant areas far better (Saariluoma 1991, Acta Psychologica; Saariluoma 1998, Memory). Hence "track what changed," not "memorize the picture."

Visualization rides on pattern knowledge. Computer analysis of hundreds of serious GM games found blindfold play barely raised blunder rates versus rapid with sight of the board — time pressure, not sight, drove errors (Chabris & Hearst 2003, Cognitive Science). Caveat that cuts in your favor: that's GMs, whose chunks carry the image for free; at club level the image plausibly still binds — which is exactly what the depth meter measures. Eye-tracking agrees on what to encode: experts fixate between pieces, on relations, not piece-by-piece (Reingold & Charness 2001, Psychological Science).

Blindfold solving demonstrably transfers. An experimental study (FIDE 1650–2100, n=38) trained notation-only tactics — no board, repeated exposures weeks apart — and found a 31% average reduction in tactic solving time (p<0.005) versus no change in controls, with 1900+ players improving most. Mode 4 replicates that protocol; the woodpecker pass system covers the repetition. Small study, real signal.

Frozen-board line reading is the established format. Commercial visualization trainers (Visualwize's 4 depth levels; calculation workbooks from Thinkers Publishing; Justesen's notation-only blindfold puzzle books) all use it: position stays put, the line unfolds in your head, depth scales with skill — exactly modes 2–3, with the depth meter measuring where your board corrupts.

Waypoints beat one giant leap. The "stepping stones" technique — deliberately fixing an intermediate position in the mind's eye and calculating onward from it — comes from Tisdall (Improve Your Chess Now!, 1997); "every move does four things — a square empties, a piece's control leaves, a square fills, new control arrives" is Rayner's operational version. Mode 2 drills both: mid-line waypoint checks every two plies, then full reconstruction at the end. Its rigid opposite, Kotov's calculate-each-branch-once tree, is widely criticized as unrealistic. Support here fades as you climb (adaptive depth, no hand-holding) per expertise-reversal findings (Kalyuga).

Why you answer on the board, not in words. Dual-task experiments show chess thinking is crippled by visuospatial and executive interference but essentially untouched by articulatory suppression (Robbins et al. 1996, Memory & Cognition) — the calculating mind is spatial-executive, and verbal narration is a light mnemonic at best. Imagery format also varies person to person, down to full aphantasia (Pearson & Kosslyn 2015; Zeman et al. 2015) — propositional trackers calculate fine.

Honesty clause. Direct evidence that visualization training raises playing strength is one small controlled study plus strong mechanistic theory — thin. That's why this tab measures your own curve (reliable depth, ghost rate, seconds per answer) instead of promising transfer: your data decides whether it's working.