We model each stroke with a power-law fit t = A · d^B, where we use:

• Any real PBs at that stroke and distances (after a small correction explained below).
• The stroke-specific shape (B) anchored to world-record pacing if data is sparse.
• A cross-stroke synthesis if no PB exists for a stroke: scale from another stroke using world records at the same distance.

World-record correction: WRs across distances are fitted to reduce pacing bias, then used as a multiplicative correction by distance. Output times are formatted as m:ss:cc (hundredths).