The story of the game
James, twelve moves, one decision. The engine's ledger for this game holds exactly one entry: 9...Nxd4 (46% → 19%) — every move after it was forced bleeding. That makes it strike three for the ...Qxb2 raid, and this time the raid was lost before the queen ever touched b2.
The key moment
9...Nxd4 — the grab that was a trap. 8...Qb6 put the queen on the classic poison-pawn diagonal, staring at b2 and d4, and it feels like White forgot both pawns. He didn't: after 10.Nxd4 the recapturing knight's next hop is the whole point. 10...Qxb2 11.Ndb5 cut the queen's road home while eyeing c7; 11...Qb4 12.Nc7+ forked king and rook. The engine's move was 9...Bd7 — develop, castle, and keep ...Qxb2 alive as a threat. A threat you don't execute still ties the opponent down; executing it handed the initiative to a knight with better plans.
The scan that was skipped: your own two questions — "what defends this pawn?" and "what does the recapturing piece do from its new square?" Nd4's answer was b5; b5's answer was c7. Two forcing hops, five seconds each.
What went right
The other eleven moves were clean — 88.8% accuracy in a lost cause. This was one decision, made in "I'm being clever" mode, which is exactly the trigger your profile documents. One decision is fixable.
Profile check
Fire alarm, third ring: this graduates ...Qxb2 from heuristic to hard rule. And clock the symmetry — two days after your 18.Nd7+ forked David, his 12.Nc7+ forked you. The family fork ledger is even.
What to drill next
- The absolute rule: no ...Qxb2 — and no committing to Qb6 poison-pawn setups — before move 15. Decline by policy, not calculation.
- Recapture-square scan before any grab: name the recapturing piece's next two forcing moves out loud.
- Play 9...Bd7 against the engine from this position a few times — feel how the unplayed threat keeps working.