In　this　paper,　we　provide　a　novel　dynamic　decision　method　of　blockchain　selfish　mining　by　applying　the　sensitivity-based　optimization　theory.　Our　aim　is　to　find　the　optimal　dynamic　blockchain-pegged　policy　of　the　dishonest　mining　pool.　To　study　the　selfish　mining　attacks,　two　mining　pools　are　designed　by　means　of　different　competitive　criterions,　where　the　honest　mining　pool　follows　a　two-block　leading　competitive　criterion,　while　the　dishonest　mining　pool　follows　a　modification　of　the　two-block　leading　competitive　criterion　through　using　a　blockchain-pegged　policy.　To　find　the　optimal　blockchain-pegged　policy,　we　set　up　a　policy-based　continuous-time　Markov　process　and　analyze　some　key　factors.　Based　on　this,　we　discuss　monotonicity　and　optimality　of　the　long-run　average　profit　with　respect　to　the　blockchain-pegged　policy　and　prove　the　structure　of　the　optimal　blockchain-pegged　policy.　We　hope　the　methodology　and　results　derived　in　this　paper　can　shed　light　on　the　dynamic　decision　research　on　the　selfish　mining　attacks　of　blockchain.　©　2021,　Springer　Nature　Switzerland　AG.