In the following code segment from blake2s.cu (ccminer source code - cuda C)

uint32_t vhashcpu[8];
         uint32_t nonce = sph_bswap32(resNonces[1]);
         be32enc(&endiandata[19], nonce);
         blake2s_hash(vhashcpu, endiandata);

         *hashes_done = pdata[19] - first_nonce + throughput;

How do pool miners prevent coders from modifying the source code and recompiling ccminer so they fake the hash count and get an unequal share of the winning pools??  This would apply in pretty much all the algos...?