Hi @sweaty_dog,
Bit-packing itself is permitted, and its use alone will not affect prize eligibility. As @dipam noted, we cannot impose a blanket ban. A packed word is indistinguishable from an ordinary integer value at the metering level, and the same bitwise operations can have legitimate uses.
The final determination will nevertheless be made by ARC during the manual validation of prize-contending submissions. At a high level, the research goal of the challenge is to develop estimation methods beat black-box sampling.
We have seen some teams primarily focussed on increasing sampling throughput by moving substantial numerical work outside Flopscope, for example through custom numerical kernels, or by relying on operations that are materially underpriced by the cost model to use a higher “effective sampling budget” than the competition spec allows.
The concern is not the use of bit-packing, custom code, or any particular operation by itself. It is whether those techniques are being used primarily to obtain substantially more effective sampling budget than the accounting is intended to permit.
Accordingly, a legitimate use of bit-packing as part of an estimation method should be fine. However, if it is used primarily to circumvent the intended compute accounting and create an effectively larger sampling budget, ARC may adjust or re-evaluate the submission and, if required, disqualify it.
So there is no automatic penalty based solely on the use of bit-packing. Its role will be assessed in the context of the submission as a whole.