The majority hashrate (aka “51%”) attack which orphans some or all of the minority’s blocks is not objectively distinguishable from a random result. Some blocks are orphaned and others are not, yet there is no way to objectively prove from the information stored in the blockchain that a 51% attack was present. Clues might be indicative of a 51% attack such as a higher orphan rate, but this can also be caused by network delays which is what the 51% attack masquerades to other mining nodes. Tracking down nodes which are participating in the attack is not plausible because IP addresses and even pools can be a Sybil attack.[41] There can’t exist an objective perspective in the Byzantine Generals Problem (aka “BGP”) without a total perspective.[42] Yet a total perspective is not a BGP.
Unless a minority hashrate miner has, or is in a pool that has, at least ¹/₃ of the systemic hashrate, they won’t be able to triangulate to differentiate between random bad luck or a 51% attack. Even if they do have sufficient hashrate to probabilistically distinguish, they can’t necessarily convince the rest of the minority, because it is not objectively provable by anything that can be put on the blockchain. Propagation is not proof because every node’s perspective of propagation is different, i.e. propagation is Einstein’s special relativity.[45] Who is telling the truth? This is the self-referential insolvability of the BGP.[42] And because propagation is not proof, the timing gap delays analysis[43] [44] is inconclusive and not even wrong.
Thus there could be a vacuous point by Bitcoin maximalists (or supporters of any PoW blockchain) who argue that selfish mining attacks have never been confirmed in the wild, because we won’t be able to objectively prove it when they are and we can’t objectively prove they aren’t. Unlike a selfish mining attacks which employ only a minority of the systemic hashrate, a 51% or greater attack on minority blocks wouldn’t necessitate large chain reorganizations. Bitcoin Core apparently has a feature to warn of large reorganizations which thus would be not be triggered,[46] and which is inconclusive any way for detecting the less-than-majority-hashrate selfish mining case.[41]
Although a total perspective is theoretically observable by the community-at-large via coordinated reputation, reputation begets politics which is manipulable. For example, w.r.t. to the timing gap method of detection,[43] [44] we can’t objectively prove Blockchain.info isn’t complicit and can only rely on reputation. Political reaction requires a (manipulable and otherwise unlikely) monolithic political hardfork action with the entire loss of security to due to the necessity of replacing the proof-of-work algorithm to prevent the existing majority miners from unduly influencing, manipulating, or controlling the outcome. Yet monolithic political action is typically divided-and-conquered by multiple, simultaneous, divisive issues. Voting is a power vacuum, voting is divided-and-conquered by one election pertaining to multiple divisive underlying issues, and objectivity of reputation is impossible when obscured manipulative deals are possible behind-the-scenes[47] while profit due to such malfeasance is not effectively bounded.[3] [4]
Another form of 51% attack which is not provable and thus is “invisible” (i.e. not objectively distinguishable) is for example the miners refusal to accept a protocol change such as might be the case for the Bitcoin block size increase political battle.
PoW allows insidious attacks that no one can prove nor disprove. What better financial weapon could one design to aid the gazillionaire capitalists who don't want their power and actions to be observable.[48]
[3] Eric S. Raymond, Some Iron Laws of Political Economics. Armed and Dangerous blog, May 27, 2009.
[4] Paul Sztorc. Nothing is Cheaper than Proof of Work. Truthcoin.info blog, §Money and Politics, Aug 4, 2015.
[41] Ittay Eyal, Emin Gün Sirer. How to Detect Selfish Miners. Hacking, Distributed blog, Jan 15, 2014.
[42] Shelby Moore III. Self-referential insolvability of the Byzantine Generals Problem. Bitcointalk.org, “Satoshi didn't solve the Byzantine generals problem” thread, post #119, Feb 9, 2016.
[43] Matt Springer. Is Bitcoin Currently Experiencing a Selfish Miner Attack?. Scienceblogs.com blog, Jan 11, 2014.
[44] Fangyang Cui. Detecting Selfish Mining in Bitcoin. Dec 7, 2015.
[45] Shelby Moore III. The Universe. Unheresy.com blog, §Observer invariance, Jul 19, 2013.
[46] Murch. Implicit/Tacit mechanisms in Bitcoin Core disclosing Selfish Mining. Bitcoin.stackexchange.com, Jun 10, 2015.
[47] smoothie, smooth, r0ach, monsterer, DecentralizeEconomics. Bitshares' DPoS "behind-the-scenes" politics. Bitcointalk.org, “The state of crypto - The only serious thread on the subforum” thread, posts #137–155, Sep 9, 2015.
[48]: Shelby Moore III. Bitcoin: The Digital Kill Switch. Bitcointalk.org, MarketOracle.co.uk, Mar 29, 2013.