I didn't have a chance to look at this in detail, but if the sender knows the private key, then double spends are a concern and the incomplete mitigation (to spend asap) is a timing analysis vector and afaics brings the traceability right back onto the blockchain.

Edit: myself (and I assume smooth and others) have thought deeply about all these sort of schemes and dismissed them. There isn't likely anything new under the sun except for some exotic new crypto that improves upon the Zerocash, Zerocoin, or fully homomorphic schemes. I would like to see a quantum resistant reformulation of Cryptonote.

Untraceability (e.g. when merging change from prior txns) can only be secured mixing of txns on the blockchain. Mixing can be done on the blockchain using an orthogonal protocol such as CoinJoin, but as explained by myself and smooth upthread, CoinJoin is not autonomous, is not immune to being jammed by a determined attacker, and has a user simultaneity requirement that will prevent it from scaling. Darkcoin mitigated the jammability by implementing masternodes which compromise anonymity against a ubiquitous attacker such as the NSA and they were the last time I checked adding some pre-mixing scheme to mitigate the simultaneity issue but that pre-mixing comes at a cost (and I think also a cost to scalability as well). Afaics after much thought and study, onchain mixing such as Cryptonote's one-time ring sigs (or Zerocoin and Zerocash but we pointed out their vulnerabilities upthread) is the only way to achieve anonymity with ledger-based crypto-currency. Note I also pointed out upthread that one-time ring sigs are not sufficient, and we need IP obfuscation as well. I argue that Tor (and likely I2P as well) are not entirely reliable IP obfuscation (argue they are subject to various attacks on the anonymity).