It works today and can be used to prove exact size: the key observation is that all you need to show the length and hash of a transaction is the final SHA256 midstate and chunk (max 64 bytes). It also uses the observation that a valid transaction must be at least 60 bytes long for compression (though much of that compression possibility goes away if you're proving something other than "too large").

On March 22, 2017 1:49:08 PM PDT, Bram Cohen via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:
>Some questions:
>
>Does this require information to be added to blocks, or can it work
>today
>on the existing format?
>
>Does this count number of transactions or their total length? The block
>limit is in bytes rather than number of transactions, but transaction
>number can be a reasonable proxy if you allow for some false negatives
>but
>want a basic sanity check.
>
>Does this allow for proofs of length in the positive direction,
>demonstrating that a block is good, or does it only serve to show that
>blocks are bad? Ideally we'd like an extension to SPV protocol so light
>clients require proofs of blocks not being too big, given the credible
>threat of there being an extremely large-scale attack on the network of
>that form.
>
>
>On Wed, Mar 22, 2017 at 1:47 AM, Luke Dashjr via bitcoin-dev <
>bitcoin-dev@lists.linuxfoundation.org> wrote:
>
>> Despite the generalised case of fraud proofs being likely impossible,
>there
>> have recently been regular active proposals of miners attacking with
>simply
>> oversized blocks in an attempt to force a hardfork. This specific
>attack
>> can
>> be proven, and reliably so, since the proof cannot be broken without
>also
>> breaking their attempted hardfork at the same time.
>>
>> While ideally all users ought to use their own full node for
>validation
>> (even
>> when using a light client for their wallet), many bitcoin holders
>still do
>> not. As such, they are likely to need protection from these attacks,
>to
>> ensure
>> they remain on the Bitcoin blockchain.
>>
>> I've written up a draft BIP for fraud proofs and how light clients
>can
>> detect
>> blockchains that are simply invalid due to excess size and/or weight:
>>
>>    
>https://github.com/luke-jr/bips/blob/bip-sizefp/bip-sizefp.mediawiki
>>
>> I believe this draft is probably ready for implementation already,
>but if
>> anyone has any idea on how it might first be improved, please feel
>free to
>> make suggestions.
>>
>> Luke
>> _______________________________________________
>> bitcoin-dev mailing list
>> bitcoin-dev@lists.linuxfoundation.org
>> https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev
>>