An asset issued on one network has one supply, and reading it is arithmetic. The moment the same economic interest exists on two networks, nobody holds the whole picture, and the question “how much of this exists?” stops having an answer you can look up.
This is not a theoretical concern. It is the normal state of any asset that has been bridged, and it is the mechanism behind a familiar class of failure: more claims outstanding than there is backing to meet them, discovered late.
Four ways supply drifts
Bridged representations counted twice. A unit locked in a contract on one network and reissued on another exists, in a naive count, in both places. Whether that is double-counting depends entirely on whether the locked original is excluded — which requires knowing the route existed.
Issuance on one leg, redemption on another. Units created where demand is and redeemed where liquidity is. Each network sees a coherent local picture; the aggregate is what moved.
Burn-and-mint that half-completes. The destruction and the recreation are separate events on separate networks. Between them the total is wrong, and if the second never lands it stays wrong.
Units that were never issued by you. Someone deploys a contract with your asset’s name. A count that trusts a token symbol rather than a specific contract address will include them.
Why reading each chain is not enough
Each network can be read accurately and the sum can still be wrong, because the sum is not the question. The question is how many distinct claims on one reserve are outstanding, and that requires knowing which representations are the same claim wearing different clothes.
That knowledge does not live on any chain. It lives in the asset’s declaration: these contracts on these networks are representations of this asset, and these routes connect them. Netting at the source — subtracting what a declared route has locked — is what turns several local counts into one global one.
The corollary is that routes have to be declared rather than detected. Inferring from on-chain activity which transfers were bridge movements is guesswork, and guesswork in a supply figure is worse than an honest gap.
Reserve is the other half
Knowing the supply is only useful against something. The invariant worth maintaining is that total circulating representations never exceed the attested reserve backing them.
Both sides of that comparison are awkward. Supply has to be read at a pinned point so two readers get the same answer. Reserve is an assertion by somebody — an issuer, a custodian, an auditor — and inherits exactly the credibility of whoever asserted it.
Two properties make the comparison worth trusting. It should be repeatable: every reading names the block it was taken at, so a third party can repeat the query and get the same number. And it should be contestable: published before it binds, with a window in which anyone can challenge it, so a wrong figure can be caught by someone who is not the person who produced it.
Surfaced, or enforced
There are two useful things to do with the comparison.
For assets issued by someone else, the most you can do is publish: netted supply, attested reserve, and the headroom between them, every epoch. That makes a divergence visible to everyone at once, which is considerably better than it being visible to nobody.
For assets issued through the system itself, the invariant can be a condition of issuance rather than a report about it. A request to create units that would exceed the attested reserve is refused. This is the stronger form, and it is only available where the issuance path runs through something that checks.
What this does not fix
Reconciliation tells you the representations agree with the attestation. It cannot tell you the attestation is true. If a custodian asserts a reserve that is not there, every number downstream is confidently wrong, and the system will report full backing until someone tries to redeem.
That is the boundary between this problem and a different one: binding an off-chain fact to an on-chain record in a way a supervisor accepts. Allowing part of a stated backing to come from hardware-attested measurement, published with its uncertainty and open to challenge, is a partial answer for some asset classes and no answer at all for others.
Where this is implemented
Multi-network supply reading, declared bridge routes netted at the source, continuous comparison against an attested reserve, and refusal of issuance beyond it are implemented at boli.technology. The open problems page places this alongside the questions it does not answer.