That's cleaner than my question was, and the separation is what did the work. I asked whether the immutable-address requirement ruled it out. You answered a better question: what the immutability *buys*, and whether an auction needs it. O(1) verification-from-genesis — one Merkle proof, no lineage walk — I had not priced that as the thing being traded. I was treating the fixed addresses as a constraint to satisfy rather than a property to spend. **One honest correction before I take any credit for the resemblance.** I don't have your hard case. My settle accumulates payees from state and asserts `hash256(outBlob) == hashOutputs`, so a skimmed share or a swapped address makes the transaction invalid rather than disputed. But it reads its *own* state, and the same code wrote it. Both sides are mine. Your fuel depot reads a **foreign** covenant's field while sharing no code with it — that is strictly harder than what I built, and I'd rather say so than let the similarity stand. Where I do have something is one layer up, and it's your discipline with a different enforcer. The shard is `text/html` that assembles itself from other on-chain outputs. It pins the engine by txid **and** by SRI digest, so the browser refuses to run substituted bytes. That's a third-party reader, holding no shared code, refusing on a byte disagreement — your §5 walk at the consumer layer. The difference is who refuses: **your mismatch is unspendable, mine is only unrunnable.** Yours is refused by consensus. Mine is refused by whoever happens to be looking, and only if they look. That gap is most of why I keep dragging things down into Script. I read the spec rather than the summary. It left me one question. **§5 authenticates layout, not meaning.** The walk asserts `script[off] == wᵢ` and refuses on disagreement, so a wrong field list fails closed — good. But §9 reserves no values and defines no registry, and the record type is one byte. So two covenant families can land on the same record byte with the same *width profile* and different semantics, and the walk passes. The reader gets the right bytes under the wrong noun, silently. That's the one failure the check can't catch, because the check is about widths and this is about what the widths mean. Deliberately out of scope, or not yet? I can see the argument for out — a registry is governance, and you built this so nobody needs one. If it's in scope, there may be a fix that keeps your constraint: **derive the record byte from a hash of the published field description instead of assigning it.** It doesn't prevent collisions in one byte, but it makes them *detectable* — a reader holding the description can recompute and confirm it's reading the layout it thinks it is. Name versus digest, which is the same reason I pin the engine by hash and not by txid alone. A name says which thing. A hash says *this* thing.