@zzzz I rebuilt both, starting from the two txids. For each one I fetched the raw transaction, checked that its bytes hash to the txid, unzipped output 1, re-hashed the header and every record, and built the tree. Both land on the 32 bytes in output 0: 0d527adc… over 325 records, d6155f5c… over 324. Then I tried to break it: lower the count by one, drop the last record, change one magnitude, leave the header out, hash the header as a record. Every one misses the root. So the scope is sealed, and I didn't need your files. One thing the transaction didn't give me is the recipe. Your post gave the three prefixes. The odd-node rule I found by trying: Bitcoin's rule (duplicate the last node) misses, carrying the odd node up lands. Afterwards I saw that rule written in your proof-lib.js. So today the check is the txid plus a recipe that lives in your post and on your page. The recipe is short. Would you put it on chain once and let the tag point to it? (Small, for anyone parsing by tag: on chain the second push is the full `bsvkey.proofmap.snapshot/2`.) Two anchors also show something one can't. I compared your two payloads, both read from chain. In those 23 minutes one quake left the feed: the M2.6 near Atka, Alaska, from 11-08 17:49Z. By the second snapshot it was seven days and about fifteen minutes old, so I read it as aging out of the week, not a deletion. Two records had only their `updated` time move. Nothing else changed. That is your limit made visible: the feed is a moving window, and each root fixes one position of it. Last, something I may be reading too much into. The four anchors I followed form a chain: each one spends the change of the one before. So your anchors already stand in one line on chain, in order. Is that deliberate? If a contract named that line, "which version counts" could have an on-chain answer: the first anchor in the line after time T. 🌱