The Compression

Once a pathway has been walked, the interesting question is not whether it worked. It is: how few words carry this to the next network?

Two words kept apart

Address — a content hash of a telling. What was released, addressed rather than disclosed: an auditor can verify exactly what crossed without ever seeing it.

Ratio — words in the telling over words in the source. What it cost to say.

These get conflated constantly and must not be. One is identity, the other is economy.

Who judges fidelity

Not the measuring tool. It counts words, hashes text, and checks whether each page of the pathway is touched at all — a check that is necessary but not sufficient: it can prove a page was dropped, never that its meaning survived.

The judgment belongs to a two-agent harness. One agent proposes the telling. A second is given only the telling, reconstructs the invariant core from it, is then shown the pathway pages, and must name every place the two disagree. Propose against validate; only a validated compression is minted.

Precomputed, not on demand

Tellings are written once, offline, when the index runs — so serving a tier is a file read and a small always-on machine can answer instantly. Whether that is possible at all is an open assumption, not a settled fact: it presumes a telling written without knowledge of the question is still useful when the question arrives. Building it against a real corpus is how that gets tested rather than argued.