unchallenged_chain anchors on the wrong end of the chain?

[justinstimatze]

Ran into this while looking at #150. It's a separate thing and it predates that fix, so I'm filing it on its own.

extract-prompt.ts describes depends_on as "one claim only holds if another premise is true", so in {from, to} the premise is the to. detectLoadBearingVibes reads it that way too — it counts incoming edges and calls them "holding up a few things".

The chain detectors go the other direction. longestChainNodesFrom follows outgoing edges, so chain[0] is the most-derived claim and the premise ends up at the far end. Both chain detectors anchor on chain[0].

Small repro — with d1 depends_on p, d2 depends_on d1, d3 depends_on d2, the premise is p:

detectUnchallengedChain → anchors on d3

The rendered phrasing is "A lot of things follow from '{claim}' … worth a 'what if not' pass", which describes a premise, but {claim} gets filled with the conclusion. So the nudge points at the wrong claim. productive_stress_test behaves the same way.

I can't tell which half is wrong:

• if the anchor is wrong, it wants to be chain[chain.length - 1], and anchor-head.test.ts needs flipping (it currently asserts the head and calls it the premise)
• if the anchor is right, then the comments and the phrasing are what need fixing

Happy to send a PR either way. Also happy to close it if I've got the edge convention backwards.

One unrelated thing in the same function: the byAnchor dedupe looks dead, since chain[0] is always the start node so the anchors are already unique.