p1 · the assumptions registry — eidos replaces lean
claim
the owner (2026-10-27): a clear registry of the stack's scientific assumptions; lean is foreign, eidos is native, and whatever eidos lacks to replace lean must be built. the unit is one proposal page, proposals/assumptions-registry.md, and its index row. the pull request: soft3#398.
for
one list of everything the stack assumes, classed, with a response plan per entry; the list is the axiom file of a soft3 eidos corpus, so a theorem can only rest on a named assumption; 2026-10-05 to 10-26 (3SUM, APSP refuted 10-05; UGC proved 10-06; ETH, SETH, Gap-ETH, NSETH claimed refuted by a preprint of 10-17 rev. 10-25, under review) showed that nothing in consensus rested on a conjecture, by habit — this makes it a rule.
against
eidos is small today and its record is not yet trustworthy (a Rust kernel, the nox kernel a draft spec, an FNV-1a certificate placeholder, sorry accepted as proved, one-file check, unary Nat, no Int, 139 tests); the ladder E0–E5 is work, and E2 is weeks; the registry's class-G and class-H rows are choices, not science, listed in the same table.
review
node 5 of deliver: a second body reads this unit at one commit. the lenses it reads by, its verdicts and its findings are its own and live on the pull request, soft3#398; each finding keeps its id until disposed, and this page does not restate them. round 1's disposition is on that thread; the observed results it produced moved to the receipt of 2026-10-30.
the verdict below is the owner's on this unit, in the shape every tasks/ decision card carries; it is not the reviewing body's.
verdict
—