The latest progress is not a PDF. It is the round-69 self-reference boundary machine (2026-10-07/08); Paper A / B / C are slices of it, not the whole library.
T1 balance equation has no θ; T2 naive fixed point is degenerate; T3 extremal principle depends on the distribution; L5 endpoint is impossible. The surviving route is T-θ′: L(θ) = 1/L(θ), with the unique solution θ* = 2/(1+e). Boundary: the strong claim "only self-reference" is not proved; w=2 is a machine-checked counterexample to the bare selector claim.
BoundaryReportIdentity (13 theorems / 5 definitions): within the up-closed class, R=R' ⇔ c_R=c_R', and R=Fix(c_I) ⇔ c_R=c_I; ReportCountermodels (4 theorems / 1 definition): non-up-closed counterexamples where the premise cannot be dropped. The unconditional identity is false; the correct theorem is conditional.
7 criteria, 6 pass, 1 fail (criterion 6: Fix(c_I) is not up-closed). The runtime gate flags boundary=known ∧ report=absent as illegal, with a three-valued allow / review / illegal output. Zero LLM, zero GPU.
polarity_change_iff; assembly_not_more_cohesion (only the negative half); commit_from_adjunction / commit_is_optimal; the positive existence theorem is not yet proved.
Unified body: 56 files / 467 theorems+lemmas / 90 definitions; Paper B slice: 53/417/77; A module: 1/33/7; 0 sorry, 0 custom axioms.
This is a research line, not a paper. The evidence — pre-registration, the machine, the results, and the unfinished items — is frozen in a MANIFEST and checked item by item. No "global axiom" is claimed on this basis.