Conjecture tests: EM-annihilator killed (F-6); condensate-as-collapse and hypercone-projection survive

[tensorrent]

Summary

Three conjecture tests under the Adversarial Compression Cycle — each stated in precise kill-able form before computing, then read as it lies. One kill, two survivors. All decisions in exact rational arithmetic (sympy); the single measured quantity is labeled T3.

1. EM-as-torsion-annihilator — KILLED, both forms (T4) → ledger F-6

code/acs_codebase/extras/test_conjecture_em_torsion_annihilator.py

• Strong form (Q unique annihilator of the full torsion sector): the torsion sector is exactly Sym₀(4), and its centralizer in sl(4) is {0} — the full sector annihilates nothing.
• Weak form (Q unique in ker ad_{T_BL}): Q = J3+K3+T_BL/2 is not even in the kernel; the kernel is the full 9-dim sl(3)⊕u(1) block.
• Positive residue (new T1 locked invariant): the canonical torsion-coupling operator is block-scalar with exact spectrum {4 (×9, Sym₀ block), 6 (×6, o(4) block)} — torsion coupling resolves only the Palatini split; no direction inside either block is graded.

2. Condensate-as-collapse ("slag of the furnace") — SURVIVED, all four forms (T1)

code/acs_codebase/extras/test_conjecture_condensate_collapse.py

The image/ω-limit dual of the killed kernel form. ad_{T_BL} is semisimple with exact spectrum {−4/3 (×3), 0 (×9), +4/3 (×3)}; every generic direction collapses projectively onto the 3-dim dominant sector V₊, which is abelian, nilpotent of order 2, and consists exactly of the three lepton→quark transition operators; collapse rate 4/3 = Δ(B−L) exactly. The invariant (uncollapsed) sector is exactly sl(3)⊕u(1)_{B−L}.

3. Hypercone-through-the-slice — SURVIVED, all three forms (T1/T1/T3)

code/acs_codebase/extras/test_conjecture_hypercone_projection.py

• Real 2-param slice x·D + y·S03 in the fermion 4-rep has eigenvalue sheets exactly ±√(x²+y²) (true double cone); 1-param sub-slices see hyperbolae — the cone poking through, apex off-slice (codim 2, β=1 class).
• Adding the chirality direction z·(i·A03) — the same i introduced by the J-map (Prop 9.7) — gives sheets exactly ±√(x²+y²+z²): codim 3, β=2 class. The repulsion exponent is a dimension counter (β = codim − 1).
• Measured on the repo's 100k Riemann zeros (unfolded): fitted small-spacing exponent β = 2.019 — GUE/chirality class, the shadow of a codim-3 conical structure.

Ledger updates

• docs/Elimination_Ledger.md: kill entry F-6 + two survivor entries (2026-07-17), each with mechanism and scope boundary.
• docs/ACS_Antigravity_Feed_Package/key_parameters_ledger.json: F-6 falsification recorded; new T1 locked invariant torsion_coupling_spectrum = 4:6.

Scope boundaries (enforced in each script)

All results are exact statements about the sl(4,R) Palatini structure, its fermion 4-rep, and parameter-space spectral geometry. None establishes physics of spacetime, measurement, or cosmology; open problem #2 (constructing the Hilbert–Pólya operator) stands. Narrative names correspond structurally; identification with physical collapse/projection remains at the conjecture layer.

Verification

python -m pytest code/acs_codebase → 42 passed on every commit; all three new scripts run standalone with deterministic output.

🤖 Generated with Claude Code

https://claude.ai/code/session_01ByfauYHs7CN8VHXGMBotFp

---

Generated by Claude Code