{ "record_id": "T2-NETI-NETI", "semantic_name": "T2 · Neti Neti — structural (a)symmetry between unification and reportability at ML rate-distortion equilibrium", "record_class": "SIMULATION", "program": "Structural Claims of the Valleys (T-series)", "version": "1.2.0", "date": "2026-07-31", "authorship": { "primary": "T-series research thread, Ozone Archive Research Program", "model_attribution": "designed and executed autonomously by AI research thread under user oversight" }, "status": "UNSUPPORTED at preregistered aggregation level (v1.0); H_MIXED_WINS on T2-R2 (v1.1.0); T2-R2a (v1.2.0) power-analysis verdicts: F1_INTERMEDIATE (A3 stabilizes H_STRICT at D1 β=128 at 9/15 = 60% vs null 0/15; A2 short at 5/15); F2_STABILIZES (A3 stabilizes H_LINEARLY_HIDDEN at D3 β=8 at 10/15 = 66.7% vs null 3/15). Capacity-dependent effects verified as real, above-null-baseline, and reachable-but-not-inevitable in higher-capacity Gaussian bottlenecks.", "evidence_type": "COMPUTATIONAL_SIMULATION", "physical_valid": "NONE", "replication_status": "internal_followup_T2R2_and_T2R2a_completed", "claim_status_banner": "UNSUPPORTED at aggregation (v1.0); H_MIXED_WINS (v1.1.0 T2-R2); F2_STABILIZES (v1.2.0 T2-R2a) — A3 at D3 β=8 stabilizes to 10/15 = 66.7% (Wilson 95% CI [41.7, 84.8]) far above null 3/15 = 20% for H_LINEARLY_HIDDEN. F1_INTERMEDIATE (v1.2.0 T2-R2a) — A3 at D1 β=128 stabilizes H_STRICT at 9/15 = 60% vs null 0/15; A2 short of hurdle at 5/15. All test archs beat null on D1 β=128 (0/15). Capacity-dependent phenomena are real, not pipeline noise; not universal, not inevitable.", "source_proposition": "Advaita 'neti neti' / Bahá'í 'seventh valley' — a kenotic end-state in which unified perception, retained substrate content, and preserved reportability coexist; operationalized as joint conditions on MI(X;Z), decoder-output classification skill, single-sample latent classification skill, and substrate classification skill in variational bottleneck models", "related_records": [ "T1-EMPTY-THRONE", "O0-CRP-021" ], "companion_record": "T1-EMPTY-THRONE", "primary_verdict": "H_OTHER (v1.0 main); H_MIXED_WINS (v1.1.0 T2-R2); F1_INTERMEDIATE and F2_STABILIZES (v1.2.0 T2-R2a) — capacity-dependent H_STRICT and H_LINEARLY_HIDDEN signatures verified above null baseline in A3 at specific (dataset, β) regimes", "hypotheses": { "H_STRICT": "FAILS · 0/3 test arches meet aggregation criterion (v1.0)", "H_ASYMMETRY": "FAILS · 1/3 arches (A3 only) exceeds null baseline on 2/3 datasets (v1.0)", "H_UNREACHABLE": "FAILS · not uniformly monotone (v1.0)", "H_OTHER": "TRUE · v1.0 final verdict", "R2_H_ABSENT": "PARTIAL · holds at β=128,β=512 but not at β=8 (v1.1.0)", "R2_H_LINEARLY_HIDDEN": "FAILS · 0/3 test arches meet aggregation criterion; A3 has individual-seed hits (v1.1.0)", "R2_H_MIXED": "TRUE · 2/3 test arches (A1, A2) satisfy partial gap condition at β=8, β=32 (v1.1.0)", "R2a_F1_D1_beta128_H_STRICT": "F1_INTERMEDIATE · A3 stabilizes at 9/15 = 60.0% [35.7, 80.2]; A2 short at 5/15 = 33.3%; A1 at 6/15 = 40.0%; C_null 0/15 = 0.0% [0.0, 20.4] (v1.2.0)", "R2a_F2_D3_H_LINEARLY_HIDDEN": "F2_STABILIZES · A3 at β=8 reaches 10/15 = 66.7% [41.7, 84.8], far above null 3/15 = 20.0% [7.0, 45.2]; effect disappears at β=32 (A3 3/15 vs null 1/15) (v1.2.0)" }, "controls": { "C_null_linear_gaussian_pPCA_VAE": "H_STRICT hits: 0/21 (as expected). Substrate-perception baseline gap up to 0.598 at D1 β=128. On R2: shows H_MIXED at β=8 on D3 (mlp_μ up to 0.556, gap up to +0.176) — confirming part of the mixed effect is generic Gaussian-bottleneck geometry.", "C_positive_two_pathway_VAE": "H_STRICT hits: 12/21 (D1: 7, D2: 5, D3: 0). Positive control passes. On R2: mixed pattern on D3 at β=8 similar to test arches." }, "deviations": [ "DEV-001 (2026-07-31, pre-execution) — Rule A(1) reinterpretation (null H_ASYMMETRY = baseline, not abort trigger), skill_substrate = max(skill_μ, skill_h_x), H_ASYMMETRY revised to require test-arch gap ≥ null gap + 0.10. New disjoint test-arch seeds {20260940, 20260950, 20260960}." ], "preregistration": "_internal/t2-neti-neti/preregistration.md", "preregistration_r2": "_internal/t2-neti-neti/prereg_r2.md", "preregistration_r2a": "_internal/t2-neti-neti/prereg_r2a.md", "code": [ "_internal/t2-neti-neti/src/common.py", "_internal/t2-neti-neti/src/datasets.py", "_internal/t2-neti-neti/src/metrics.py", "_internal/t2-neti-neti/src/vae.py", "_internal/t2-neti-neti/src/two_pathway.py", "_internal/t2-neti-neti/src/linear_gaussian.py", "_internal/t2-neti-neti/src/pilot_controls.py", "_internal/t2-neti-neti/src/pilot_controls_v2.py", "_internal/t2-neti-neti/src/full_run.py", "_internal/t2-neti-neti/src/nonlinear_probe.py", "_internal/t2-neti-neti/src/r2_run.py", "_internal/t2-neti-neti/src/r2a_run.py" ], "results": [ "_internal/t2-neti-neti/runs/controls/pilot/pilot.log", "_internal/t2-neti-neti/runs/controls/pilot/results.json", "_internal/t2-neti-neti/runs/controls/pilot/verdict.json", "_internal/t2-neti-neti/runs/controls/pilot_v2/pilot_v2.log", "_internal/t2-neti-neti/runs/controls/pilot_v2/results.json", "_internal/t2-neti-neti/runs/controls/pilot_v2/verdict.json", "_internal/t2-neti-neti/runs/main/v1/main.log", "_internal/t2-neti-neti/runs/main/v1/results.json", "_internal/t2-neti-neti/runs/main/v1/verdict.json", "_internal/t2-neti-neti/runs/r2/v1/r2.log", "_internal/t2-neti-neti/runs/r2/v1/results.json", "_internal/t2-neti-neti/runs/r2/v1/verdict.json", "_internal/t2-neti-neti/runs/r2a/v1/r2a.log", "_internal/t2-neti-neti/runs/r2a/v1/results.json", "_internal/t2-neti-neti/runs/r2a/v1/verdict.json" ], "replication_queue": [ { "id": "T2-R1", "status": "REGISTERED", "purpose": "Scale to larger latent dims and non-Gaussian datasets; test whether A3's capacity-dependent effect strengthens" }, { "id": "T2-R2", "status": "COMPLETED (2026-07-31, v1.1.0)", "verdict": "H_MIXED_WINS", "purpose": "Nonlinear probes on μ for D3 rings to distinguish 'signature absent' from 'signature linearly hidden'" }, { "id": "T2-R3", "status": "REGISTERED", "purpose": "Alternative objectives (WAE, VQ-VAE) to test whether fragile positive instances persist outside β-VAE" }, { "id": "T2-R2a", "status": "COMPLETED (2026-07-31, v1.2.0)", "verdicts": { "F1_D1_beta128_H_STRICT": "F1_INTERMEDIATE (A3 stabilizes 9/15 = 60.0%; A2 short of 7/15 hurdle at 5/15; A1 6/15; C_null 0/15)", "F2_D3_H_LINEARLY_HIDDEN": "F2_STABILIZES (A3 at β=8 reaches 10/15 = 66.7% vs null 3/15 = 20.0%)" }, "purpose": "Power-analysis follow-up: 12 additional disjoint seeds merged with T2 v1.0's 3 for 15-seed pool on focused (kind, dataset, β) scope. Determined that fragile positive instances at D1 β=128 and A3 individual hits at D3 β=8 are systematic capacity-dependent effects above null baseline, not seed-lottery noise." }, { "id": "T2-R2b", "status": "REGISTERED (from R2 §11.8)", "purpose": "Larger nonlinear probes to test whether mixed signal at moderate β is probe-capacity-bounded" }, { "id": "T2-R2c", "status": "REGISTERED (from R2 §11.8)", "purpose": "Alternative non-linearly-separable datasets to test whether mixed pattern is D3-specific or generalizes" }, { "id": "T2-R2a-1", "status": "REGISTERED (from R2a §12.7)", "purpose": "A2a = ReLU with dz=4 to disentangle activation-vs-capacity contribution to F1 D1 β=128 stabilization gap between A2 (5/15) and A3 (9/15)" }, { "id": "T2-R2a-2", "status": "REGISTERED (from R2a §12.7)", "purpose": "30-seed replication at A3 D3 β=8 to tighten Wilson CI around ~67% point estimate" }, { "id": "T2-R4", "status": "REGISTERED (from R2a §12.7)", "purpose": "Latent-geometry visualization + KL-per-dimension analysis for A3's systematic strict-signature behavior" } ], "predecessor_probe": { "id": "T2-P0", "date": "2026-07-30", "notes": "Internal exploratory probe on A1_tanh_dz2 alone, single seed, on D1 only. Found SWEET_SPOT_MU_ONLY pattern. T2 v1.0 supersedes P0 with disjoint seeds and full study rigor; P0 is retained only for provenance." }, "downloads": { "scientific_record_md": "scientific_record.md", "manifest_json": "manifest.json", "preregistration_md": "preregistration.md", "preregistration_r2_md": "prereg_r2.md", "preregistration_r2a_md": "prereg_r2a.md", "deviations_md": "deviations.md", "main_verdict_json": "verdict.json", "r2_verdict_json": "verdict_r2.json", "r2a_verdict_json": "verdict_r2a.json" }, "revision_history": [ { "version": "1.0.0", "date": "2026-07-31", "summary": "Initial publication. Preregistration locked. DEV-001 logged pre-execution. Full run: 315 fits, ~19 min wall time. Verdict H_OTHER (UNSUPPORTED at aggregation with fragile positive instances)." }, { "version": "1.1.0", "date": "2026-07-31", "summary": "T2-R2 addendum. Preregistration locked. Nonlinear MLP probe over 315 configurations, deterministic reuse of T2 fits verified (0 failures). Verdict H_MIXED_WINS: partial linearly-hidden effect on D3 at moderate β (β=8, β=32) in A1 and A2; genuine absence at deep collapse; capacity enhancement in A3. Null-control also shows partial mixed effect, confirming part of the effect is generic Gaussian-bottleneck geometry." }, { "version": "1.2.0", "date": "2026-07-31", "summary": "T2-R2a addendum. Preregistration locked pre-execution (2026-07-31, after v1.1.0 publish, before R2a code). 12 new disjoint seeds × 4 kinds × 2 datasets × 3 β = 288 new fits + 72 old-seed re-evaluations for 15-seed pool (360 total). Determinism gate 0/360 failures. F1 verdict F1_INTERMEDIATE (A3 clears at 9/15 = 60% for D1 β=128 H_STRICT; A2 short at 5/15). F2 verdict F2_STABILIZES (A3 clears at 10/15 = 66.7% for D3 β=8 H_LINEARLY_HIDDEN vs null 3/15 = 20%). Null baselines: 0/15 (D1 β=128) and 3/15 (D3 β=8) — all test archs beat null on D1; A3 beats null decisively on D3 β=8. Fragile positive instances established as systematic capacity-dependent effects, not seed-lottery noise. Wall-clock 25 min." } ] }