MELV Failure-Boundary ABM V1.0: Empirical Validation of the Equation 7 Gateway Surface

Description This dataset contains the complete run data, summary files, and source code for the MELV Failure-Boundary ABM (FB-ABM) Version 1.0, executed June 2026. The FB-ABM was designed to empirically validate three pre-registered mathematical predictions derived from the C5 derivation (Track B Jacobian stability analysis of the MELV two-species substrate under the saturation form, June 2026). Framework context The MELV (Modified Energetic Lotka-Volterra) framework models cooperative maturity (φ) dynamics in two-species systems. The master equation (Equation 1a, saturation form) is: i(t) = i₀ × (1 − η × tanh(ε × φ(t) × β_norm(t) / η)) The φ dynamics equation (Equation 7) was under active derivation review during 2026. The prior gateway threshold R < 0.50 was retired following a three-error audit (equation-model mismatch, binning artefact, circular criterion). C5 derivation produced the replacement: the Jacobian stability condition β×i_∞ < 1, where i_∞ = i₀×(1−η×tanh(ε×β_norm/η)) evaluated at φ=1. The resulting canonical Equation 7 is: dφ/dt = α(1−φ) × H(1−β×i_∞) × max(0,1−i(t)) − δ×D(t)×φ×H(β×i_∞−1) The FB-ABM was designed to test this result empirically under stochastic finite-population dynamics. Design logic Unlike prior MELV ABM generations (V2.1, V2.2) which mapped the cooperative basin, the FB-ABM inverts the design: it starts systems inside the cooperative basin and drives them toward failure, recording the precise parameter values at which the cooperative equilibrium loses stability. This avoids the survivorship bias inherent in cooperative-basin designs. Full design specification: MELV_FB_ABM_Preregistration.md (included in this deposit). Pre-registered predictions tested P1 (C5-1): Mean empirical β×i_∞ at tipping = 1.00 ± 0.05 across the parameter sweep. P2 (C5-3): Empirical β_crit within ±5% of the analytic small-argument expression 1/(i₀−1) in the small-argument regime (sar = ε×β_norm/η ≪ 1). P3 (C5-2 / Claim 3): NEW-CANONICAL gate form H(1−β×i_∞)/H(β×i_∞−1) produces no pathological artifacts; CLAIM3-SIMPLIFIED (D(t) alone as decay gate) shows premature cessation; OLD-PLACEHOLDER (H(0.5−R)/H(R−0.5)) shows structural misalignment. Results summary P1 — CONFIRMED: Mean empirical β×i_∞ at tipping = 1.0049 ± 0.0007 on canonical pathways (C-increase and TAX-increase, 2,700 runs), 0.49% deviation from the theoretical prediction of 1.000. Across all pathways including beta-increase (4,050 runs): 1.0077 ± 0.0041. Full analytic surface (4,320 tuples) confirms β×i_∞ = 1.000000 ± 5×10⁻⁸ to machine precision via bisection. P1 criterion met. P2 — CONFIRMED with refined regime scope: 19/19 eligible i₀ values within ±5% of numeric β_crit. Analytic expression 1/(i₀−1) accurate to ±5% for sar < 0.5; degrades predictably above. Regime boundary empirically refined from "sar ≪ 1" to sar < 0.5. P3 — CONFIRMED: Gate configuration Tipped Mean β×i_∞ Deviation Mean tip step NEW-CANONICAL 522/522 1.0042 +0.42% 2,420 OLD-PLACEHOLDER 378/522 0.4479 −55.2% 19 CLAIM3-SIMPLIFIED 522/522 0.4670 −53.3% 19 NEW-CANONICAL: no oscillation, no premature collapse, no stochastic resonance. OLD-PLACEHOLDER: 144/522 runs failed to tip at all; when tipping occurred it was premature by 55%. CLAIM3-SIMPLIFIED: premature cessation confirmed at step 19. Stochastic boundary: All 30 conditions (6 parameter tuples × 5 offset fractions), 100/100 replicates tipped. Soft boundary width: mean 0.0017, range 0.0006–0.0047 β-units. Boundary is sharp under demographic noise. File inventory Filename Rows Description FB_ABM_SmallArg_SubSweep_Raw.csv 400 P2 — individual run records (20 i₀ values × 20 replicates) FB_ABM_SmallArg_SubSweep_Summary.csv 20 P2 — per-i₀ summary with regime classification FB_ABM_GateComparison_Raw.csv 1,566 P3 — individual gate comparison records (174 tuples × 3 configs × 3 reps) FB_ABM_GateComparison_Summary.csv 3 P3 — per-gate-configuration summary FB_ABM_GateComparison_Trajectories.csv — P3 — φ and i(t) trajectories for 5 representative tuples × 3 configs FB_ABM_PrimarySweep_AnalyticSurface.csv 4,320 P1 — full analytic tipping surface (1,440 parameter tuples × 3 domains) FB_ABM_PrimarySweep_Simulated.csv 4,050 P1 — simulated verification runs (150 tuples × 3 domains × 3 pathways × 3 reps) FB_ABM_PrimarySweep_Summary.csv 9 P1 — domain × pathway breakdown FB_ABM_StochasticBoundary.csv 30 Stochastic boundary (6 tuples × 5 offsets × 100 replicates per condition) FB_ABM_AdjudicationSummary.csv 6 Summary of pre-registered criterion results fb_abm_core.py — Core ABM engine (gate logic, domain parameterisations, run engine) run_small_arg_sweep.py — P2 small-argument sub-sweep runner run_gate_comparison.py — P3 gate comparison sub-sweep runner run_primary_sweep.py — P1 primary sweep runner (analytic surface + simulated verification) run_primary_sweep_b.py — P1 simulated verification only (trimmed for session constraints) run_stochastic_boundary.py — Stochastic boundary characterisation runner compile_adjudication_package.py — Adjudication package assembly and criterion assessment MELV_FB_ABM_Preregistration.md — Pre-registration document (design authority) MELV_Framework_Whats_Changed_and_Why_v4_3.md — Supplemental — framework amendment trail and context (see below) Supplemental files Two files are included as supplemental documentation alongside the primary dataset: MELV_FB_ABM_Preregistration.md (primary supplemental): The pre-registration document that served as the design authority for this ABM. Specifies all parameter ranges, success criteria, tipping detection criterion, three-configuration gate comparison design, biological calibration anchors, and the anti-circularity protocol. Included so the dataset is fully self-contained: readers can verify that the run design matches the pre-registered specification without accessing external documents. MELV_Framework_Whats_Changed_and_Why_v4_3.md (framework context supplemental): The MELV framework's amendment narrative — a structured record of what changed in the framework, why, and what evidence supported each change. Provides the intellectual context for understanding what the FB-ABM was testing: why R<0.50 was retired, why the 0.50 threshold needed replacement, what the three-error audit found, and how the C5 derivation produced the gate forms tested here. Without this document, the dataset is interpretable in isolation but not in framework context. Readers approaching the dataset from outside the MELV project will find this document essential for understanding the scientific question the runs were designed to answer.