Manhattan Pinball: lifted Schur Lyapunov certificate at R=2, q=0.2 (collar-depth 4)

Computational artifact accompanying ongoing work on the Manhattan Pinball localization conjecture for q in (0, 1/2). The Manhattan Pinball model is a Lorentz-mirror lattice walk; Grimmett records the all-density localization statement as an open problem, and Li proves only the near-critical regime q > 1/2 - epsilon. This deposition contains the lifted prefix-wise row database, sparse upper-bound transition matrix, trunk/bubble classification, Lyapunov function, and full Schur-complement Lyapunov certificate at R=2, q=0.2 with collar-depth 4 lifted profiles. The certificate verifies 0 violations across 2,634,103 states (17,294 trunk + 2,616,809 bubble), with rho_cert = 1.0e-4 and rho_star_upper = 7.55e-9 from solve. Contents (see README.md in the tar.gz): lifted_q02_R2.sqlite (4.6 GB) - lifted row database (profiles, rows, entries) matrix_q02_R2_schur.npz (260 MB) - sparse 2.63M x 2.63M transition matrix, 30.6M nnz kind_q02_R2.json (119 MB) - trunk/bubble classification of all states h_q02_R2_schur.npy (21 MB) - Lyapunov function h over all states cert_q02_R2_schur.json - certificate metadata (rho, margins) worst_q02_R2_schur.jsonl - top 50 tightest rows MD5SUMS - integrity hashes Reproduction: Source code at https://github.com/okezue/ManhattanPinball. Pipeline: prefix_lyapunov_v2.py build-shards (R=2, q=0.2, L=5, K=0.5, split-depth=10, collar-depth=4, collar-width=1, collar-mode=mask, backward-dual-collar, dedupe) then merge-shards, then schur_certificate.py kind-map (mode=child-only-bubble) then from-db (max-entry aggregation) then solve (rho-hi=0.999999, rho-safety=1e-4) then verify.